I looked across 20 Western countries, and 32 companies passed.

Some of them have delivered amazing returns

Let’s look at a few examples:

For the most part, these are exceptional long-term returns.

The studies

In terms of stock market performance, Bain & Co compared in 2025 companies’ annualized total shareholder returns (TSR) according to their acquisition frequency. The firm defines frequent acquirers as companies making more than one acquisition per year.

The gap is large: from 2012 to 2022, frequent acquirers achieved 8.5% per year, compared with 3.7% for companies that made no acquisitions.

It is not the acquisition itself that drives performance, as will be shown in the studies later in this article. The initial quality of the companies, their size, or their industry can also play a role.

McKinsey compares acquisition strategies among large global companies. The firm defines programmatic M&A as repeated small or medium-sized acquisitions, generally at least two per year, cumulatively representing a significant share of the market capitalization.

This suggests that acquisition discipline and regularity matter more than a few exceptional large deals. The latter are often harmful to shareholders.

There is a potential survivorship bias: companies that were no longer publicly traded at the end of the period are not included in the study’s sample.

In a study published in 2025 in The Review of Corporate Finance Studies (Solving Serial Acquirer Puzzles), on 55,482 acquisitions carried out by 8,640 publicly traded US companies, between 1989 and 2018, Macias, Rau & Stouraitis identified four profiles:

• Loners: one or two acquisitions.

• Occasional acquirers: infrequent acquisitions.

• Sprinters: acquisitions concentrated over short periods.

• Marathoners: regular acquisitions, practically without interruption.

Their main results:

• 17.5% of acquirers carried out 60% of acquisitions.

• The most frequent acquirers were more likely to acquire private companies.

• After adjusting for market expectations, the returns associated with the marathoners’ acquisitions do not decrease over successive acquisitions.

• Marathoners accumulate significant economic gains thanks to numerous small transactions.

In a study published in the Journal of Financial Economics (Extraordinary acquirers), Golubov, Yawson & Zhang found that:

• The abnormal returns at acquisition announcements vary considerably from one company to another.

• The best companies show some persistence in their strong performance.

• This persistence does not necessarily disappear when an executive is replaced.

• Company-specific characteristics explain a significant part of the differences in performance.

This provides academic evidence supporting the idea of organizational skill in making acquisitions.

I also found studies that challenge the outperformance of serial acquirers:

• Laamanen in 2008 (Performance of serial acquirers: toward an acquisition program perspective) found that frequent and irregular acquisitions were associated with poorer performance.

• Hossain, Pham and Islam in 2021 (The performance and motivation of serial acquisitions: Evidence from Australia) suggest that serial acquirers obtain lower stock-market returns around announcements, with higher risk. Saccuman & Schraa report similar findings in Europe in their 2024 study, The Impact of Deal Frequency and Acquisition Blocks on Bidder Performance.

What Makes a Good Serial Acquirer?

I see six characteristics specific to a good serial acquirer:

-The acquisition strategy is clear and disciplined. It knows what type of company it is looking for, consistent with its business model, and what price it is willing to pay. It also knows when to walk away if the price becomes excessive.

-The return on invested capital is high. The acquisitions must generate sufficient profits relative to the amounts invested. The incremental return on invested capital (ROIIC), which can reflect both acquisitions and organic investments, remains high over time.

-The company can finance its acquisitions internally, which helps limit new share issuance and shareholder dilution. Here, we look at the growth in earnings per share and FCF per share.

-Debt remains under control. With high debt levels, the company can accelerate its growth, but at the cost of greater vulnerability if interest rates rise or the economy slows down. The ability to weather difficult periods without endangering the whole group is essential.

-The acquired companies must continue to function well. They must at a minimum maintain their activity, and ideally continue to develop their sales and their margins. Persistently negative organic growth can hide operational problems.

-The serial acquirer must be able to reinvest for a long time and on a large scale. As it grows, it must also deploy more capital to maintain the same pace of growth. This is why small serial acquirers operating in very fragmented markets can offer particularly attractive potential.

The central principle: a good serial acquirer is not the one that carries out the most acquisitions, but the one that consistently reinvests substantial amounts at high returns, for the benefit of shareholders.

For readers who want to learn how to evaluate these companies themselves, I strongly recommend FluentInQuality’s article:

Here’s How to Spot and Evaluate a Great Serial Acquirer in 10 Minutes.

But buying companies is not a guarantee of success. Acquisitions can also destroy considerable value. The risks are the negatives of the points mentioned above, with operational consequences.

What’s Included in My List

A conventional stock screener cannot identify serial acquirers reliably. I therefore identified historical serial acquirers, companies that follow similar models and companies that regularly make acquisitions, even when acquisitions are not the sole driver of their growth.

Once I had identified an initial pool of companies, I applied the following filters:

I added a Serial Acquirer score, comprising six criteria. Each criterion receives a score from 0 to 3, for a maximum score of 18. The criteria are the following:

1. Capital Returns

2. M&A Discipline

3. Balance Sheet

4. Funding

5. Organic Growth

6. Reinvestment Capacity & Runway

I have also included a few companies that do not strictly fit the definition of a serial acquirer, but whose acquisition strategies are interesting enough to fit in this list.

SUBSCRIBE FOR FREE TO SEE THE FULL LIST

Subscribe now

Everyone has their own edge in certain sectors; your profession or your personal experience will certainly steer you toward one of them.

You can also stay on generalist subjects where logical analysis and some market research can bring you to a sufficient level of understanding.

Remember nonetheless the following quote from Warren Buffett:

He is also credited with:

One could give the company that I am going to present to you the advantage of simplicity.

The structure of my deep dive is standard. Learn the process in the dedicated articles in the learning section. The whole article is long because it covers a lot of ground, and short because I try to give you a synthesis of the points. So I start with a “short version” for those of you in a hurry.

Subscribe now

Today, we’ll talk about Storytel, a company based in Sweden and running a leading audiobook and e-book subscription service, with a stake in book publishing.

The Short Version

Storytel is a Swedish audiobook and e-book company with 2.75 million paying subscribers. It operates mainly in the Nordics and Europe, but also owns Audiobooks.com in the US.

Its business model allows it to acquire publishers and then distribute their content internationally through audiobooks, while gaining insights into what its consumers want. The model is profitable, enabling the company to acquire additional publishing houses.

Since 2023, the company has gone from heavy losses to profitability, improved its margins and now has almost no net debt. Returns on invested capital are rising, and Storytel can finance further acquisitions with its own cash.

But the stock has already increased fivefold since its 2023 low. A large part of that increase comes from investors being willing to pay a higher multiple.

The main question is competition. Spotify has entered the Nordic audiobook market, where Storytel makes most of its money. Subscriber growth is slowing there, and we still don’t know how much pressure Spotify and potentially YouTube could put on Storytel in the coming years.

I’m not investing in the stock for now, and I’m keeping an eye on how the share price and competition evolve. A great company with long-term prospects, and one to add to my watchlist.

1. What Storytel does

History

In 2005, Jonas Tellander and Jon Hauksson founded Bokilur, an audiobook service in Sweden with a unique product. The launch of the iPhone in 2007 made listening on the move practical, and in 2011 the first audiobook produced by Storytel came out on the Swedish market.

Jonas Tellander, CEO from 2005 to 2022

The company then grows by buying what it lacks. In 2013, it acquires the audiobook publisher Storyside and then embarks on buying publishing houses and an audiobook subscription service in the Nordic countries. The company has been listed on the Nasdaq First North Growth Market since December 2018 and passes one million paying subscribers in 2019.

In 2022, it buys the American service Audiobooks.com, reaches two million paying subscribers and its catalogue exceeds one million titles. The same year, Jonas Tellander hands over the reins of the company, remaining on the board and as a shareholder. Johannes Larcher then took over until 2024, when he was replaced by the current CEO, Bodil Eriksson Torp.

On 8 June 2026, the company transferred its shares to Nasdaq Stockholm.

What Storytel sells

The company sells stories, in two forms: access to a catalogue of audiobooks and e-books by subscription, and books that the group publishes itself (paper, audio, digital).

The group describes itself as an “ecosystem” that combines technology, distribution and content creation, through its two complementary businesses: Streaming and Publishing.

At the end of 2025, the subscription business had more than 2.67M paying subscribers, in more than 25 markets, with a catalogue of more than 1.8M titles in more than 55 languages. The market is essentially local: about 85% of listening relating to local authors and languages. A catalogue is therefore not sold in the same way in every country: publishers and authors on the ground are needed.

The company lives mainly off its subscriptions, which are billed in advance. A very appreciable part, as this revenue is recurring and predictable. The main costs are content (copyright fees, third-party publishers, printing) and marketing.

Streaming: the platform

Streaming is the consumer-facing platform, under three brands: Storytel, Mofibo and Audiobooks.com. The customer pays for a monthly subscription to listen or read with several offers that we will review in the competition section.

Added to this are distribution partnerships such as Klarna (Premium members in 14 markets access the catalogue), Vodafone Turkey, Onet Premium in Poland, and RDF Media in Chile. In addition, the company launched single-title sales in 2025 for best-sellers in English, to complement the subscription.

Publishing: the content factory

Publishing is the creative base of the group: it secures rights and works with authors on print and digital titles. It groups together publishing houses, mostly in the Nordics.

Publishing also has a licensing agreement with Spotify in the Nordic countries, so Spotify is both a competitor of Storytel’s Streaming business and a customer of Publishing.

Customers

Storytel mostly sells to individuals.

At the end of June 2026, the service counted 2.67M paying subscribers. Usage is a strong habit: 37% of paying subscribers listen every day.

The company divides its revenue by operating segments, i.e. Nordics, Europe, Americas and APAC.

Unfortunately, this segmentation does not clearly show the dynamics in the US, as the US is grouped with all of Latin America. This suggests that the figures are not exceptional in the US; otherwise, the company would show them separately.

The streaming segment represents 82.3% of 2025 consolidated revenue. No customer accounts for 10% or more of revenue.

Suppliers

Storytel mostly buys stories. In 2025, the cost of goods sold is 2,190.1 MSEK, and its main item, 1,858.1 MSEK, or 46% of consolidated revenue, consists of payments to the rights holders of digital and printed books. To this are added the fees of payment providers and the cost of printed books. The other suppliers account for much less. The catalogue contains more than 1.8M titles, with more than 600,000 authors.

Some of them publish through publishing houses belonging to Storytel.

In the company’s documents, no content supplier is named as dominant: the group publishes neither the list nor the weight of its main publishers.

For the other supplier items, there is paper and printing, as well as technical infrastructure (such as Google Cloud), and payment providers.

Competitors

85% of Storytel’s listening relates to local authors and languages. Indeed, people still often prefer to listen to a book in their native language. Storytel faces four types of players: a local leader (itself), challengers backed by publishers, global platforms that serve each language with translated titles, and a few small local players. Here is an overview of direct competition:

Storytel is linked to Spotify because its publishers have signed a licensing agreement with it in the Nordic countries. Spotify is therefore both a customer of Publishing and a competitor of Streaming.

We can also add indirect competition: every day the user will choose how to be entertained. This can be through listening to a book, watching a series, watching a video or listening to a podcast, browsing a social network, or any other distraction. Storytel says it fights for subscribers and content, but also for consumers’ time in general, and expects this competition to increase. I had already talked about it in the Netflix deep dive on the notion of “moment of truth”.

Here is a breakdown of usage in Sweden, compiled by Nordicom in 2025 based on nearly 6,000 people aged 9 to 85.

This image shows that audiobook listening remains a minority activity: 15% of Swedes on an average day, all services combined (Storytel, BookBeat, Spotify, libraries), far behind social networks, streaming video and YouTube. However, it is a lasting habit: of these 15%, 12 percentage points listen for more than 30 minutes, i.e. four listeners out of five.

Competitive advantage

Let’s take the different types of advantages:

Switching costs: low

The subscription is monthly (3, 6 or 12 months in certain markets), with no long-term commitment or loyalty program. People tend to stay out of habit, but they can leave very quickly when they see another offer.

Brand: medium

This point is difficult to evaluate quantitatively, but I suggest drawing a parallel with luxury: products belong to brands, grouped within “houses” (Vuitton, Tiffany, Gucci, etc.), and those houses are grouped within a company.

Each author is like a “house” in luxury whose books are unique products, and each publishing house is like a portfolio (or a collection) of authors to whom the public becomes attached.

Storytel owns publishing houses and therefore the distribution rights of the catalogue.

But if the author leaves, the public leaves, which puts the strength of the “Storytel brand” into perspective. The public’s attachment goes to the author before going to the service.

Its brand advantage currently exists mostly in the Nordics, where Storytel says it has been the leader of Nordic audiobooks for some twenty years, ranking first by number of subscribers in Sweden, Denmark, Norway and Iceland, with 65% of Streaming sales.

Norstedts, founded in 1823, saw one of its authors, László Krasznahorkai, receive the 2025 Nobel Prize in Literature.

Product (what Storytel does) hard to copy: low

The service itself (app, recommendations, synchronized listening between audio and text) is not rare: Spotify added it to its subscription in four Nordic countries in November 2025, and Bonnier operates BookBeat in 10 markets.

Product (audiobooks): protected by intellectual property law

Publishing also produces its own content through its publishing houses and Storyside, including Storytel Originals reserved for the service. The rest of the catalogue (more than 1.8M titles) is licensed. Backlist titles, released more than a year ago, make up 62% of consumption, and according to Storytel about 85% of listening is tied to local authors and languages: an entrant must redo this licensing work market by market, and it cannot be improvised.

This advantage protects the country-specific share of the catalogue, not the translated international titles, which any service can license in each language.

And we do not know if these licenses are exclusive: if not, a rival can license the same titles, and Spotify already does so with local publishers, including Bonnier.

AI changes the game particularly in distribution, making it easier to translate an existing book.

Network effects: the company has, in my view, a very interesting flywheel, described below.

This flywheel means the moat strengthens over time, through external acquisitions, provided that they are made at a good price.

2. Financials

Key financial metrics

Commentary

Revenue

Growth is slowing: from 22.1% in 2022 to 8.8% in 2024 and 5.9% in 2025. The majority of the decline comes from the exchange rate, with the appreciation of the Swedish krona: growth at constant exchange rates is 9.2%.

This growth comes mainly from volumes, because the number of paying subscribers increases by 7.2% over one year (2.75M at the end of June 2026), whereas ARPU is stable at 116.4 SEK per month (116.5 a year earlier).

A word on the publishing business, which is a bit complicated from an accounting perspective.

Say Norstedts publishes a novel. It sells the book to bookstores (external sale), and it also gives the audio and e-book rights to the Storytel service, which puts them in its catalogue (internal sale).

In the group’s accounts, both are recorded as revenue in the Publishing segment. But when the accounts are consolidated, the internal sale disappears from revenue, and the corresponding purchase disappears from Streaming’s content costs. Only the real costs remain (copyright fees, printing, etc.).

Operating Margin

The 2023 loss includes 671.6 MSEK of exceptional items, including 465.4 MSEK of goodwill impairment of Audiobooks.com and 167.6 MSEK of content impairments, mostly in expansion markets.

The improvement since then comes from operations with the increase in operating leverage: in 2025, each additional krona of revenue generated about 0.78 krona of operating profit. In the first half of 2026, it was about 0.51, depending on the evolution of costs and revenue.

Net Income and Free Cash Flow

The 2025 net income is inflated by tax. Storytel accumulated losses in the past, and these losses are deductible from its future profits. Until now, it had not recorded them on its balance sheet. In 2025, it judged that it would have enough profits to use them. It therefore recorded them as an asset, which creates an accounting tax gain of 205.0 MSEK. Without this gain, net income would be 299.0 MSEK, about 40% more than in 2024.

This is the main cause of the gap between net income and FCF in 2025. For H1 2026, they are within about 5% of each other, because depreciation and amortization exceed investments, alongside changes in working capital.

Balance sheet

As of 31 December 2025, goodwill and intangible assets make up 51% of the balance sheet total (3,570.4 MSEK). Equity is 1,899.3 MSEK: once goodwill and intangibles are deducted, 68.4 MSEK remains.

Necessarily, for a company that grows through acquisitions, the amount of goodwill is bound to be large. The risk, in the event of a downturn in activity, is that an impairment will be recorded as a net loss. With the next acquisitions, this ratio of goodwill to balance sheet total could increase, which would weaken the balance sheet.

At the end of June 2026, net debt is 13.7 MSEK, or 6.6% of EBIT, which leaves room for acquisitions, or returns to shareholders via dividends or share buybacks. All covenants, including the one on net debt relative to EBITDA, were met in 2025.

3. Management

The new CEO

Bodil Eriksson Torp has led Storytel since 1 October 2024, taking over as head of the group after the founder, who stepped back from his duties in 2022, and after Johannes Larcher (who came from HBO Max) who served until 2024.

Born in 1968, she spent fifteen years at Bonnier, eight of them in general management positions, then eight years at the head of the press group Aller Media, and she most recently led the investment company VO2 Cap. Her profile is that of an executive from subscription media and major consumer brands. She comes from Bonnier, owner of BookBeat, Storytel’s direct competitor.

Under her leadership, the company reaps the fruits of the turnaround carried out by her predecessor. Almost all the executives have held their current roles since 2025.

The CEO received about 9.7 MSEK in 2025. Her personal exposure mainly goes through the share plan, as she only holds 12,634 B shares, or about 1.5 MSEK at the current price (about $150,000). That is a small amount and does not, on its own, ensure alignment with shareholders.

She also holds 108,010 share rights (about 12.7 MSEK if all are vested), linked 60% to the performance of the share and 40% to sustainability criteria, exercisable in the summer of 2028.

The other executives hold few shares (50,000 for the chief financial officer, a few thousand or tens of thousands for the others, i.e. about 0.13% of the capital for the whole team, general manager included), and 430,098 share rights.

The shareholders

Storytel has no controlling shareholder. The capital is dispersed (27,099 shareholders at the end of 2025), the top ten hold 40.55%, and each B share gives one vote (the 635 A shares give ten, which is negligible). The company has been listed since June 2026 on the main market of Nasdaq Stockholm, after nearly eight years on First North.

Roxette Photo SA is a holding company located in Luxembourg, with Jonas Sjögren as co-owner. It owned 20% of the capital in 2019; its stake fell to 7.31% at the end of 2025, then, in 2026, it bought 110,000 shares at 83.21 SEK (9.2 MSEK) and now holds 5.75 million shares, or 7.44%.

Capital allocation

The company has capacity for profitable reinvestment with a ROIC clearly above the cost of capital.

Capital finances what it should finance: first internal growth, then acquisitions, then returns to shareholders.

In 2025, the group invested 193.9 MSEK, or a third of its operating cash flow, and spent 883.6 MSEK on sales and marketing (22.0% of revenue). It also devoted 154.9 MSEK to development, of which 25% were capitalized.

In M&A, Storytel acquired two publishers between January 2025 and May 2026, paying in cash: Bokfabriken in Sweden, at 7x its EBIT, which is a good price, and Overamstel in the Netherlands and Belgium, at about 11.9 times its 2025 EBITDA, allowing it to enter these markets.

The company has a habit of paying a dividend, which represented 24% of 2025 net income attributable to shareholders. Storytel also buys back shares, in amounts corresponding to the incentive plans. This part therefore aims only to counter dilution.

4. The two theses

Why did the stock go up?

The stock trades at about five times its 2023 low, while profit went from -813M SEK for FY 2023 to 504M SEK at the end of FY 2025.

In logarithmic terms, about a quarter of the rise comes from earnings and three quarters from valuation. The market therefore pays more for the same company, or judges it less risky. I see several reasons:

Storytel returned to profitability after its 2023 losses, linked to the purchase and impairment of the goodwill of Audiobooks.com, while improving its operating leverage.

Meanwhile in 2024, the group exceeded its guidance, at the same time as it progressively improved its balance sheet, and as subscribers increased by 25% since 2023.

The fear of competition from Spotify, a potential competitor in 2025, has diminished, and it is now seen more as a distribution partner.

Bear thesis

The increase in competition

For me this is the major point of the bear thesis, which is why I am not entering the stock for the moment.

The arrival of Spotify on the audiobook market could weigh. The Nordics, Storytel’s main market, account for about 59% of the group’s revenue, with growth of only 2.4% in 2025 and subscriber growth of 3.6% over one year at the end of June 2026. Spotify integrated audiobooks into its Premium subscription in the Nordic countries starting in November 2025, and announced on 7 October 2026 that it is extending its offer from 22 to more than 180 markets by the end of the year.

Net subscriber additions are already slowing: 83,000 in the first half of 2026 versus 110,000 a year earlier, and 12,000 in the second quarter versus 32,000 (my calculation from the tables by segment). Management expects net growth in 2026 at least equal to that of 2025: about 129,000 net subscribers would therefore be needed in the second half, versus 102,000 in the second half of 2025.

YouTube is reportedly considering officially integrating an audiobook offer, potentially attached to its YouTube Premium subscription, as competitors such as Spotify do.

Growth through pricing is limited

The group’s ARPU is stable: price increases in local currency are absorbed by exchange rates and by the subscriber mix, with more subscribers in Europe, where ARPU is 72 SEK (about $7), versus 154 SEK in the Nordic countries (about $14). In America, ARPU falls by 6.5%. Europe’s gross margin declines (40.7% versus 42.1%). Growth therefore comes from markets where the subscriber brings in less, and Storytel cannot raise its prices indefinitely.

The price of acquisitions grows over time

Bokfabriken cost about seven times its operating profit, Overamstel about 11.9 times its 2025 EBITDA (14 MSEK), and when entering the US with Audiobooks.com, Storytel clearly overpaid for an acquisition that is not developing for the moment.

Bull thesis

We are talking about a company without debt whose margin is rising, which finances its acquisitions and its returns to shareholders with its own cash.

The company has a flywheel

Storytel’s cycle links a publishing activity (authors, rights, backlist) to a streaming platform (audience, data). See the section “The flywheel” for details. Its size is an advantage here. YouTube and Spotify are much bigger than Storytel. For groups of that size, buying a small publishing house changes nothing, and buying a publisher large enough to matter is difficult: the larger the publisher, the greater the risk of paying a high multiple. Storytel, on the contrary, can buy houses whose size matters to it, without endangering its balance sheet.

Furthermore, the company is profitable enough to finance its growth through acquisitions. After several purchases, the company manages to reinvest and to pay a dividend, with almost no debt, while increasing its operating leverage quarter after quarter.

The market is growing and Storytel still has a lot to conquer before reaching saturation

According to the group, the number of monthly active audiobook users in its core markets (Europe and North America) almost doubled between 2020 and 2025, and industry projections point to continued growth. Storytel has 2.75 million paying subscribers for an addressable audience that the group estimates at about 400 million people (130 million excluding the United States) in its ten core markets, i.e. less than 1%. It therefore still has a lot to conquer before saturation.

And the growth figures in Europe are good, with Europe’s revenue growing by 18.2% at constant exchange rates in the second quarter, including 12.8% more subscribers (1.07 million). Europe contributed 121,000 of the 185,000 net subscribers of the last twelve months.

Competitors have not stopped the trajectory for the moment

Spotify has offered audiobooks in its Premium subscription in the Nordic countries since November 2025, and yet the group’s growth at constant exchange rates is accelerating: 9.2% in 2025, 10.3% in the first half of 2026, 12.7% in the second quarter.

The giants of the sector have, moreover, made other choices: Spotify announced in November 2021 the purchase of Findaway, an audiobook distributor, and licenses catalogues, including that of Storytel in the Nordic countries, rather than buying publishers.

If competitors want to oust Storytel, they will not be able to remove its catalogue without harming themselves.

They would therefore have an interest in making a distribution agreement with Storytel, and that is what can be seen in the figures, with external Publishing sales outside the group growing by 31.6% at constant exchange rates in the second quarter of 2026. This is also the path Storytel is taking with its agreements with Klarna in the United States (for audiobooks.com), Vodafone in Turkey, or RDF Media in Chile.

Blind spots

Neither the licensing terms nor the contracts with Spotify are published, so it is unclear whether Spotify will be more of a competitor or a distributor in the future.

The agreements with Klarna, Vodafone Turkey and RDF Media

The group is not very clear on the churn rate, and the share of family subscriptions and the conversion of end-of-quarter trials are not documented.

5. Price scorecard

The numbers only make sense if you’ve read the section on the two theses.

As Aswath Damodaran puts it, “numbers without stories to back them up are exercises in financial modeling.”

The scorecard below requires an important assumption, namely, the continuity of a normal economic environment, i.e. without a recession, a major credit event, or a market crash. I will someday write an article about how to deal with these conditions.

Any of these could push the stock below the bear case shown here, temporarily, independently of the company’s underlying performance.

All figures in the scorecard are pre-tax and pre-fees. Run the numbers for your own situation before drawing conclusions.

The horizon is three years, in October 2029.

The assumptions are the following:

• Revenue growth goes from 0% per year in the bear case, to 7.5% per year in the central case, and 10% per year in the bull case.

• The margins go from 8.3% in 2026 (net margin) to 7% in the bear case, 10% in the central case, and 12% in the bull case.

• Payout ratio of 34% in the bull case, with none in the bear case; dividends are reinvested with no tax and no fees.

The selected range of P/E ratios displays the different estimated scenarios. We cannot display a 10-year history, the company having been loss-making over most of the last 10 years.

My conviction sits at the SEK 133 scenario, which isn’t enough for me to invest in the company, given the growth assumptions and the factors in the bear thesis.

The re-rating will mainly depend on the catalysts covered in the next section.

6. Momentum

Combining fundamentals (covered above) with momentum is an excellent way to improve your returns compared with using fundamentals alone. You’re never right on your own, especially since a significant part of returns comes from the multiple, particularly if your horizon is under 5 years.

The catalysts are the third-quarter results on 27 October, then the annual results on 10 February 2027.

As mentioned in the bull thesis, the development of acquisitions and the conditions of content distribution will be important.

On the competition side, the Swedish competition authority must rule on Bonnier’s acquisition of Nextory.

The arrival of YouTube in the audiobook field is another possibility in the bear thesis.

Final Thoughts

Storytel is a good example of why I put together my list of 48 quality small-caps. There are some genuinely interesting businesses out there that I would probably never have looked at otherwise.

investments. But the list is full of interesting businesses, across different industries and countries.

I have already invested in Baviera in the past, and I recently exited my position in EverQuote. For more on these companies, have a look at my previous article.

Not all 48 companies are cheap, and not all of them will turn out to be good

I suspect there are still plenty of good discoveries waiting for us. And that’s the whole point of the list: finding businesses worth understanding, and then waiting for the right opportunity.

Subscribe now

Disclaimer: This content is for educational purposes only and does not constitute investment advice. See our full disclaimer.

Last June, I published a deep dive on Booking Holdings, whose valuation was under pressure due to fears of disruption from AI. Then the Q2 results came out: no disruption in sight.

Then, on September 8, Meta launched Muse, and Booking’s stock price fell nearly 10%, with its PE ratio under 18, one of its lowest levels in recent years.

The purpose of this article is to see the threats that Muse and AI agents pose to Booking.

Subscribe now

1. Muse AI

What it does

Muse is Meta’s personal assistant, launched on September 8, 2026, initially in the United States. It can be accessed via an app, on the web, or through WhatsApp. It is powered by Muse Spark, Meta’s model.

Each user has a dedicated computing environment, and Muse takes action for the user, even when the app is closed. It can also connect to their email, calendar, or Instagram and Facebook accounts.To process payments, it uses Stripe Link, which provides a one-time virtual card limited to the approved amount. Each purchase must be approved by the user within the conversation.

Here’s a video from Meta showing how it works:

The app has been downloaded more than 9 million times since its launch. For now, it is only available in North America.

How Muse Books Trips

For flights, Muse connects to Duffel, a company that provides API access to airline inventory.

It allows users to search for, book, cancel, and manage flights through Muse, with human assistance available in case of any issues.

The reservation is issued through Duffel, while the airlines generally remain the primary merchant.

For hotels, Muse opens a browser and visits websites (like Expedia or Hotels.com), just like a human would. Expedia announced on September 22 that it would join Muse.

Duffel said that accommodations would come later. Booking, for its part, has not partnered with Muse.

Performance

Gizmodo published a test earlier this week, where the agent booked a restaurant (but one that was over budget) and failed to book flights: it couldn’t connect to Expedia; United Airlines (and Amazon) blocked bots, and the conversation proved slow and confusing.

Its performance could improve very quickly. For now, however, the experience is still far from perfect.

2. The easier the purchase, the more prone it is to disruption

To assess the risk of disruption, I distinguish between three levels of purchasing and how a personal agent can transform each one.

The first category includes routine purchases and changes in commoditized products, like curbside grocery pickup (Meta just partnered with Walmart), standardized products, cell phone plans, or basic insurance. Here, the products are comparable, and you know what you will get. This is where the agent can operate almost independently.

In case of error, the stakes are low. This is the area most prone to disruption.

In the case of travel, this would mean returning to a hotel where you are a loyal customer for a repeat stay.

The second level involves more complex processes and decisions, like comparing insurance policies, or switching providers when the procedures are more difficult, or filing a complaint with a company or government agency. The AI can run almost the entire operation, but this requires more interaction with the user, and the AI must be reliable. The potential for disruption is moderate, and many see this as a step forward for users.

The third level involves high-stakes purchases, like vacations, cars, or healthcare.

Here, the decision is too important and the information sometimes too sensitive to be entirely entrusted to an AI.

Booking a two-week family vacation involves a large budget, personal preferences, and the risk of a bad experience. Personally, I don’t want to ruin what is often the best time of the year (and the memories that go with it) because an AI that’s still in its early stages might have misunderstood or mishandled my requests. Especially when several thousand euros or dollars are at stake.

AI could become the universal consumer interface without necessarily replacing the companies that carry out transactions and provide services.

My core scenario is that AI plans and compares options, the traveler chooses and confirms, and then Booking makes the reservation and provides support. The agent can even process the payment through Booking. Other scenarios are also possible.

3. Booking’s Moat Evolution

The advantage of Genius remains the same

In the second quarter, Genius members at Levels 2 and 3 accounted for nearly 60% of Booking.com’s room nights, up from a year ago. Travelers still have every reason to take advantage of these benefits, which should be reflected in Muse’s comparison.

Brand Erosion

In my deep dive, I explained that Booking’s customers often return to the platform out of habit. This is a distribution moat, and likely the one that Muse could weaken the most.

If a traveler starts their search with a personal assistant, they may no longer think to open Booking. AI can influence which hotels are displayed and how they are displayed. Booking may still process the reservation, but it loses visibility early in the journey.

To counter this, Booking may need to invest more in its brand through ads and pay to increase its visibility among travel agents.

Direct Access to Hotels

Even though Booking continues to list 4.4 million properties, the temptation for a hotel to accept direct bookings with no commission is very real. European regulations already limit Booking’s use of rate parity clauses, and hotels can offer lower rates on their own websites. An agent could highlight these differences more prominently.

The risk for Booking is that hotels will have a new channel to reach their customers and thus gain greater bargaining power, even if they continue to sell their services through Booking. They could also use SiteMinder.

However, the hotel would have to handle sales and manage the service, which isn’t a problem for large chains but is more complicated for an independent hotels, the Booking’s core target market.

Trust and Problem-Solving: A Reason to Stay

I once tried to contact Abritel (Vrbo, part of Expedia Group) on a weekend because of a booking issue, but no one answered. I had to wait half a day to check out because the host didn’t respond to my messages. I swore to myself I’d never use that platform again. It was a bit less expensive than Booking, but in the end, it wasn’t worth it.

With Booking, I only had one issue out of more than ten reservations: I was billed twice. It was quickly resolved by customer support. I was actually glad to have a real person on the other end of the line who understood my situation and followed up with me to confirm the resolution.

My only major incident happened last year with Airbnb, at a beachfront house. Right in the middle of summer, we had an electrical problem. Despite our host’s reassuring words, the issue wasn’t resolved quickly. Airbnb refunded our reservation and added 10% so we could book another place to stay on the platform. That allowed us to bounce back without too much trouble.

For me, one of the main advantages of platforms like Airbnb and Booking is their ability to step in when a reservation goes wrong. An agent couldn’t provide that.

AI replaces part of the pre-booking process

Today, the “standard” process is to search on Google, then make a selection on Booking, then book the hotel on Booking. Tomorrow, the search and selection could take place on Muse, ChatGPT, or Gemini, followed by the reservation on Booking. That’s what I was explaining in the third image. The acquisition cost could change, depending on the scenario.

If the agent brings a customer who previously came through an ad, it is only a change in the type of acquisition.

If the agent brings a customer who was too lazy to book, it’s a new business.

If the agent directs the customer to the hotel’s website, and Booking is removed from the transaction, Booking loses a customer. That’s the worst-case scenario.

4. Competitors’ Response

Agentic AI and the commoditization of a competitive industry

I remain skeptical about the existence of a sustainable competitive advantage based only on automated web browsing. Google, OpenAI, and others are certainly also developing agents that will have similar capabilities in the future.

Agent-based transactions are likely to become a low-margin commodity.

In this case, Meta’s potential strength lies instead in the distribution and customization enabled by its applications. But a stronger distribution than its competitors? And if the brands and applications of Meta stop growing while others continue to grow, that marks the end of its distribution advantage.

Lola: the recommendation agent

Booking now offers Lola.com, which provides access to travel and experience bookings, with offers that remain within the group.

Lola is free for most uses, with a freemium model.

Vayo: Booking’s latest response

Yesterday, October 14, Booking announced a planned investment of $100 million in Vayo, a personal travel assistant.

According to the website, it uses the booking history in your email inbox to understand your preferences and offer you a personalized trip, without feeding your data to an AI.

It will allow users to book hotels and flights directly, and will redirect them to the hotel’s website or another platform if a better deal is available. Thus, Booking is seeking to maintain its customer relationships through this tool and appears to be engaging in a price war to preserve its distribution network.

A price war is never good for shareholders, at least in the short term.

How will Vayo be funded? It could be through subscriptions, advertising or transaction fees. Today, the question remains open.

Metrics

It will be important to monitor how these trends develop, particularly through upcoming reports from Booking, Airbnb, and Expedia, which are scheduled for November 3, 5, and 5, respectively. For Booking’s initial report, you can check out my latest Q2 update. The metrics to track are in the picture below:

Since Muse has just been launched and Vayo is still in private beta, these figures are likely to fluctuate in either direction over the next few quarters.

Here are the metrics to track:

Final thoughts

Muse shows that agents are at the dawn of a new way of making choices and placing orders, which will impact the travel industry. While there are certainly threats to its competitive moat, Booking has all the tools it needs to maintain or strengthen its competitive advantage over time. The risk of disruption is well offset by the platform’s security advantages and its ability to support the AI agent revolution within its ecosystem.

Valuation poses another problem, because as long as investors fear future disintermediation, they may refuse to quickly restore this multiple for Booking, despite solid results or the announcement of a promising product. It may take years to dispel this uncertainty, with no guarantee that the stock price will rebound any time soon.

I think Booking can remain a solid company in the age of AI agents. Whether investors will share my view next quarter, next year, or three years from now remains the unknown factor in this investment. But I’m in no rush.

We hear a lot about bottlenecks related to the rise of AI, particularly in the areas of electricity supply and memory production.

The other day, I came across an article about INNIO N.V., a company benefiting from the growing demand for decentralized power generation.

It caught my attention, so I decided to take a closer look.

In this article, I’ll walk you through the business, its products, competitors, and competitive advantages.

Then I’ll share what I took away from this research, and the bigger problem it revealed about investing in bottlenecks.

Subscribe now

1. What INNIO does

History

INNIO was carved out of General Electric. This is a transaction that separates a business unit from its parent, often through a sale to an outside investor or a separate stock market listing. How does this differ from a spin-off? In a carve-out, the parent company may retain a majority or significant stake. Advent International, a private equity firm, acquired what was General Electric’s Distributed Power business, which includes two brands, Jenbacher, based in Austria and Waukesha Motor Co. in Wisconsin.

The company went public in June 2026 at $27. The principal shareholder (44.6%) is now AI Alpine (Luxembourg) S.à r.l., co-owned by funds managed and/or advised by Advent and ADIA (the Abu Dhabi sovereign wealth fund).

What INNIO sells

INNIO sells gas engines ranging from 220 kW to 10.6 MW, fueled by natural gas, biogas, landfill gas, industrial gases, or mixtures with hydrogen. The group has two segments.

The first is Equipment, which is divided into three business lines:

-The data center business line delivers the prime and backup power required for AI workloads. This is the main area we’ll be focusing on during this Deep Dive, as it’s experiencing very strong growth.

Due to the massive growth of data centers, connecting to the power grid has become a major bottleneck. It takes years to connect a new site to the existing grid.

To bypass these delays, data centers can generate electricity on-site without relying on the public grid, and we call it “behind the meter”. These installations can serve as temporary solutions or remain in operation over the long term. Once supplied through the public network, we refer to it as “in front of the meter”

-The Power Solutions business line provides baseload and peaking power to stabilize utility grids (in-front-of-the-meter) and power independent microgrids (behind-the-meter). There are three main uses:

Electric utilities and independent power producers, to adjust to peaks of demand or low supply due to low wind or sun.

Factories, hospitals, and airports generate their own electricity, often through cogeneration. In this system, the heat generated by the engine is used to heat buildings or power an industrial process.

A third group of buyers recovers the gas that would otherwise escape or be flared. Engines convert this gas into electricity.

-The compression business line includes the engines that drive natural gas compressors to force gas through wells, gathering lines, and gas pipelines. I won’t develop much on this one, as it is just 8% of the total revenue and its part will shrink over time.

The second segment is the Services segment. To deploy and maintain the installations in the Equipment segment, it is necessary to provide services through qualified technicians and to organize the implementation. This segment includes spare parts (approximately 85% of service revenue), maintenance, overhauls, and multi-year contracts indexed to inflation.

Here is a summary from the company’s prospectus:

Clients

All of the clients are businesses.

In the Data Center segment, INNIO sells directly to hyperscalers, colocation providers, data center developers, and “Energy-as-a-Service” providers.

For Compression, INNIO sells exclusively through indirect channels, such as system integrators or distributors. Customers include operators of landfills, wastewater treatment plants, livestock farms, and oil fields.

For Power Solutions, INNIO’s subsidiaries sell directly, and distributors resell under the INNIO brand. Most have an exclusive contract limited to a specific territory, lasting one to three years. Major customers (“key accounts”) are not subject to exclusivity agreements. The customer base is very broad: electricity producers, public utilities, municipalities, factories, hospitals, airports, greenhouses, and operators of landfills, wastewater treatment plants, livestock farms, and oil fields.

Concentration on a few customers is increasing:

The top five customers account for approximately 39% in the first quarter of 2026; Clarke Energy, the top indirect customer, accounts for 15% of 2025 revenue;

VoltaGrid, an Energy as a Service Provider, accounted for approximately 41% of Équipement’s order intake in 2025.

Data centers accounted for 63% of equipment orders as of June 30, 2026, even though they represented only 21% of revenue for the half-year.

Management states that no single customer accounts for a disproportionate share of the order book and indicates most hyperscalers as customers.

Suppliers

INNIO uses two suppliers for most components and three for some, and manufactures critical parts in-house. Some equipment comes from a single supplier. Steel is the primary raw material, followed by copper, aluminum, rubber and plastics.

In most cases, INNIO does not purchase raw materials itself; its suppliers handle that. Their prices often include a cost-pass-through mechanism or an indexation clause.

Rare earth elements are needed in the manufacture of equipment. Restrictions on Chinese exports are a key risk.

The names and weights of the main suppliers are not disclosed.

Since August 2025, INNIO’s equipment has been subject to tariffs of up to 200 percent, corresponding to the proportion of aluminum and steel imported into the United States.

In response, the company launched the “local-for-local” initiative, which involves producing and purchasing within the markets it serves to minimize customs duties. This limits tariffs on trade between regions.

Competitors

Competition Overview

Here is a list of competitors ranked by how similar their technologies are.

Customers are looking for time-to-power, reliability (99.9% availability in data centers), fast startup (approximately 15 seconds), resilience to sudden load fluctuations, part-load efficiency, modularity, and the service network.

As long as there is a bottleneck in the electrical grid connection, the customer is paying for speed and reliability of power, and the price is less important.

Goldman has raised its forecast for on-site capacity to 67 GW by 2030, including approximately 18 GW of fuel cells worldwide, or about 27 percent.

Why gas engines, and not turbines, other engines or fuel cells?

The image below is from INNIO:

(1) Transient capability: ability to rapidly change power output in response to sudden changes in load or grid conditions while maintaining stable operation.

(2) Part load efficiency: calculated as the electrical output divided by the fuel energy input at a defined load level. It is commonly assessed by comparing efficiency at partial and total load.

With a gas-powered engine, an AI-based data center is essentially buying time. This allows the data center to start operating without waiting years for a grid connection. These engines aren’t the best in every aspect. Their strength lies in the balance they offer between delivery times, flexibility, and cost, especially compared to gas turbines, whose order books are full.

Here is another comparison chart of the different technologies:

A combined-cycle power plant is a facility that generates electricity in two stages using the same gas. First, a gas turbine burns the gas and drives a generator. This is the simple cycle. Next, the (very hot) exhaust gases pass through a boiler that produces steam. This steam drives a second, steam-powered turbine connected to a second generator.

The benefits cited by INNIO, Wärtsilä, and Bloom Energy come partly from the manufacturers themselves, so these comparisons deserve some caution.

Compared to turbines, gas engines are faster, more adaptable, offer better tuning performance, are available more quickly, and are less expensive.

The prospectus cites SemiAnalysis, which states that the larger the production unit, the more reserve capacity must be planned for: a factor that gives 5 MW engines an advantage over 50 MW turbines.

Large-scale and combined-cycle turbines remain better suited for centralized generation, larger installations and stable loads. Aeroderivative turbines compete with engines for peak load and rapid response.

Fuel cells can be more expensive (around $117/MWh, approximately 45% above the gas-turbine and combined-cycle alternatives), but offer higher efficiency, faster response times, and potentially shorter delivery times.

In short, there is no perfect solution here.

The best choice depends on what customers need, when they need it, and how much they’re willing to pay. And these choices may look very different in a few years.

Compared to other gas engines

Even within gas engines, the competition is tough and so is finding a decisive technological advantage in this group.

Competitive advantage

Here are the five typical sources of a moat.

• Intangible assets: low to moderate; no patents are essential.

• Switching costs: high after installation; this is the core of the moat (Services account for 48% of 2025 revenue).

• Network effect: none.

• Cost advantage: low compared to other engines.

• Efficiency scale: average, depending on the market.

INNIO’s products are good, but not unique.

Its competitive moat is narrow and comes primarily from the switching cost of the service segment.

The equipment order backlog and production slot reservations exceed 15 GW, more than four times the power delivered over the past twelve months.

According to the CEO, 2026 and 2027 are nearly fully booked; available slots are in 2029 and 2030; and projects are already being discussed for 2030 and 2031. He also explained that if demand is strong and supply is limited, the company is always in a strong position when it comes to pricing, and he sees no signs of a decline.

But this strong position could be temporary and linked to a surge in business. How long these conditions will last is another question.

That’s not enough for me to invest in this company, as there are too many questions.

I usually devote the rest of my deep dives to examining the financials, management, bear and bull cases, and then to setting a price and reviewing catalysts.

However, bottlenecks are among the most discussed investment opportunities on Substack and X, from electricity and memory to photonics.

So I think it’s worth going a step further.

Why am I passing on INNIO despite its impressive growth? And more importantly, how should investors approach these bottlenecks in the first place?

That’s what I’ll explore in the second part of this article.

Subscribe for free to discover the six questions I believe every investor should ask before investing in a bottleneck and how I decide between stock picking, ETFs, and simply staying away.

Subscribe now

Small companies that few people follow get far less attention, and some of them have the numbers of a future giant.

I ran the screen across 21 Western countries and 48 companies passed.

Small-cap stocks are among the best opportunities in the market: they are covered by fewer analysts and often too small for an institution to build a position.

Less attention leaves more room for mispricing.

They start with a liquidity discount and become more closely tracked as they grow, and the discount vanishes.

Warren Buffett has pointed this out many times in the past:

Small-caps are interesting, but not all small companies are equal. The higher-quality end of the universe has historically outperformed over long periods.

The result is even more striking when you look at some of the best individual winners that grew significantly over time.

In Berkshire Hathaway’s early years, Warren Buffett also invested in a high-quality small-cap: See’s Candy, in 1972.

That is the kind of profile I am looking for: a small business that can grow without needing proportionally more capital.

What’s Included in the List

I left out entire industries for several reasons:

Too complex, like banks, asset managers or too speculative, like biotechnology.

Too risky, like passenger airlines and commodity producers, with low pricing power, or chemicals and trading companies, where margins swing with the cycle.

Against my principles. Tobacco, defense, and gambling.

The idea is to keep the list understandable, especially for newer investors, without filling it with businesses where a screen can look great right before the cycle turns.

Here are the selection criteria for the Tier 1

It is deliberately restrictive. This selection naturally excludes cyclicality, or companies that could also perform well, but with more risks, like value or deep value.

I also provide a Tier 2 for companies that meet the other criteria, but have increased their share count by less than 2% per year over the past three years. That still leaves room for some excellent businesses while keeping dilution limited.

Moat Score

The screen gets me to the first group of companies, but good numbers are only the starting point.

I also give each company a Moat Score, a personal assessment of the competitive advantage of the firm, which explains how the company will be able to protect its future profits. I use it to rank the companies that pass the financial filters.

The screen tells me whether the numbers are good; the Moat Score asks why they might stay good.

The companies are ranked by Moat Score, then alphabetically.

Three names already covered

• Baviera: I recently published a full deep dive on the company.

• EverQuote: I owned it and followed the company quarterly. I sold in September for reasons unrelated to the fundamentals.

• Kri-Kri: Here is a link to an excellent article by Joe | Goldman Won’t Cover This.

I will also publish a new deep dive on one of the other companies on this list over the coming quarters.

Why I won’t keep the full list public forever

Many of these companies have a market cap under $1B, and most trade only a few hundred thousand dollars a day. So an article that became too popular would end up backfiring on its readers, and I need to prevent too many buyers from driving up the price because they all come from the same source.

That’s why the full list is free for the first 500 subscribers.

After that, it will move behind the paywall and remain available to paid subscribers only.

You’ve seen 3 of the 48 companies. The other 45 are below.

SUBSCRIBE FOR FREE TO SEE THE OTHER 45

Subscribe now

The market is full of extremely attractive small-cap stocks with a long-term growth outlook and appealing valuations. These are compounders to hold for years, providing a stable foundation for your portfolio. The real difference isn’t in making 50% a year for 3 years, but 20–30% for 10 years or more.

And the company I’m recommending today is one of them. Far removed from the AI hype and its speculative valuations, it’s a long-term play in a sector with a lot of room to grow.

This is the logic of Chris Mayer’s book, *100 Baggers*, which examined 365 stocks that increased 100-fold between 1962 and 2014.

Mayer’s criteria are described in the image below:

And the company I’m suggesting to you meets almost all of the criteria.

Subscribe now

It is Baviera, a Spanish eye surgery group, formerly Clínica Baviera.

The structure of my deep dive is standard. Learn the process in the dedicated articles in the learning section.

The whole article is long because it covers a lot of ground, and short because I try to give you a synthesis of the points. So I start with a “short version” for those of you in a hurry.

The Short Version

Baviera is a Spanish eye surgery group that has grown steadily for almost twenty years. It has not raised capital since its 2007 IPO, it earned 21.6% on invested capital in 2025, and its new clinics are paid for by its own cash flow.

The growth is not spectacular, but it is steady. And what could happen to it? The business is very solid, and it relies on opening new clinics and making acquisitions to expand internationally or consolidate its position in a given country. The market is regulated because it requires doctors to perform surgeries. They have already succeeded in doing so in Spain, where they are the market leader; they have established a presence in Germany and Italy; and they are well on their way to establishing a profitable presence in the United Kingdom.

I think Baviera is the kind of stock you can hold for ten years and end up with around four times your money, with peace of mind.

1. What Baviera does

History

Baviera was founded in Valencia in February 1992 as Clínica Oftalmológica Baviera, a single eye consultation center started by Dr. Julio Baviera and his family. Starting in 1996, the company began opening clinics throughout Spain, at a rate of about three per year beginning in 1997. The company was led for nearly thirty years by Eduardo Baviera, Julio’s brother.

The 3i fund, which currently owns the Action stores, acquired a stake in the company in 2005 and then sold all of its shares during the initial public offering in March 2007 on the Madrid Stock Exchange. At that time, the group had 36 clinics, all located in Spain.

The company’s expansion into Europe took place in June 2008 with Baviera’s acquisition of CARE Vision Germany GmbH and Care Medical Services Holland, giving it a presence in Germany and the Netherlands. By early 2009, the group had 15 clinics in Germany, the Netherlands, Italy, and Austria.

In August 2017, the Chinese group Aier Eye Hospital acquired an 86.83% stake in the company. It has since sold a portion of that stake and still holds 73.23%.

In July 2024, Baviera acquired Optimax, a British group of 19 clinics that was undergoing restructuring. Finally, in December 2025, the company was renamed simply Baviera. Luis Grávalos succeeded Eduardo Baviera as CEO on March 26, 2026.

Clients

Baviera treats patients who come in for various types of treatment:

Refractive (laser) surgery to treat presbyopia, myopia, hyperopia, or astigmatism. These are generally patients who want to stop wearing glasses. They pay for the procedure in full, or almost in full, with a reimbursement or a discount from an insurer in such cases.

Cataract surgery: replacement of the natural lens with an intraocular lens, for older patients. Cataracts are covered by the public health systems, so who pays depends on the country. In Spain, Baviera presents private surgery as the answer to public waiting lists, and its patients pay themselves or go through an insurer that has an agreement with the group. In Germany, statutory insurance pays for the operation with a standard lens, and the patient pays for premium lenses.

Patients also come in for other conditions related to general ophthalmology, for consultations and examinations.

In 2025, about 16% of ophthalmology revenue came from agreements with insurers, which provides information on the overall level of patient reimbursement.

The cost of the surgery will depend on the technology used by the healthcare professional, and does not depend on the refractive error being corrected (myopia, hyperopia, etc.). The company also offers 0% interest loans through its partners.

The company, which began in Spain, subsequently expanded into Germany, Austria, Italy, and the United Kingdom, as outlined in the presentation below.

The customer base is very spread out: Baviera received about 180,000 patients in 2024 and says it has performed some 2 million operations since it started. In 2006, about 60% of new ophthalmology patients in Spain came on the recommendation of former patients. Because a laser operation is bought once, growth depends on winning new patients every year.

The data I could find indicate that the market is growing by about 5% a year.

Suppliers

Baviera’s main resource is its doctors, and most of them work for the group as employees. At the end of 2025 it employed 272 doctors: 224 in Spain, 38 in Germany, 9 in Italy and one in the UK. Almost all have permanent contracts, and just over half work part-time. In the UK the doctors are not employees: they work under a commercial contract with the group. The group also calls on outside collaborators, doctors and companies, to cover peaks in demand.

Employed doctors cost €49.2 million in 2025 and outside doctors are included in the €23.6 million “professional services” line item, which could include all service providers, including the housekeeper and the plumber.

Like any ophthalmology center, it requires medical equipment, which is divided into consumables and capital assets (for example, from Bausch & Lomb or Carl Zeiss Meditec). Replacing a machine is complicated because the entire operation relies on mastery of that machine, and there is a risk of delays or failure.

Almost all clinics are rented, so landlords are suppliers too.

Competitors

Competition in this type of business is local, since people generally don’t walk 500 miles to undergo a procedure that is fairly common.

For non-reimbursed procedures, competition comes from other ophthalmology groups, which are listed below by country:

In Spain, it’s mainly Miranza, launched in 2020, and it reached €128 million of revenue in 2025, vs €204 million for Baviera. Our company is more standardized and highly advanced in refractive surgery, while Miranza positions itself more as a network of centers of excellence offering specialized care.

In 2024, the ophthalmology clinic market was worth approximately €720 million, with Baviera holding a market share of about 25% and Miranza holding 15%.

In Great Britain, there is Optical Express, the leader in laser technology, with £146.5 million in revenue in 2022 vs £16 million for Baviera in 2025.

In Germany, while Baviera generated €59 million in 2025 with 32 clinics, there are numerous alternatives, including Sanoptis, with its 220 locations, Artemis (120), Veonet (84), and Augenzentrum Eckert (72).

In Italy, Baviera Italia accounts for approximately 7% of the company’s 2025 revenue with its 11 clinics, or about 21 million euros, less than Vista Vision, its main competitor, which generates 65 million euros with its 12 clinics, with a broader range of ophthalmology services.

In most cases, the market remains fragmented. When it is not large groups, it consists of individual practitioners or small groups of a few practitioners.

Competitive advantage

Since 91% of revenue comes from surgical procedures, this section focuses primarily on that area, with the remainder split roughly evenly between consultations and other conditions (retina, oculoplasty, glaucoma, etc.).

There is little competition with the public hospital as laser surgery and presbyopia treatments are not covered by public systems, so Baviera competes there with private clinics.

Cataract surgery is different: the public system pays for it, with a standard lens and no bill for the patient. What Baviera sells on top is choice of the date, the surgeon, and a better lens if the patient wants one. The average wait for cataract surgery in the public system was 68 days at the end of 2025, and only 2.6% of patients waited more than six months.

Critics also argue that private clinics take the simple, profitable cases and compete with public hospitals for the same doctors. I found no data showing that Baviera weakens the public system.

As for barriers to entry, this is an industry that depends on the number of practitioners, and regulations prevent unregulated entry into the market. Not just anyone can put on a white coat and become an ophthalmologist overnight. And currently in the EU, it often takes quite a long time to find a doctor.

As a result, there is cost pressure to retain doctors within these facilities.

Compared to practitioners working alone or in small groups, the opportunities for reinvestment and cost-sharing result in higher profitability.

The company therefore enjoys, like its competitors—the other groups—undeniable advantages:

• brand visibility,

• very extensive network,

• more consistent processes,

• a strong clinical focus on refractive surgery,

• greater transparency regarding prices and financial performance.

It’s a very profitable business with an ROIC ranging from 20 to 30 percent depending on the year (except for 2020, but that’s because of COVID), far above any reasonable cost of capital.

Buyers are scattered and have no bargaining power. The same goes for insurers, which are fragmented. It’s a network that grows through word of mouth, so switching costs are based on reputation. Buyers don’t have any particular incentive to return. People don’t visit an ophthalmologist regularly to have surgery.

Furthermore, this is a request related to operations that will depend on economic conditions. Although it is a healthcare expense, it can be deferred so that the cost is covered later.

And while this may seem obvious, it is a key aspect of any competitive advantage: customers do not automatically seek out the cheapest option; instead, they prioritize quality and safety.

Here is the summary slide. Note that this image should be read against the alternatives.

2. Financials

Key financial metrics

Watch out when interpreting the figures; these are for a half-year period in 2026. You must double the amounts and adjust for one-time items to compare the years.

Commentary

Revenue

Since its IPO, the company has grown in revenue at a rate of 9% per year. One observes that growth has been slowing since 2023: the trend varies as a function of the opening of new clinics.

Useful clarification: the figures published for 2024-2025 overestimate growth by a few percentage points because of the integration of Optimax (United Kingdom). In the first half of 2026, without revenue from the UK, growth would have been 8.8%.

Operating Margin

The EBIT margin (which accounts for most of the rents in depreciation according to IFRS standards) declined in 2024 and 2025, notably because of the acquisition of Optimax in the United Kingdom. In Germany, profitability fell because of the increase in salaries paid to retain doctors. In Italy, it decreased because the group hired staff in anticipation of its openings and increased its marketing expenses to move from a B2B model to a B2C model. Spain experienced the opposite evolution and recorded an improvement.

The first half of 2026 highlighted an operating leverage effect: expenses increased by 8%, while revenue grew by 10%.

Net Income and Free Cash Flow

The slowdown in net income growth in 2025 is explained by an entire year of losses in the United Kingdom, still linked to the integration and turnaround of Optimax, as well as by a higher tax rate (28.6% versus 26.5%).

The company compares the year 2025 to a pro forma 2024 fiscal year, which gives a net income growth of +15%. The published comparison indicates +4%, and it is the one that should be used. The number of shares remaining unchanged (16.3 million), earnings per share therefore simply follows the evolution of net income.

Baviera also pays a negligible sum in respect of share-based compensation, of approximately 0.44 million euros per year, and 163,043 stock options at 23 euros (1% of the shares) will become exercisable in 2027.

In the first half of 2026, the financial result represents about two points of the 17% increase. The decline in free cash flow (FCF) in the first half of 2026 (−1.6%) is not worrying, because it reflects the timing of payments. Meanwhile, the operating income rose by 12%. After lease payments, the group converted between 66% and 84% of its net income into cash between 2023 and 2025. Quite a good score!

Balance sheet

As of June 30, 2026, Baviera reported a net financial position of +41.7 million euros after paying a dividend of 25.1 million euros during the first half of the year. Including lease obligations, net debt amounts to 19.4 million euros. And given its EBIT, this debt can be repaid in about four months.

The assets held by Baviera consist mainly of what a clinic needs to operate. It would be an overstatement to call this an “asset-light” business, where capex is 3-5% of revenue. Baviera’s capex, spent on medical equipment and clinic fit-out (the interior renovation of a rented space to turn it into a clinic), is closer to 10% of revenue.

3. Management

The new CEO

Eduardo Baviera, the company’s co-founder, led Baviera for 30 years. There is no information available regarding the reasons for the change in CEO. It is reasonable to assume that, after 30 years at the helm and at nearly 60, he would like to slow down while maintaining a more behind-the-scenes role within the company. He remains on the board as vice president and continues to oversee strategy and medical direction.

Luis Grávalos Soria succeeded him on March 26, 2026. He has spent more than 16 years with the group. He first led the Germany division, then the United Kingdom division following the acquisition of Optimax. The board says it is appointing him “to drive a new phase of growth, while maintaining the continuity of the model that has been successful in recent years and strengthening a professional, international management model focused on responsible growth.”

The data does not indicate that the successor created a plan. The only guidance is the board’s statement of continuity and a new growth phase.

Yellow flag: Not only can I find no record of the CEO currently holding shares in the company, but I also see regular sales of shares dating back to 2025. The only known new incentive is the long-term cash bonus approved on May 27, 2026. It pays out in cash, with no shares involved, and neither the amount nor the conditions have been disclosed. Alignment with management cannot be assumed.

The Majority Shareholder

A Chinese group controls Baviera. Aier Eye Hospital Group, based in Changsha, is China’s largest ophthalmology chain and is listed on the Shenzhen Stock Exchange. It acquired 86.83% of Baviera through a tender offer in August 2017. Its stake has since fallen to 73.23%. Its founder, Chen Bang, serves as chairman of both groups. Having a majority shareholder (furthermore a Chinese one) in the company has several implications:

First, the positives. The company is not at risk of a hostile takeover, and Aier is an industrial shareholder in the same industry with a long-term perspective. Also, the company has been doing well for the past nine years with this majority shareholder.

Next, the disadvantages.

Management is in charge of day-to-day capital allocation (clinic openings, capital expenditures), but the board (on which Aier holds four of the seven seats) has the final say.

Liquidity is limited, with a free float of about 18% of the market capitalization. A large seller can cause the price to drop sharply. The same is true on the upside. This also hinders the stock’s inclusion in indices and institutional portfolios.

Having a majority shareholder always carries the risk of delisting. Any significant drop in the stock price (even with solid fundamentals) would be a bargain for the majority shareholder, which could buy out the minority stake at a favorable price (with a premium on the current price), at its sole discretion.

There are currently no major tensions with China, so a scenario of serious deterioration of relations with China is one with significant impact but very low probability. The recent tensions between China and the EU have had no impact on the company.

Capital allocation

Baviera pays for its growth, its dividend and its acquisitions out of its operating cash flow.

Openings come first: capex was about €30 million in 2025, close to 10% of revenue: €15.9 million for new clinics and €14.3 million to renovate existing ones. The group opened 11 clinics in 2025 and 8 in the first half of 2026, taking the network from 148 to 156.

The dividend comes second. Baviera has paid €1.57 per share in each of the last two years, a payout of 68% of 2024 earnings and 62% of 2025 earnings. The cash paid out was €25.9 million in 2024 and €26.4 million in 2025.

Management knows how to seize acquisition opportunities to keep growing, notably abroad. Examples include the acquisitions of Optimax (UK) and Care Vision (Germany). Although Optimax is not yet profitable, we can expect it to become so over the coming quarters, as the business is turned around.

4. The two theses

I believe the best way to decide whether an investment is sound is to pit the bullish and bearish theses against each other. That way, I know what could happen, and knowing your position well helps you endure downturns and protects you psychologically. Considering only the bear thesis is alarmism; considering only the bull thesis is wishful thinking. This section includes both the arguments and the catalysts for each version of the story.

Why did the stock go up?

I usually write a post explaining why the stock price has fallen. But in this case, there’s no problem; the price has been rising steadily since 2023.

This chart shows that the rise in 2025 is driven primarily by the P/E ratio, while the rise in 2026 is driven by earnings growth. The rise in 2024 is driven by both factors.

One might think that the rise in the 2025 multiple stems from the doubling of the free float in April (from 10% to 18%), which reduced the liquidity discount. Another reason is the relatively low starting point for the P/E ratio in early 2025, which was only 13.

Currently, the company is trading near its highest relative valuation levels, with a P/E ratio of 19.

Bear thesis

A technology that reduces the need for laser surgery

We haven’t yet found a cure for presbyopia, myopia, hyperopia, or astigmatism, but who knows? Maybe one day there will be a treatment that can prevent these vision problems. If that happens, it will be the end of Baviera’s core business.

Currently, alternatives such as implantable lenses do exist, but they still require the assistance of a doctor, so this would simply be a shift in technology rather than a true replacement. A 2017 analyst report (Renta 4) indicates that intraocular lenses already accounted for 40% of revenue, so this is a well-established practice.

There’s also the issue of preventing nearsightedness, but if glasses are still needed, that doesn’t change the bottom line: people who undergo surgery want to live without glasses.

People prefer glasses

More people keeping their glasses means fewer surgeries, and therefore lower revenue. In the case of cataracts, surgery is essential to prevent blindness, so revenue from cataract procedures remains steady. The reports do not specify the portion of revenue attributable to cataract procedures.

No recovery in the U.K.

Revenue in the United Kingdom still accounts for a small portion of the group’s total (€10.6 million in H1 2026, or 6%), but investors expect this segment to become a new growth driver. There are 19 clinics out of a total of 156, representing a potential 12% of revenue. If the turnaround is successful, this metric is expected to double. Profitability is a goal for 2027 that has yet to be confirmed.

A price war or higher competition

That is not currently the case for the price war, and healthcare services tend to be priced consistently, but this scenario is still possible if one of the competitors were determined to win the market at any cost. This is especially true since, in most cases, patients pay out of pocket.

For higher competition, this would involve in particular poaching doctors in the event of a shortage of job openings.

Economic conditions that could cause operations to be postponed

These are mostly non-essential operations that can be postponed. The economic situation in Spain has improved significantly in recent years, but that period of improvement may be coming to an end. Spain accounted for 68% of sales and 81% of EBITDA in the first half of 2026, with 52% of the clinics located there, so the company’s fate remains closely tied to the economic situation in Spain.

No opportunities left for M&A in Spain

In Spain, a transaction must be notified to the competition authority (CNMC) when the combined market share reaches 30 percent, as defined by product or service and by geographic area. This may refer to a sub-segment (such as laser surgery rather than ophthalmology as a whole) and “a substantial part of the territory.” With a 25% market share, Baviera could be affected, as it would have to grow solely through organic growth. It is a notification, and the deal could still be made (or not), depending on the regulator.

Bull thesis

The UK is on the road to recovery, and Italy continues its momentum

UK: In the first half of 2026, revenue rose 28% to 10.7 million euros, and the EBITDA loss narrowed from 3.0 to 1.6 million euros compared with the first half of 2025 (+1.4 million euros in a year), even as the company continued to hire.

Italy: The network grew from 8 clinics at the end of 2024 to 11 in June 2026, with revenue up 9% year-over-year and EBITDA up 25%.

Changes in technology or alternatives are fairly stable

I find it hard to imagine people collectively changing their minds about glasses, whether to wear them or have surgery. In a 2024 survey commissioned by a manufacturer of implantable lenses, 88% of Spaniards said they wanted to live without glasses. Demand shows no signs of waning, and people continue to seek surgery if they can afford it.

As for the possibility of alternatives that would eliminate the need for glasses, such as a treatment that cures nearsightedness or something similar, we can only wait and see for now.

Spain is doing well, and other countries are growing

GDP grew by 0.7% in the second quarter of 2026 (2.7% YoY), driven by domestic demand: household consumption rose by 3.2% over the past year. The unemployment rate fell to 9.87%, below 10%. There will come a time when Spain’s economic performance will slow, and this could affect Baviera. For now, Spain is doing OK.
But until then, other countries continue to grow, driving the push to internationalize sales so that Baviera’s growth rests on multiple pillars. Today, sales in Spain account for 68% of total sales; tomorrow, that figure could be 40%, then 20%, to reach in the long term (or come close to) the revenue share within the EU.

The Moat grows stronger over time

I don’t see any competitive advantage of Baviera over other clinic groups; the main factor is the size and operational efficiency of management. But in terms of competitive advantage over other smaller players, Baviera operates 156 clinics under a single brand across all its countries, has a global advertising budget of 15.1 million euros (which a small practice cannot match), and offers enormous opportunities for profitable reinvestment, with a return on invested capital of 21.6% in 2025.

I’ll finish with the key point of the bull case:

There are still significant opportunities for new clinics

The network has 156 clinics, and the goal reported by the press is 180, about 16 new openings per year. There is still room for growth in countries where the group already operates: Germany has 34 clinics, the United Kingdom has 19, and Italy has 11, compared to 92 in Spain.

Beyond that, there is the rest of the EU to expand into: France, the Nordic countries, the Benelux countries (and Switzerland). Baviera has not announced any plans for these countries, but such expansion remains entirely possible.

Each clinic opening costs approximately 1.4 million euros and is funded by operating cash flow, without borrowing, and generates approximately 0.4 million euros in free cash flow per year. The payback period is therefore approximately 3 years.

When it’s not a new location, it may be an acquisition or a merger. Baviera has already completed three such transactions: Care Vision in Germany in 2008 (8 centers in Germany and 1 in Austria), Clínica Torre Eva in Milan in 2019, and Optimax in the United Kingdom in 2024.

Even in Spain, the two largest firms account for about 40 percent of a market worth approximately 720 million euros, meaning that more than half goes to other firms and independent practices, which remain a driver of growth. Organically though, as mentioned in the last point of the bear case.

Opportunities for external growth remain strong, and the sector could decide to consolidate in order to increase economies of scale and brand strength, including Baviera. The sector remains highly fragmented (outside of Spain), so the prospect of regulatory authorities blocking such moves is still a long way off.

Blind spots

• We cannot yet assess the M&A opportunities in the target countries and the consideration of the regulator

• Will the CEO be successful? We could end up with a “Steve Ballmer scenario” at Microsoft, where the company’s stock price stayed flat for over a decade.

• Will the company seek to expand internationally?

• It is quite difficult to assess the performance of the other groups, which are private and therefore don’t disclose information.

5. Price scorecard

The numbers only make sense if you’ve read the section on the two theses.

As Aswath Damodaran puts it, “numbers without stories to back them up are exercises in financial modeling.”

The scorecard below requires an important assumption, namely, the continuity of a normal economic environment, i.e. without recession, major credit event, market crash. I will someday write an article about how to deal with these conditions.

Any of these could push the stock below the bear case shown here, temporarily, independently of the company’s underlying performance.

All figures in the scorecard are pre-tax and pre-fees. Run the numbers for your own situation before drawing conclusions.

The horizon is in three years, in October 2029.

The assumptions are the following:

• Baviera will open 6 clinics per year in the bear case, +8% openings per year in the central case, and +10% in the bull case. The last one corresponds to the historical trend of the company.

• For the revenue, it goes from 4.3% per year in the bear case, to 8.2% per year in the central case, and 10.1% per year in the bull case. Revenue per clinic is assumed to stay flat.

• The margins go from 14.8% in 2026 (net margin) to 12% in the bear case, 15% in the central case, and 17% in the bull case.

• Payout ratio of 60% in the central case and 55% in the bull case, with none in the bear case; dividends are reinvested with no tax and no fees.

The selected P/E ratios range from the 2022 low to the June 2026 high, and add the possibility of raising the valuation ratio to 25, taking into account the company’s market capitalization exceeding €1 billion and reducing its liquidity discount.

My conviction sits at the €91 scenario or higher (bull case x PE of 20).

The re-rating will mainly depend on the catalysts covered in the next section.

6. Momentum

Crossing fundamentals (covered above) with momentum is an excellent way to improve your returns from doing just fundamentals. You’re never right on your own, especially since a significant part of returns comes from the multiple, particularly if your horizon is under 5 years.

Catalysts

Results for the upcoming quarters: It is necessary to verify whether the overall figures are in line with the forecasts in Part 5, namely, approximately 8–10% revenue growth and a stable margin of around 15–17%.

Any downward trend will need to be examined; in that case, I’ll post a deep-dive update.

Evolution of the low-margin countries: In addition, the United Kingdom and Italy must report margin improvements and increases in revenue.

Announcement of acquisitions: Baviera has made three acquisitions since 2008 (Care Vision in Germany in 2008, Clínica Torre Eva in Milan in 2019, Optimax in the U.K. in 2024), all small, and none is announced today. Its net financial position and profitability leave room for more.

A deal would be the first sign that the last point of the bull case is starting to play out. So far, Baviera has acquired a group to establish a presence in a new country, and may continue to do so.

Increase in the free float by Aier: Aier holds 73% of the capital and the free float is only about 18%. In April 2025, Aier and Vito placed 8% of the capital with a discount of 8.7%, which took the float from about 10% to about 18%. The stock first fell 5.3%, to €33.80, then rose. Aier has sold in steps before, so another placement is possible, but it has said nothing. A new placement would improve liquidity and open the shareholder base to new funds. In the short term, it would put pressure on the price, and would be a good time to add to the stock, all other things being equal.

Review of the strategic plan by the new CEO: For now, Luis Grávalos Soria is following in his predecessor’s footsteps, but he may introduce a new plan whose content and execution could drive the company to a new level of growth.

When to enter the stock?

The main risk for the stock is a decrease in the PE ratio, which could be a good opportunity (depending on the reasons for the selloff) to build a bigger position, because usually, in this situation of top PE, I could start a small position.

From that point on, the price may fall solely because of the multiple, in which case you should increase your position when the decline stops.

For my part, I prefer to wait until the stock price drops before buying more, which also gives me a chance to see and hear what the new CEO has to say.

Final Thoughts

This is a company with good prospects over the next 10 years or more. The growth isn’t spectacular: it’s not Nebius Group. But the model is simple and funds itself. New clinics are paid for with operating cash flow, any excess is available for organic growth, dividends or new acquisitions, and the return on invested capital was 21.6% in 2025.

But it’s the kind of stock you can hold onto for 10 years and potentially end up with 4 to 5 times your initial investment.

As the company had a price of €8 during Covid, it is up 7-fold since. If it keeps up the pace of fundamentals and quadruples in value over 10 years, it will have grown 28-fold. Another ten years or so in the same direction, and it will be a 100-bagger in 26 years, not from this point though.

I see a clear long-term advantage for this company that supports the bullish case over the bearish case. But it will take time. Compounders come out ahead in the long run. This one, with peace of mind as well.

Subscribe now

You liked this article? You may also like the list below, within which Bavaria is located

Disclaimer: This content is for educational purposes only and does not constitute investment advice. See our full disclaimer.

We have three bank cards in our European household and we will keep only one, because we won’t need more.

It all began in April 2024 when I started taking up positions on Visa and Mastercard. I was up 30% one year later, but I decided to exit both, rotating into Nvidia and Alphabet.

The main trigger was an event that I saw coming for them in the European Union; and that I still haven’t seen written about anywhere, not even on Substack.

I am convinced that in a few years, the developments described in this article will come to pass (at least in the EU) with major implications for Visa, Mastercard, and PayPal.

Subscribe now

I. The flywheel

The spectacular growth and competitive advantage of Visa and Mastercard are based on the following flywheel: more issuers attract more merchants, more merchants attract more cardholders, more cardholders generate more transactions, more transactions produce more data, more data enables better value-added services, and better services attract more issuers.

And so on.

Once the scale effect is reached, you’ll have a network of 150 million acceptance points across 200+ countries, with switching costs, that are based on decades of embedded infrastructure in every bank’s core systems.

Visa’s FY2025 results look solid, especially in the United States, where you get rewards: miles, cashback, airport lounges.

II. the vectors of the broken flywheel

1. In EU, fees is all you get

In the US, Visa and Mastercard can afford incentives. In the EU, it’s just a commodity to pay, as the regulator capped interchange (0.20-0.30%), and rewards barely exist.

So when an alternative shows up, nobody has a reason to stay and the cheapest option wins.

Germany is ahead on most tech topics, but not on payments: cash is still widely used and a lot of transfers go through the banks. Paypal commands approximately 30% of German e-commerce transactions, the dominant payment method online. That position was built on a single use case: I want to pay online without giving my card number, with buyer protection.

There is, of course, Wero, but the system launched in 2024 and backed by European banks offers exactly the same features as Paypal. The only benefit is simply that it is a European public service. An “economic patriotism” in short, without much interest beyond the love of the flag.

2. A2A: the disruption already underway

Disruption has already begun in two countries with national initiatives. Credit cards aren’t the only losers; cash is also a collateral victim of the operation. But that doesn’t translate into investment losses for anyone for the latter.

PIX, Brazil, launched November 2020

The initiative was mandated by the Banco Central do Brasil. It is free for individuals; merchants pay 0.33%, compared to 2.3% for cards. Participation is mandatory for any institution with more than 500,000 accounts.

Five years later, in 2024, there were 64 billion transactions (exceeding the combined volume of Visa and Mastercard) and 93% of Brazilian adults were using PIX. A resounding success.

Meanwhile, the use of cash fell from 43% to 6% of all payments. Visa and Mastercard incured combined losses of R$12B between 2021 and 2024, or nearly $2.3B.

It was such a success that in September 2025, the U.S. Trade Representative launched a formal investigation into PIX, calling it an “unfair competitive disadvantage” for U.S. card networks, a claim echoed in a White House report in April 2026. President Lula responded: “No one will make us change PIX.”

UPI, India, launched in 2016

Nine years after its launch, UPI had processed 644 million transactions in a single day in June 2025. In May 2025, the value of UPI transactions was 12 times greater than the cumulative total of all card transactions in India. UPI is growing at an annual rate of 30–40 percent, compared with 10 percent for Visa. Another resounding success.

3. The Digital Euro: Once Bit, Twice Bit, Three Times Bit

The digital euro is a central bank digital currency (CBDC). It is a form of public currency issued directly by the European Central Bank, with the same legal status as cash.

It is a sovereign currency, backed by the ECB, and accessible to all citizens and businesses in the eurozone.

Currently, when you pay by card, you are using bank money (i.e. private money) processed through a private infrastructure owned by U.S. companies. The digital euro would be the first form of public digital currency directly accessible to EU citizens. It would be the equivalent of a banknote, but in digital form. So it is not a stablecoin, but an official version of a currency.

The ECB has made it clear that its goal is to reduce dependence on foreign payment infrastructures.

In practice, you store your digital euros in a wallet provided by your bank, a post office, or any authorized payment service provider. You can make payments in stores, online, or between individuals, whether online or offline. No internet connection is required for offline transactions. There are no transaction fees for users.

Here are the main design features confirmed by the ECB. These may change between now and the go-live date.

Acceptance is mandatory for all merchants in the eurozone from day one. Transactions are off-line, and only the payer and the payee know the details; the Eurosystem cannot directly link transactions to individuals. In principle, holding limits would be capped at €3,000 per individual. Cash and existing payment methods would still be available.

As for the timeline, following a study phase in 2021 and preparatory phases, on July 9, 2026, Parliament approved its negotiating mandate in plenary session by a vote of 416 to 169, with 22 abstentions. Next, a political agreement is expected by the end of 2026, then a 12-month pilot from the second half of 2027, and first issuance as early as 2029 if nothing slips. Some delay is likely. Below is the list of PSPs selected for the pilot.

4. What about me ?

We have three credit cards at home, at about €9 each per month. We’ll keep one for trips abroad, for the insurance and the best international transaction fees.

The other two will go. The digital euro will cover the same things at no cost, and every merchant in the eurozone will have to accept it.

The €9 includes my bank’s fees. What Visa and Mastercard take is about 0.1% of each transaction. The impact could be to 150 million households in the same situation.

As an independent analyst, I may end up with no business card at all. If the digital euro covers everyday payments and domestic B2B transactions, a corporate card isn’t needed anymore, out of business travel abroad. One more brake on the flywheel.

5. The Guillou Case

On August 20, 2025, the U.S. Treasury put Nicolas Guillou, a French judge at the International Criminal Court, on its sanctions list, next to drug traffickers and Al-Qaeda members, for issuing arrest warrants against Prime Minister Netanyahu and former Defense Minister Gallant.

Visa and Mastercard had to comply, and so did PayPal. In France, paying in euros, a French judge couldn’t buy groceries or withdraw cash with his card anymore.

His own words, spoken at the Union of Judges’ convention in October 2025:

I realized that Europe no longer has any sovereignty when it comes to payment systems, since we rely almost exclusively on American payment cards: Visa, Mastercard, and American Express. Even if your bank doesn’t close your account, you no longer have any means of payment.

Nicolas Guillou (French judge at the International Criminal Court)

As reported by Le Monde on November 19, 2025 (in French): the U.S. executive can cut off any European citizen from the banking system and digital life of their own country.

Everyone at the ECB and in Brussels saw it. For the EU, this is a matter of survival, and they know it.

6. Risks perceived by intermediaries

Visa’s 10-K for fiscal 2025 does mention this risk:

The European Central Bank has launched a multi-year project to study the introduction of a digital euro, which would serve as an alternative to foreign digital currencies and payment service providers. If their rollout proves successful, these national payment platforms and digital currencies could have a significant impact on Visa’s domestic and cross-border payments, particularly in terms of potential disintermediation.

Mastercard’s 10-K barely mentions the topic, and PayPal’s report doesn’t mention it at all.

III. The Impact on Visa, Mastercard, and PayPal

1. Expansion Scenario

At first, the digital euro will be open only to eurozone residents, and to non-residents with an account at a eurozone payment provider. Non-residents will still pay with their Visa card. But it could also expand to neighbouring countries, over the coming years.

Central banks could also connect their systems directly and skip the intermediaries. Then, moving money would become a utility run by public institutions, a bit like water or electricity.

With the emergence of nationalism and national pride, paying with the national network could become a way to show your ideological views.

3. Financial impacts

61% of Visa’s revenue comes from outside the United States, and the Eurozone is a big part of it. Since Visa is asset-light, any downturn in revenue would severely affect the operational leverage, and the flywheel would start turning backwards, as shown in the picture below.

Data processing and service revenue follow the decline in revenue proportionally. International transactions and value-added services should hold better.

A drop in revenue hits ROIC immediately, and the premium on the multiple (because of the stability of the business) would shrink. For me, today’s price does not justify the risk premium of Visa or Mastercard.

PayPal has the same exposure as Visa and Mastercard in Europe, but nothing like their international transactions as a last line of defense. So it’s even worse for them.

Conclusion

Given the successful examples of Pix in Brazil and UPI in India, I see no reason why the model wouldn’t be a success in the Eurozone. And since the Guillou affair, this approach has become a necessity, there’s no turning back.

Meanwhile, fintech companies continue to emerge, and the account-to-account payments are growing. I find it hard to look ahead to the future of payment methods with confidence. I think there will be a name for what is happening to Visa, Mastercard and PayPal in Europe.

We’ll call it the “Finpocalypse.”

Subscribe now

Disclaimer

• “The streaming market is saturated”

• “Netflix has gotten too expensive”

• “There’s no more growth”

These points may be true, to a certain extent. But the market consensus loves to reiterate its pessimism when the price is at its lowest, while every new rally is met with optimism when the price hits record highs.

Netflix had been riding a wave of excessive optimism, which was reversed by a drop of more than 40% from its highs. But like all great companies, it never ceases to surprise us. Let’s see what the future holds for Netflix.

Like AppLovin, Netflix belongs to my favorite type of investment: a category definer, temporarily beaten down by the market on a below-expectations print.

A category definer is a company that created a new category and occupies the leadership seat in it, crushing the competition in its segment through its products and services.

This is a favorite strategy of Bill Ackman at Pershing Square, where he made a purchase in Q2 2026 for a position that now accounts for nearly 5% of his portfolio, with an entry price of $71.

I had already noticed Netflix when its stock price fell in Q2, but it was hard to know how far the price would drop. It seems to have stabilized since then.

The structure of my deep dive is standard. Learn the process in the dedicated articles in the learning section.

The whole article is long because it covers a lot of ground, and short because I try to give you a synthesis of the points. So I start with a “short version” for those of you in a hurry.

Subscribe now

The Short Version

Netflix is a category definer down 40% after several quarters that pointed to a slowdown in growth. The streaming giant serves 330 million households, yet still holds only 7% of a $670 billion market. The company has a competitive advantage due to its size, which gives it a cost advantage in production and distribution, and due to its proprietary catalog. But the following question is, in my opinion, the key reason for choosing Netflix as an investment:

If you had to keep just one streaming service, which one would you choose?

Everyone I spoke with chose Netflix. The double-digit annual revenue growth over the past several years has resulted in earnings growth of more than 30% per year. The steady increase in revenue continues to improve operating leverage and, consequently, profitability potential. The company has growth drivers in advertising, video games, and live events, particularly in sports.

I’m convinced that Netflix could quickly double in value within a year, or grow by 30% annually for about three years. The Q3 results on October 20 will indicate whether I should buy the stock; I’ll keep an eye on whether the price stabilizes or if market concerns persist.

1. What Netflix does

History

Netflix was founded in 1997 by Reed Hastings and Marc Randolph in Scotts Valley, near Santa Cruz, California. It began as a mail-order DVD rental service. The company went public in 2002. Reed Hastings took sole control in 2004 and launched its streaming service in 2007, which would go on to bring the company spectacular success. It began producing original content with *House of Cards* in 2013, marking a turning point in its history.

The first season of *House of Cards*, a Netflix original series, marks a turning point in its story

The 2010s were a period of growth and international expansion. The company experienced its first-ever net subscriber loss in Q1 2022, which, combined with the ongoing bear market, led to a nearly 75% decline in its stock price. Starting in 2023, the company launched its ad-supported service.

What it sells

Netflix primarily sells subscriptions directly to viewers. It’s also possible to get Netflix through partners or service providers. Subscriptions are monthly and can be canceled at any time, with prices ranging from $1 to $37 per month depending on the country and the plan (ad-supported or ad-free, number of screens, video quality, etc.). The company reserves the right to change its prices and plans “based on perceived value rather than on a fixed schedule.” A small portion of its revenue comes from advertising.

Companies typically report on multiple product lines. Netflix, however, reports under a single operating segment: “Streaming Revenues.” The 2026 guidance is $51 billion, of which $3 billion comes from advertising, with the remainder coming from subscription revenue. Netflix is best known for its catalog of movies and TV shows, but other segments are expanding, including video games and podcasts. As management noted during the Q2 2026 earnings call:

Another big positive sign is that since last October, so 8 months ago, when we really sort of scaled up this cloud initiative, monthly active players for cloud games have increased 11x and adoption is significantly ahead of that curve that we had for mobile games with even higher retention value. So we’re definitely excited about that and focused on scaling up cloud games. We’re also seeing positive signals with kids games. So Netflix Playground, which is our app for kids games, no ads, no in-app purchases, curated set of games, very safe space. We’ve seen 3x growth in daily players since that launch. That’s driven more engagement in kids mobile games, which is up 600% year-over-year. So that’s super exciting to see as well.

Gregory K. Peters, Co-CEO, President & Director

Management also cites podcast distribution as showing encouraging signs. It remains to be seen whether this segment will develop into a growth driver in the future.

Netflix is also developing an offer on live events, as they “do a lot of lifting for us for acquisition: [...] 6 out of top 10 new member sign-up days over the past 5 years have come from live events”.

The company spends between $18 and $20 billion a year on content.

Customers

Typically, understanding a company’s customers (especially in the B2B sector) is particularly complex. In this case, Netflix has the undeniable advantage of catering to the general public and applying Peter Lynch’s key principle: “Invest in what you know.”

Its 330 million subscriber households, which initially were more prevalent among those under 45, now span all ages and family types, ranging from those who are more price-sensitive and opt for the ad-supported tier to those seeking a premium experience who are willing to pay more.

Here’s a little anecdote: I know quite a few families where the grown-up kids come back to visit their parents to get their Netflix account back, since access gets blocked after about two weeks without a Wi-Fi connection. It’s a sign of family bonds, and it’s touching.

Below is a geographic breakdown of revenue. This overview shows a disproportionate concentration in countries where subscriptions are most expensive.

One type of minority client is advertisers in the advertising sector.

According to Netflix, its Total Addressable Market (TAM) is estimated at $670 billion, and the company captures 7% of it, which represents 5% of the global TV audience share. This second metric measures time spent, while the first focuses on revenue.

Netflix still has enormous room for growth in its core segment, as the company states they serve 330 million households out of 800 million.

The market itself is growing, according to the company, from $600B+ in 2023 to $670B in 2026, but Netflix does not publish market share figures before 2024: the first historical figure is 6%, rising to 7% in 2026.

Regarding the video game segment, the company indicates a TAM of $140-150B excluding China/Russia, with revenue still nascent for now.

Suppliers

Netflix runs the vast majority of its computing on Amazon Web Services. Quite funny when you know it’s its closest competitor.

The Hollywood guilds govern collective bargaining agreements for writers, actors, performers, and directors. Contract renegotiations are potential moments of risk.

The company also purchases content rights, and has studios and rights holders as suppliers for its content licenses.

Netflix operates its own delivery network, and must interconnect with internet service providers and other telecom operators. This regularly leads to disputes and court summonses over net neutrality and fee calculations.

Competitors

The first area to consider is direct competition. The graph below shows major players’ market shares:

2025/2026 was in this sector a true stock market soap opera, full of rumors and dramatic twists: The industry was in negotiations to acquire Warner Bros. Discovery’s streaming and studio businesses, with Netflix having made an offer, then withdrawing in the face of a higher bid from Paramount-Skydance.

Every time a user wants to be entertained, they will choose their medium, and the real battle is for the user’s attention. That is why, in its talks, management prefers to talk about another metric called moment of truth, rather than viewing time. It describes the instant when the consumer chooses a brand or product over an alternative.

If the moment of truth is the battle for attention, the war is fought in the mind.

And in this arena, the competition is fierce: YouTube and its billions of videos, Meta and its endless feed, television, video games, and every form of entertainment under the sun. For example, YouTube accounts for 13.4% of U.S. TV viewing time, compared to 7.8% for Netflix.

A French television executive (TF1) made an infamous remark when he said he was selling “available brain time.” Behind this cynicism lies an incentive for an ad-supported network to maximize captive brain time at all costs (through addictive, superficial, and repetitive content) because that’s what sells to advertisers. Netflix, being subscription-based, is incentivized to maximize perceived value (and as metrics, quality and retention) so that the subscriber keeps paying every month.

Competitive advantage

From a consumer’s perspective, I invite you to ask the people around you: If you had to keep just one streaming service, which one would you choose?

I may be biased, but as far as I’m concerned, the unanimous answer is Netflix. Even for those with children, where you might expect competition from Disney+.

It is my deep conviction and it resonates whenever I hear that Netflix has won the streaming war, beyond just market share.

On subscription pricing:

• The US/Mexico/Spain price increases in H1 2026 were absorbed as expected

• The end of password sharing had already converted free users into paying ones.

• We can infer that a price increase in Europe would have the same effects.

A major point: users can cancel their subscriptions at any time. The switching costs are mainly psychological, but subscribers could potentially leave overnight. The assumption made above is that a multi-streaming user would cancel their subscription to another service in order to keep only one: Netflix.

The company also has an advantage of better use of catalog and greater predictability of success on pre-project (i.e. when you choose which shows you’ll produce) that comes primarily from behavioral data collected over more than a decade and engagement metrics that Netflix refuses to disclose:

Netflix produces most of its shows. It carries some advantages like the IP Ownership: In 2019–2021, WarnerMedia/NBCUniversal removed Friends and The Office from its catalog. When you have the ownership, you cannot be canceled. It also gives the ability to monetize the content, like with games, merchandise or spin-offs.

It also eliminates the middlemen, for a better cost basis of the content, and when Netflix produces real hits, it can distribute them across 330 million subscribers in ~190 countries. It is something a traditional studio cannot replicate.

Netflix has the largest engaged audience in the world: it gives the company a decisive cost advantage from economies of scale. In addition to this advantage, there is a network effect that attracts creators and studios, because the return on investment for content distributed at this scale is better than elsewhere. That improves Netflix’s bargaining power in future deals. And because it has such a scale, it was able to develop a proprietary delivery network (OpenConnect), with a cost advantage over a third party.

Here is the summary slide. Note that this image should be read against the alternatives. Netflix will face less competitive intensity than Hulu, because this is David against Goliath, so Netflix has an advantage over its competitors. It is not Apple, nor Coca-Cola, in terms of competitive advantage, but it is still a very good moat.

2. Financials

Key financial metrics

Watch out when interpreting the figures; these are for a half-year period in 2026. You must double the amounts and adjust for one-time items to compare the years.

Commentary

Revenue

The increase in revenue is generally linear. Note that the deferred revenue mechanism (subscription fees billed in advance but only recognized in earnings for the delivered part) can affect sales. The increase is due either to the growing number of subscribers or to pricing.

It’s worth noting that Netflix has immense growth potential in its core market. A shift from 7% to 14% of the company’s total addressable market (TAM), based solely on volume, would double annual revenue and generate between 100% and 200% profit growth, depending on assumptions regarding operating leverage.

Operating margin

Every increase in revenue leads to an increase in operating leverage, which in turn leads to higher margins.

Without going into too much detail, the company will record its content on its balance sheet (capitalize it) at the outset, and then amortize it over time. More than 90% of the cost of a title (license or production) is amortized within four years of its initial release, with movies amortized more quickly than TV series. Netflix does not amortize titles individually, but rather on an aggregate basis.

Production tax credits reduce the cost basis of the content, thereby reducing future depreciation over the title’s useful life.

The value of the catalog and intellectual property does not appear on the balance sheet; it is a cost that is amortized. This intellectual property can therefore be considered an unrecognized asset.

Net income and free cash flow

Net income for H1 26 must be restated because it includes $2.8 billion in termination fees related to the WBD contract negotiations, which are reported under “Interest and other income.” This accounts for almost the entire difference from FCF.

Differences in the amounts capitalized and amortized can account for discrepancies between net income and free cash flow:

• If Netflix capitalizes more than it amortizes, there is no impact on net income, but free cash flow decreases (because operating cash flow decreases).

• If Netflix amortizes more than it capitalizes, this reduces net income but has no effect on free cash flow, since the cash was already spent at the time of capitalization.

It’s kind of like working capital.

The annual growth in net income and free cash flow over the past five years has been impressive, especially given the company’s market capitalization, at approximately +31.8% per year and +37.4% per year, respectively.

Balance sheet

The balance sheet is dominated by content, with net content assets accounting for 59% of it. Netflix has $9 billion in cash and net debt (including cash on hand) of $5.24 billion. It would take the company nearly five months of operating income to pay off its debt. No problem there.

Netflix also has off-balance-sheet commitments related to contractual obligations: these are productions that Netflix has agreed to produce but that are not yet available in its catalog. The total amount is nearly $18 billion, after all.

3. Management

A Two-Headed Head

Historically, Reed Hastings was at the helm, but since 2020, the company has had two CEOs. Normally, this violates the principle of unity of command so dear to Fayol, which is why this model remains very rare.

Here, responsibilities are divided by area:

• Ted Sarandos: content, marketing, legal, communications, public relations. He has held this position since July 2020 and has been with Netflix since 2000.

• Greg Peters: Product and Technology, Advertising, Human Resources, Finance, Video Games. He has held this position since 2023, taking over from Reed Hastings. He joined Netflix in 2008.

The principle is specific to Netflix: for each significant decision, a single person is designated as responsible (the “captain”), chosen based on who has the most expertise and context on the subject. It is up to that person to consult widely before deciding, but the final decision rests with them, with no need for consensus. They call it “informed captaincy”.

When there is a disagreement between the two, Peters summed it up this way: “When there’s a space where we disagree, but then we ultimately defer to each other’s area of expertise, I think we get to better decision-making.”

A safety net exists as well: the two work with an executive coach (whose identity remains confidential at Netflix) to manage their relationship and coordinate with the board of directors, described as a “light touch” intervention.

Ted Sarandos and Greg Peters

Regarding their alignment (skin in the game),

• Ted Sarandos earned a total of $53.9 million in 2025, including $3 million in salary, $7 million in bonuses, stock awards worth $41.4 million (77% of his total compensation), and other compensation totaling $2.4 million. He owns 178,954 shares, worth approximately $13 million.

• Greg Peters received a total of $53.2 million in 2025, including $3 million in salary, $7 million in bonuses, and the vast majority of the remainder in stock. He owns 120,931 shares, worth approximately $8 million.

Surprisingly, Netflix lets each employee choose the cash-to-stock-option split of their compensation, a rather unusual form of shareholder alignment.

We’d need to know what portion of their wealth is in Netflix stock. I don’t have that information at this time. We note that, since their stock holdings are worth roughly one year of cash compensation, this does not point toward “skin in the game” on this point. Regarding stock-based compensation, it vests over three to four years.

We can also assume that the CEOs’ strong profile is linked to a career spent almost exclusively at Netflix. A mercenary would have already switched companies several times. Furthermore, the separation of powers at the top is a positive sign that helps avoid leaders obsessed with personal glory. Likewise, the sensible decision to refuse to pay an excessive price for WBD is also a step in the right direction.

Capital allocation

First, a company will seek to reinvest its profits in organic growth or, alternatively, through acquisitions; otherwise, it will return capital through dividends or share buybacks.

Here, Netflix generates enough profits to exceed its capacity for organic reinvestment and finances its share buybacks through positive cash flows.

As mentioned, the decision not to outbid Paramount-Skydance’s offer is a positive development, described as “nice to have at the right price, not a must-have at any cost.”

One thing I’m trying to avoid over time is a downward revision of guidance, which hasn’t occurred in recent years.

In previous fiscal years (2023–2026), the company has delivered on its guidance and addressed challenging issues during earnings calls. The company’s policy does not encourage sandbagging: “The guidance we provide is our actual internal forecast at the time we report, and we strive for accuracy.”

The only issue that needs to be addressed is that the company reduced the granularity of its non-financial data (subscribers, engagement) at the very moment when that data was showing signs of weakness.

Subscribe now

4. The two theses

I believe the best way to decide whether an investment is sound is to pit the bullish and bearish theses against each other. That way, I know what could happen, and knowing your position well helps you endure downturns and protects you psychologically. Considering only the bear thesis is alarmism; considering only the bull thesis is wishful thinking.

Why did the stock fall?

Netflix has lost nearly 40% from its highs due to several factors:

• The decline started with Q2 ‘26 results that were in line with Netflix’s guidance but below analyst consensus, along with the simultaneous announcement of a cut to the “What We Watched” engagement report, moving from semi-annual to annual publication starting in 2027. It’s the second disclosure reduction in ~15 months, as Netflix hasn’t published subscriber details since 2024.

• Then, Q4 ‘25 to Q2 ‘26 marks the third consecutive quarter of slowing growth. Even though overall growth remains positive on an annual basis, the market is concerned about these consecutive declines.

• Additionally, on October 27, 2025, Netflix reported a one-time Brazilian tax charge of $619 million, which lowered analysts’ estimates and triggered a 12% drop in the stock price.

• Furthermore, the decline continued in connection with negotiations over the acquisition of Warner Bros. in December 2025, when the stock lost 12.9%. Shareholders and the market were wary of the deal’s terms. In February ‘26, a rebound occurred following the halt of discussions, which was a positive signal for capital allocation.

Bear thesis

Slowdown in Growth

Growth is slowing quarter after quarter, which ultimately hampers margin expansion and net income growth.

It is possible that increased competition from other streaming services, consolidation among existing players at Netflix’s expense, and indirect competition will ultimately prove too much for Netflix and for the entire industry.

In that case, even if the market expands, no player will turn a profit.

The U.S. and Canada market appears close to saturation, with growth of only +10% in Q2 ’26, compared to +21% in LATAM and +16% in APAC.

Furthermore, the loss of data resulting from the reduction in the “What We Watched” engagement metrics is detrimental and could be used to mask Netflix’s poor results.

We should also consider competition from all other forms of entertainment that could take over. Customers are always free to leave.

Disruption risk

Generative AI is a potential source of disruption: Netflix could benefit from it in the future or be harmed by it. Independent studios or other tech companies could leverage these new technologies to create better content, faster, and at a lower cost.

Legal risk

Given Netflix’s size, acquiring competitors would likely face regulatory opposition. It would have to settle for acquiring fledgling studios or platforms.

This last point is more hypothetical, but worth mentioning:

There is always a regulatory risk, particularly in the European Union (a region known for such regulations…) with quotas for high-cost European content or a requirement to disclose the algorithm data on which part of its competitive advantage is based.

Bull thesis

Netflix is the No. 1 streaming service

I’ll go back to the beginning of the section on competitive advantage, which, in my opinion, is the jewel in the crown:

I invite you to ask the people around you: If you had to keep just one streaming service, which one would you choose?

I may be biased, but for me, the unanimous answer is Netflix—even among those with children, where you might expect competition from Disney+.

This is my firm conviction on this matter, which resonates with the times I hear that Netflix has won the streaming war beyond just market share.

And I don’t know about you, but at the same price point as YouTube Premium and Netflix, I find Netflix to be of much higher quality. Netflix is the king of entertainment accessible from home, entertainment that’s shared much more readily than a video game.

Growth Drivers

The company still has significant opportunities to increase its revenue in its market by taking market share from television, where it currently holds only a 7% market share, in a global market that continues to grow at a single-digit rate.

In addition, the following segments can also contribute to continued growth:

• Advertising, projected to generate $3 billion in revenue

• Video games, which are growing rapidly, particularly among children, with nearly a 600% increase year-over-year, from a small base though.

• Live sports, currently held by national networks. One can imagine that Netflix’s clout allows it to secure these popular events. That said, this is no different from other streaming players. Rather, it’s a point in favor of these players over traditional TV.

While Netflix has grown in terms of subscriber numbers, it can also grow by expanding the depth of its offerings with more channels per user.

Capital Allocation Discipline

Netflix has demonstrated excellent discipline, particularly through its refusal to acquire Warner Bros. Discovery: most acquisitions result in a price that exceeds earnings, especially during periods of market optimism or euphoria. It’s reasonable to assume that this discipline can be put to good use in segments that serve as growth drivers.

Furthermore, Netflix possesses the essential assets (the brands of its internally produced hit series and its distribution network) and can acquire AI studios to address the growing threat of disruption without raising red flags with regulators, as long as these studios remain small. Generative AI is already being used in approximately 300 productions by 2026.

Amid the arguments for the bear case, there is a fine line that Netflix’s management can reasonably tread. They certainly have the cards to pull this off. It remains to be seen whether they’ll do so intelligently.

Before the bear case plays out, the bull case metrics would begin to stall:

The occurrence of a single trigger is not enough to invalidate the bull case. If the case is called into question, I will publish an update to assess the new situation.

The decline in ROIC between 2021 and 2022, combined with a slight decrease in the number of users, triggered a drop of nearly 75% between 2021 and 2022. The recovery was spectacular through 2025.

Blind spots

Some arguments aren’t getting clear enough. To give you an overview, I’d like to share them with you, as they’re worth mentioning:

A central thesis in my investment strategy is the fragmentation of the world into geopolitical blocs, which is a megatrend. In my view, Netflix fits well into a multipolar world, with its international studios, and relies less on American soft power than other players. On this point, I think it’s better than Amazon. But it’s hard to generalize beyond my personal opinion.

The argument about the advertising/live sports/AI narrative underpinning the bull case isn’t unique to Netflix, but rather applies to its sector; it’s difficult to distinguish between the players at this point.

I would also have liked to tell you that Netflix is more profitable due to the scale of its distribution. Backing up this claim with facts is more complicated than I thought. The best approximation comes from ROCE. As a reminder, this ratio is calculated as Net Operating Profit After Tax (NOPAT) divided by Capital Employed (equity + financial debt).

It should be noted that Prime Video is a division of Amazon, and that the ROCE figure is Amazon’s, which includes the Cloud segment (high margins) and the Retail segment (low margins), and is likely underestimated. Apple has a very high ROCE due to a very low capital base. Apple TV is not, a priori, intended to be a market leader.

Any observations regarding Alphabet or Meta would apply equally here.

Finally, information on Netflix’s content distribution is not publicly available. Management cites data as a competitive advantage, which leaves room for withholding information about any challenges encountered.

5. Price scorecard

The numbers only make sense if you’ve read the two theses section. As Aswath Damodaran puts it, “numbers without stories to back them up are exercises in financial modeling.”

The scorecard below requires an important assumption, being the continuity of a normal economic environment, i.e. without recession, major credit event, market crash.

I will someday write an article about how to deal with these conditions.

Any of these could push the stock below the bear case shown here, temporarily, independent of the company’s underlying performance. All figures in the scorecard are pre-tax and pre-fees. Run the numbers for your own situation before drawing conclusions.

The price scorecard map

The assumptions are the following:

• For the revenue, it goes from 6% per year in the bear case, to 10% per year in the central case, and 15% per year in the bull case. Quite conservative for the bull case.

• The margins may expand thanks to the operational leverage in the central and bull cases, but as conservative assumptions, they stay flat at 24.3% (the one from the FY2025).

• Share buybacks of 4% per year in the central and bull scenarios.

The selected P/E ratios range from the 2022 low to the July 2025 high.

My conviction sits at the $172.5 scenario or higher (bull case x PE 35). Also consider that the gap could close next year ; Netflix could double. In that case, I would probably sell and find another candidate.

The re-rating will mainly depend on the catalysts covered in the next section.

6. Momentum

Overall, crossing fundamentals (covered above) with momentum is an excellent way to improve your returns. You’re never right on your own, especially since a significant part of returns comes from the multiple, especially if your horizon is below 5 years.

Catalysts

Next quarter results: They are scheduled for October 20. We’ll see whether growth rebounds, stagnates, or continues to slow. Will the operating margin rebound in the second half of the year, as the seasonal pattern would suggest, or will it continue to decline on a year-over-year basis? Advertising revenue compared to the ~$3 billion 2026 target is also something to watch this quarter.

Completion of Paramount-Skydance's Acquisition of WBD: On September 21, 2026, Paramount settled the antitrust lawsuit filed by state attorneys general, the last major obstacle to the deal; David Ellison (CEO of Paramount-Skydance) says he expects the deal to close “in about two weeks”, so around early October 2026. The stock already seems priced in, but we’ll have to see if there are any potential surprises.

AI disruption: An Adobe-style scenario is not out of the question, with AI-generated competing movies or TV shows hitting the market, given the market’s perceived fears of disruption. This would push the PE ratio toward 10 or 15.

A rerating from superinvestors buying the stock is still possible, but as Bill Ackman already owns the stock, a part is already priced in. Berkshire Hathaway could buy and make the stock go up, but I think there are few that could do so.

When to enter the stock?

The market hates uncertainty and has already started a significant decline in the share price. So you need to track the price to see where it bottoms. I see three possible strategies:

Buy the dip:

• I would wait for analysts to stop cutting their outlooks.

• The likely target zone is a PE TTM of 20-25. It’s hard to give a PE ratio as a support level, since the company has broken through its last support (200-day moving average) on the way down.

• This works if Netflix surprises to the upside next quarter, and fails if Netflix disappoints or worries the market for a second quarter.

Capitulation:

• It’s possible we head toward a bearish scenario if the company reports well below expectations. In that case, you have to wait for capitulation, which shows up through several signals, such as above-average volume, or RSI <20.

• So far, Netflix hasn’t shown signs of capitulation.

DCA: the way to reduce risk (the non-choice) is to run a DCA every month, if that’s your approach. I personally prefer to enter in one or two tranches.

As for me, I plan to open a position once the price stabilizes, and I’ll let my subscribers know when I do. I’m prepared to add to the position on further weakness or capitulation. There’s no rush: the thesis will probably take 2-3 years to play out.

Final Thoughts

Netflix has won the streaming war, but the battle for user attention is a war of the mind where every form of entertainment fights for its place.

I’m fairly bullish on this stock because I believe the forces at work in the market structure and its competitive advantage are working in its favor, and that the average user prefers Netflix. But only time will tell if management can implement the right strategies to sustain its impressive growth. Here, memory is the key to our hope: Netflix has impressed us in the past and may well continue to do so for longer than we think.

Subscribe now

Disclaimer: This content is for educational purposes only and does not constitute investment advice. See our full disclaimer.

The best investment category

AppLovin belongs to my favorite type of investment: a category definer, temporarily beaten down by the market on a below-expectations print.

A category definer is a company that created a new category and occupies the leadership seat in it, crushing the competition in its segment through its products and services.

I’m sure you already have a few names in mind, and here are some examples.

Investing in such companies offers one of the best risk-reward ratios you can find in the market. Because if a company crushes the competition, is loved by its customers, does more volume, and keeps improving relentlessly, who’s going to threaten it?

Generally, these companies are recognized by the market as very good investments and trade at a premium. The most financially interesting moments to invest in them come during a passing scare.

The tricky part is telling apart a company in decline from a giant going through a moment of doubt.

And sure enough, on August 5th 2026, AppLovin disappointed. Its Q2 revenue landed slightly below guidance, and the stock fell 19.66%. The stock has kept sliding since its all-time high of $745.61 in November 2025. Yet performance remains remarkable, with 53% annual revenue growth and net income up 78% on a trailing-twelve-month basis.

AppLovin is a great example of the virtue of patience, where waiting a year before investing would have paid off. As for me, I hesitated to enter the position back in February near its lows in case it kept falling, but the market decided to hand back part of its loss.

A few words on method and plan

I’m a former project manager with 7 years of experience (PMP certified), and I formalize my fundamental analyses like a project trajectory, from point A (starting point) to point B (destination).

Point A, in our case, is what the company does, together with its environment (competitors, suppliers) and its competitive advantage, then the financial and management side. This is a static study, and if you already know the company, you can skim through it quickly.

Point B is a forward-looking study. I don’t claim to know the future, nor what the company is worth. So instead I look for the different scenarios that could play out for the company, the way you’d run a qualitative and quantitative risk analysis on a project.

This takes the shape of weighing opposing theses: why the stock fell, the arguments for the bear thesis, and the arguments for the bull thesis. I try to make both sides as objective as possible using quantifiable criteria.

I then lay out a map of possible prices at a given target date, without a discount rate, crossing fundamental trajectories with valuation levels, stating my starting assumptions each time. Knowing which valuation multiple to apply at a given date comes down to knowing what the risks, opportunities, and therefore the narrative of the company will be. For convenience and by convention, this section is called “Valuation” even though that’s not really the right word for it.

The whole article is long because it covers a lot of ground, and short because I try to give you a synthesis of the points. So I start with a “short version” for those of you in a hurry, here it is, without further ado.

The Short Version

After falling more than 50% from its highs, notably due to a quarterly print below expectations, one can reasonably think this is an air pocket, and that growth, spectacular so far (Q3 guidance still sits at +46-48% YoY), will trend toward 30% a year according to management. The market is punishing this result all the more because AppLovin had never posted below its own expectations since its 2021 IPO.

This looks like a classic buy-the-dip setup on a quality company momentarily tormented by the market. The results due November 11th should confirm this for us. The company could be disrupted, but that seems fairly unlikely in the near term and isn’t showing up in the numbers.

Subscribe now

1. What AppLovin does

Business model

When a company wants to find customers (gamers who’ll buy inside its game, buyers on its e-commerce site), it goes to AppLovin and states its goal (“I want X users who spend at least Y”), and AppLovin’s AI places its ads wherever that’s most profitable.

This model doesn’t require an intermediary ad agency: the company itself is in the driver’s seat, and AppLovin supplies the tool and the algorithm. AppLovin gets paid by taking a commission on what the company spends.

This business model is built on four pillars:

AppLovin Ads — the core of the business. I would have liked to give you a number, but AppLovin sticks to “substantially all of our revenue” in its 10-Q and 10-K filings. The company (the advertiser) sets its campaign goal, and the AI optimizes the placement. AppLovin pays the publisher of the app where the ad was shown and keeps the rest as margin. Pricing is dynamic (as opposed to fixed, like “10 cents a click”): the algorithm adjusts in real time how much to charge based on the advertiser’s goal. Nearly all of revenue comes from mobile apps, and mostly from mobile games.

Since Q2 2026, AppLovin Ads has been available in self-service. Before that, you had to go through an ad rep.

MAX — used by app publishers to auction off their ad inventory. AppLovin takes its commission on the winning bid before passing the rest to the publisher. As with the previous product, nearly all of revenue comes from mobile apps, and mostly from mobile games.

Adjust — a measurement tool for advertisers (how many installs, fraud detection). The only product billed as an annual subscription, unlike the others which run on commission.

Wurl — distribution of video programming across streaming/connected-TV platforms, with ad sales on top. Billed on usage and/or CPM (cost per mille: the price the advertiser pays for 1,000 impressions of its ad, regardless of whether anyone clicks or buys afterward).

Suppliers

AppLovin is hosted on Google Cloud and uses its associated services. Beyond that, AppLovin is structurally dependent on the Apple App Store and Google Play for distributing its content, collecting payments, and targeting rules, and both of these players hold significant market power.

Clients

AppLovin has three customer groups, ranging from the largest companies to small independent operators:

• Advertisers who use the platform to find users.

• Advertising networks who use AppLovin’s mediation solution to buy inventory.

• The community of mobile app publishers who use the mediation tool to monetize their ad inventory.

Having a broad customer base is an advantage from a risk standpoint. Worth noting: the advertising sector is cyclical, and all customers tend to cut spending at the same time during a downturn.

Competitors

The 10-K explicitly names Meta, Google, Amazon, and Unity Software, while making an important point: some of these “competitors” are also partners and clients of AppLovin (“several of which are also our partners and clients”), the line between competitor and customer isn’t clean. The company also states: “We operate in a fragmented advertising ecosystem composed of divisions of large, well-established companies as well as privately-held companies.”

Here’s an illustration of the three preceding sections:

Competitive advantage

The main competitive advantage lies in the quality of the company’s products and the quality of its management.

The presence of network effects and switching costs helps reinforce the company’s lead and insulate it from competition.

On the network effect, the 10-K describes it this way:

“As more advertisers use our advertising solutions to market and monetize their content, we gain access to more data regarding users and user engagement, further strengthening our scaled distribution. As our distribution grows, we gain better insights for Axon AI, which then further enhances the efficiency and effectiveness of the Axon Ads Manager.”

This is primarily a data-scale barrier. Anyone with the same level of data would erase the competitive advantage, which isn’t currently the case.

I lack information on the company’s switching costs, and therefore on how easy it is to change advertising provider. It doesn’t strike me as especially complicated, especially since it seems possible to bring on a second provider, run comparisons based on a pilot, then progressively migrate ad campaigns over. We’re not in a Microsoft-style situation where entire teams’ workflows are so embedded that they can’t switch overnight. All the more so since the company’s contracts include clauses allowing cancellation at any time.

2. Financials

Key financial metrics

Commentary

The numbers are staggering. Decelerating, sure, but staggering nonetheless, especially for a TTM PE of 26. No sign of inflation from financial shenanigans after reviewing AppLovin.

Revenue

Growth has decelerated continuously and steadily over the last five periods, going from 75% (FY ’24) to 55% (H1 ’26 YoY). AppLovin states that the composition of growth has changed in nature, and breaks its own revenue down into two factors, install volume and net revenue per install (MD&A):

• FY2024: installs +50%, net revenue per install +22% → growth driven by volume.

• FY2025: installs +3%, net revenue per install +72% → shift toward price/monetization.

• H1 2026: installs -10%, net revenue per install +75% → volume is declining, and all growth now comes from monetization per install.

• Q2 2026 alone: installs -2%, net revenue per install +58% → same read, slightly softer.

Net revenue per install is the net amount of money earned over a period divided by the total number of installs generated. The number of installs tends to mechanically shrink (you only install once), and on that shrinking base, net revenue per install climbs (which can happen repeatedly, by nature).

Worth remembering, still: continuing to grow like this at these revenue levels isn’t sustainable over time. A cumulative growth rate of 50% a year gives you roughly 57x the initial amount after 10 years. So what to make of a doubling, or 75% a year?

Operating margin

With revenue growth outpacing cost growth, AppLovin’s business is scalable and shows significant operating leverage. The rise in GAAP operating margin from 42% (FY ’23) to 78% (Q2 ’26) is spectacular. Unfortunately, since the ceiling is 100%, there isn’t much more improvement to hope for on this front going forward. Holding at these levels remains impressive in itself.

Net income and free cash flow

AppLovin’s net margin is boosted by a low tax rate, 13.1% in 2025, well below the 21% federal rate. AppLovin benefits from an FDII (Foreign-Derived Intangible Income) tax break, a US tax provision created in 2017 (the Tax Cuts and Jobs Act).

I like comparing net income and free cash flow to spot variations that might reveal hidden issues. Here, they converge toward the same figure, around 65%, which is excellent. For net income, growth outpaces revenue growth in every period, driven by a scissor effect (opex shrinking as a percentage while revenue doubles).

For FCF, I usually strip out SBC, which is a non-cash charge. Growth in this metric looks larger since it starts from a smaller base.

Balance sheet

The company is financially very solid, with a debt-minus-cash to equity ratio of 0.15x. AppLovin is holding onto its long-term debt; the ratio is shrinking through the accumulation of cash and equity, which are growing faster than everything else.

3. Management

Adam Foroughi — CEO (co-founder, since 2011)

Foroughi holds only 9% of the economic capital but 61.6% of the voting power (Class B shares carry 20 votes each), 66.9% including the voting agreement tying him to Herald Chen. This makes AppLovin legally a “controlled company,” although it doesn’t make use of the governance exemptions that status allows.

According to Forbes and Bloomberg estimates, this would represent 66% to 85% of his net worth. In terms of incentives, his salary has been capped at $400k since the IPO, with no increase since.

The fact that the CEO is also the founder is a very good sign, one that typically generates outperformance.

More broadly across the leadership team, the variable portion of compensation plans is paid in equity, not cash, a notable point of alignment.

Capital allocation

No dividend is currently paid. There are share buybacks, but they primarily serve to offset dilution from SBC. Beyond that, the amount repurchased remains marginal: 0.5% over 2025, and 0.9% over H1 2026. The declining share price will mechanically push this share higher.

4. The two theses

I believe the best way to decide whether an investment is sound is to pit the bullish and bearish theses against each other. That way, I know what could happen, and knowing your position well helps you endure downturns and protects you psychologically. Considering only the bear thesis is alarmism; considering only the bull thesis is wishful thinking. This section includes both the arguments and the catalysts for each version of the story.

Why did the stock fall?

Starting in February 2025, a series of short-seller reports (Fuzzy Panda Research, Culper Research, then Muddy Waters Research in March 2025) accused AppLovin’s AXON ad system of circumventing Apple’s App Tracking Transparency framework. These allegations fed an SEC investigation that weighed on the stock in early 2026, down to a low reached around February 12th 2026, nearly 50% below the late-2025 peak.

Next, doubts about the durability of growth. As the market re-rated software names exposed to AI disruption risk, analysts considered the ~30% long-term growth assumption against a stock still priced for more and chose to lower their price targets.

Finally, and most recently: the August 5th 2026 earnings release. Q2 2026 revenue slightly missed consensus, and AppLovin issued guidance slightly below expectations. The stock then fell 19.66% intraday, wiping out roughly $40 billion in market cap, even as the SEC closed its AXON investigation that same day with no action taken. Several downgrades and price-target cuts followed (Wells Fargo, Piper Sandler, Goldman Sachs, BofA Securities).

Bear thesis

Growth is slowing

Revenue growth has declined over the last four periods (75% → 70% → 56% → 53% YoY), and Q2 2026 marks the first guidance miss since the 2021 IPO. The market can read the combination of the two as the signal of the beginning of the end of growth.

As Peter Lynch put it, “small company, big moves, big company, small moves.” Growth is bound to mechanically shrink over time.

As growth declines, it will inevitably revert to its natural, and more cyclical pace: the model depends on the ad budgets of mobile/gaming advertisers, a sector historically more discretionary, as AppLovin doesn’t require multiannual contracts.

In gaming, a significant share of revenue, the segment is mature: growth there already comes more from existing users. Nothing rules out this slowdown continuing into Q3.

Disruption risk

Analysts explicitly cite a risk that AI tools could erode AppLovin Ads’ competitive edge, wondering if 30%+ long-term growth can hold.

Indeed, the R&D headcount remains modest given the stakes. 380 R&D employees, or 42% of the workforce, are tiny compared to the AI resources of Meta, Google, or OpenAI. The current model is mainly based on a data flywheel. If a competitor with stronger AI R&D managed to replicate AppLovin Ads with less data (via more efficient techniques), the moat could give way faster than a simple data-volume catch-up would suggest. But I don’t think it will happen in the next quarter, and the competitors have shown before a propensity to fail outside of their core products.

This point matters even more given that AppLovin’s revenue depends on essentially a single line, AppLovin Ads. Disruption is all the more likely for a single-product company.

Capital allocation

$2.19 billion was repurchased in FY2025, a year that includes the all-time high (late 2025), without a quarterly breakdown of the average price paid to confirm whether the capital was deployed in a disciplined way. This should be put in perspective, though, since the question mainly matters near highs and doesn’t really apply here.

Ongoing lawsuit

Two US law firms (Schall Law Firm and Bragar Eagel & Squire) have announced they are investigating/soliciting shareholders for a possible class action against AppLovin, over allegedly false statements or omissions of information material to investors, regarding purchases between May 10, 2023 and March 26, 2025. This is a private civil action, separate from the SEC investigation, which has just been closed.

What if the bear case were true?

In that case, management would lower its guidance and report below its own internal forecasts and analyst expectations. The indicators to watch then are revenue and net margin. Moreover, products from competitors could emerge, and claim similar results.

Bull thesis

1. A strong growth with normal deceleration

The market seems to be forgetting that the company still posted 53% revenue growth. I personally estimate this is a simple air pocket. Though I say this, it is without certainty and the decline could continue over the coming quarters.

My confidence comes from the company’s track record and from its previous margin expansion, which remains impressive.

To hit these targets, the company has a growth relay in consumer/e-commerce : a self-service platform opened to the public in 2025. Revenue is seasonal, and Q2 is usually the slow quarter. Even so, advertiser spend rose 28% versus the Q4 2025 seasonal peak. The web/e-commerce vertical now runs at a record ~$1 billion annual run-rate, and management estimates this market at 5-10x the size of mobile gaming.

2. Management commentary

The miss came from an execution problem. The CEO states that the shortfall came from the timing of AI model improvements on the gaming side, “we know what happened, and it’s already been addressed.” Advertiser appetite remained healthy during the quarter.

Management maintained its long-term outlook on the August 5th call of ~30% combined annual growth for gaming + consumer.

3. A highly profitable business

The company generated $863.3 million in FCF in Q2 2026 alone, and 32% FCF growth after deducting SBC, still well above nearly the entire software/AdTech sector, with a balance sheet carrying almost no net debt. Incremental costs are minimal, since this is a scalable business.

One of my favorite indicators, the ROIC, indicates a steady growth in profitability, which remains impressive for a company of that size. Every invested dollar is producing almost a third in cash. A declining company would show a fall in this metric.

4. Only one shortfall so far

Even though the company had a miss, it’s a single occurrence since the IPO. They’re only human, after all. By nature, this kind of stumble remains a weak signal, especially when it’s being closely watched by management itself heading into next quarter.

What would invalidate the bull case

Would a new quarter below guidance do it? No, because it’s always possible that a simple air pocket produces a continued decline in install volumes without offsetting price gains.

The real signal would be a confirmed slowdown in consumer/e-commerce growth, since that’s the growth relay put forward to justify the thesis beyond mature gaming.

5. Valuation

The numbers only make sense if you’ve read the two theses section. A valuation without a thesis is just a spreadsheet.

The valuation requires an important assumption, being the continuity of a normal economic environment, i.e. without recession, major credit event, market crash. Any of these could push the stock below the bear case shown here, temporarily, independent of the company’s underlying performance. All figures in the scorecard are pre-tax and pre-fees. Run the numbers for your own situation before drawing conclusions.

The assumptions are the following:

• For all the cases, the margin stays flat.

• For the revenue, it goes from 10% per year in the base case, to 30% per year in the central case, and 40% per year in the bull case.

The valuation scorecard map

My conviction sits on the $1,549 scenario or higher.

I’d imagine that if the gap closes next year, AppLovin could once again trade at a PE of 50, under the effect of irrational exuberance, but not by 2030, if the company holds at 30% annual revenue growth.

The re-rating will mainly depend on the catalysts covered in the next section.

6. Momentum

Overall, crossing fundamentals (covered above) with momentum is an excellent way to improve your returns. You’re never right on your own, especially since a significant share of valuation comes from the multiple.

Catalysts

Q3 results (the major point):

• If AppLovin reports below expectations: this would break confidence in the guidance more durably than an isolated accident. E-commerce would slow if management’s “it takes time” turns into disappointment rather than acceleration.

• It’s possible the air pocket continues into next quarter, or even across a run of quarters with a focus on long-term indicators, producing a scenario similar to MercadoLibre (falling share price alongside renewed long-term prospects).

• If AppLovin reports in line with expectations: the market will likely want a quarter or two of confirmation, and will gently begin a re-rating.

• If AppLovin reports above expectations: the market will forget its fear, and the stock could gain 20% in a single session and keep re-pricing higher over the following weeks.

• Analyst revisions will follow the Q3 results and amplify the move up or down in the share price. The sell-side consensus remains heavily bullish despite the miss: 29 buy / 0 sell / 3 hold across 30 analysts (S&P Global), with an average price target of $575, roughly 66% above the current price ($347). The market has punished the stock more harshly than the analysts who follow the story closely. That’s generally a good signal for a buy-the-dip.

On the shareholder lawsuit: its dismissal could trigger a modest bounce in the share price, while its continuation would weigh on it.

A re-rating could also come if a superinvestor (Pat Dorsey or Steve Mandel) builds a stake, or if buybacks continue. The company could also decide to increase its allocated buyback amount as a signal of confidence.

When to enter the stock?

The market hates uncertainty and has already started a significant decline in the share price. So you need to track the price to see where it bottoms. I see three possible strategies:

Buy the dip:

• I would wait for analysts to stop cutting their outlooks. The likely target zone is a PE of 20-25.

• It’s hard to give a PE ratio as a support level, since the company has broken through its last support on the way down.

• This works if AppLovin surprises to the upside next quarter, and fails if AppLovin disappoints or worries the market for a second quarter.

Capitulation:

• It’s possible we head toward a very bearish scenario if the company reports below expectations. In that case, you have to wait for capitulation, which shows up through several signals, such as above-average volume, or RSI <20.

• So far, AppLovin hasn’t shown signs of capitulation.

• DCA: the way to reduce risk (the non-choice) is to run a DCA every month, if that’s your approach. As for me, I prefer to enter in one or two tranches.

As for me, I plan to open a position once the price stabilizes, and I’ll let my subscribers know when I do. I’m prepared to add to the position on further weakness or capitulation. There’s no rush, the thesis playing out will probably take 2-3 years. In Fortinet’s case, which shows some similarities, it took a year between the August ’25 decline and the re-rating over April-July ’26.

Final Thoughts

I think that starting at a PE of 22, AppLovin becomes a great buy, and starting at a PE of 16, a screaming buy. Who knows where the share price will end up falling? It depends heavily on investor sentiment. I plan to enter soon, because I’m convinced the company has only hit an air pocket. It’s possible, but in my view unlikely, that next quarter comes in below expectations. It should legitimately either be in line or beat expectations. As mentioned earlier, buying the dip can be either the best or the worst investment decision there is. I published this deep dive ahead of my investment decision, unlike with Booking and Microsoft, to give you time (and to give myself time) to properly weigh every element before making a call.

Subscribe now

Disclaimer

All content published by JB Peter on this platform is strictly for educational and informational purposes. It does not constitute investment, financial, legal, or tax advice, nor does it represent a personal recommendation or solicitation to buy or sell securities. This research is operated by ORIACON (SASU) and reflects independent corporate analysis. Every reader must conduct their own independent research (Due Diligence) or consult a licensed professional before making any financial decision, as financial markets involve a high risk of capital loss. At the time of writing, ORIACON or the author DO NOT HOLD shares in the company analyzed in this article. Following this publication, ORIACON and the author reserve the right to buy, sell, or modify positions in any security mentioned at any time, without prior notice to readers or subscribers.

As you can deduce from the name, the central thesis is that harmonic reducers (HDS) and planetary roller screws (THK) are in short supply, with demand exceeding supply. It gives us every aspect we want as investors: prices increases, production and guidance increases, and multiple expansion (PE ratio, Price to Sales, etc.).

I wanted to make an update since both companies beat expectations, both raised guidance and both fell after their results. This article will provide you an update on:

1. Why they fell

2. The bull vs bear part I made in the previous article

3. The valuation scorecard

Subscribe now

1. Why HDS and THK fell

Sector rotation and macro

The market feared an intervention on the yen, as the currency was weakening. The government would sell dollars and buy yen: a stronger currency would weaken the industrials such as THK and HDS.

Moreover, there was a broad AI/semiconductor selloff on August 6-7. Other names in the sector, such as Kioxia, Taiyo Yuden, Murata and Tokyo Electron, were all down on the same days. THK and HDS got dragged along with the rest of the sector.

Pre-earnings rally

THK was up roughly 23% in the week before its own results, and roughly doubled over the past year. HDS was sitting near its 52-week high, already a 3-bagger YoY.

Results were below expectations

Here is a quick update on the results:

As the market rallied before the results, the expectations were obviously stronger.

THK’s net income is up 805% YoY. It was mostly a one-off gain from the divestiture of its automotive unit. The truth lies in a normalized net income closer to +364%.

There was no explicit mention of take-off in humanoid robotics, so the market feared the AI thesis wasn’t there: HDS’s guidance hike came with just “industrial robots and semiconductor manufacturing equipment”, and THK’s growth is also concentrated in “electronics and semiconductor” demand.

2. Bull vs bear update

I made a bull vs bear in the previous article, some points may have changed. It’s worth noting THK’s cost-to-sales ratio improved 4.6 points year over year. HDS’s operating margin for a single quarter already exceeds its full-year FY2027 guidance. It goes into the good direction.

Bear argument #1: China competition

Goldman Sachs conducted a field survey of 9 suppliers in China in November 2025. They planned an increase in capacity, from 100,000 to 1M units. Moreover, the direct competitor of HDS, LeaderDrive, mentioned booked capacity through 2027, with production expected to reach 1M units in 2026.

Bear argument #2: Price rally and cyclicality instead of robotics

As said before, both stocks went up significantly. The market is still cautious about evidence of bottleneck pricing, regarding the first bear point. Neither company has mentioned humanoid-specific segments. What is sure for the moment is the capex and the increase of competition on the supply side. If there is a bottleneck, nobody knows when it can stop.

Bull argument #1: Demographic tailwind / labor shortage

This point is unchanged from the previous article and certainly won’t be any different throughout the investment period. Demographics aren’t in quarters, but in decades.

Bull argument #2: Operational execution

Both companies raised the guidance, a good signal for the continuation of the thesis. The sustained execution is a necessary condition for the whole bull thesis to materialize.

Bull argument #3: Real barriers to entry

This point is unchanged from the previous article and certainly won’t be any different throughout the investment period. The qualification cycle for a new supplier is usually several quarters: by that time, the thesis will be played out.

Bull argument #4: The fragmentation of the world into blocs

I said in the previous article “each bloc (Western, and within the West, the EU, the US and Japan taken separately) will want to keep strategic suppliers tied to its own bloc rather than depend entirely on an outside supplier. This isn’t just a price or volume argument: it’s a supply-security logic that structurally leaves room for a “trusted” bloc supplier — even at a higher price, even with less volume than the cheapest Chinese option. Both HDS and THK benefit from this mechanism, independent of their pure price competitiveness. “

This argument has another confirmation in other compartments of semiconductors:

• On rare earth elements, China restricted access to Japan for dual use and permanent magnets over Taiwan tensions

• On chipmaking equipment, U.S. export controls target American toolmakers themselves, while Japan’s Tokyo Electron and the Netherlands’ ASML negotiated carve-outs letting them keep selling into China.

The logic is the following: if possible and for strategic sectors, the US companies would get a Tier 1 group of US companies, then a Tier 2 group of western companies (EU, Japan, Korea, etc.) and then a Tier 3 group of companies (China, rest of the world) with a focus on price.

New: Bull argument #5: The main competitor of HDS, LeaderDrive, is full until 2027

The other side of the first bear argument is that if LeaderDrive is in a supply shortage and there is room for western companies, then HDS should also be in a supply shortage for the coming quarters.

New: Bull argument #6: Despite growing capacities, the Total Addressable Market is growing significantly

HDS is adding 33% more capacity at its Beverly, Massachusetts plant by December 2026, on top of 13% already added in December 2025. In Japan, roughly ¥10bn is going specifically into humanoid-reducer production at the Ariake plant, targeting ¥10-20bn of revenue from that line in FY2027.

Goldman Sachs has also revised its humanoid market-size estimate upward since the original article: from roughly $3-5bn today to $15-19bn by 2030, and $38bn by 2035. The insight I want from this is that the runway until market saturation is longer than expected, allowing HDS and THK to compound for longer.

3. Valuation scorecard

Despite revenue growth running below the +50% YoY assumption from the previous article, the net margin of THK is already running above the scorecard’s Bull case. HDS’s Q1 margin is climbing faster than expected toward the Central case. I updated the valuation scorecard to reflect the shorter horizon now that we're getting closer to 2030. I expect the price to increase several times based on sentiment in the coming quarters, then I would sell the news before the price drop. The tricky part is to know when to sell cyclical stocks: you’ll never get the perfect tops and bottoms.

Conclusion

There’s still no clean evidence of a bottleneck, but both companies are tracking their bullish-scenario margins already — what the coming quarters should settle is how much of that holds up. The low weighting of stocks in the portfolio reflects this lack of predictability, although the balance between upside and downside potential remains positive. I would consider increasing my position in both if I had additional capital to deploy, given that the fundamentals have improved faster than the price.

Subscribe now

Disclaimer

All content published by JB Peter on this platform is strictly for educational and informational purposes. It does not constitute investment, financial, legal, or tax advice, nor does it represent a personal recommendation or solicitation to buy or sell securities. This research is operated by ORIACON (SASU) and reflects independent corporate analysis. Every reader must conduct their own independent research (Due Diligence) or consult a licensed professional before making any financial decision, as financial markets involve a high risk of capital loss. At the time of writing, ORIACON or the author HOLD shares in the companies analyzed in this article. Following this publication, ORIACON and the author reserve the right to buy, sell, or modify positions in any security mentioned at any time, without prior notice to readers or subscribers.

Despite fears of AI disruption, tourists are still booking with Booking. Room nights grew 5% YoY, the operating margin improved from 33.1 to 34%. Free cash flow minus SBC is up 17%. A pretty solid quarter, beating expectations across the board.

Verdict: I would love to add content on it, but nothing to see here, folks. Everything is fine.

My reaction

Booking is up today because such a juggernaut with a PE less than 25 times makes no sense. The numbers in the valuation section are lower to Novo Nordisk, but Booking has more love from the market, and the fears are easier to dissipate.

I think I will make a 100% return in the coming quarters and sell it for another GARP in a downturn.

As a client, I like Booking, and I really like it as a shareholder with its impressive net margin and no clear competition. Airbnb’s prices increased, so I perceive less value from it.

Booking is a clear example of “Invest in what you know” and an easy business to understand. Investing feels sometimes like an intellectual race, but the reward lies mostly in the simplest businesses.

2. The Numbers

2.1 At a glance

This table summarizes the key metrics from Booking’s Q2 FY2026 earnings release: https://ir.bookingholdings.com/overview/default.aspx

2.2 Commentary

Revenue :

Total revenue grew 8% YoY to $7.35B (7% FX neutral):

• Merchant revenues (Booking directly bills the final client, and afterwards pays the merchant) grew 15%.

• Agency revenues (Booking takes a commission on the reservation) fell 7%.

• Ads : +8%

The opposite direction between merchant and agency is following the Booking business model transition. So it is going as expected, and room nights are in line with the global increase.

Margins:

Net income margin grew to 26.5% from 13.2%. The number is overrated, since a non-cash FX gain on euro-denominated debt of $195M (vs a loss of $962M a year ago) occurred this quarter. The best metric to follow is the GAAP operating margin at 34.0% (vs. 33.1%, +0.9pt) driven by fixed-cost leverage (total operating expenses +7% vs revenue’s +8%) and cost reduction.

The Transformation Program expects an annual run-rate savings raised from $550M to ~$650M, targeted by end of 2027 — confirms the deep dive’s “more capacity to invest” thesis.

Net income and free cash-flow

As mentioned above, the FX gain on the debt is distorting the net income: 118% YoY. I prefer to track the free cash-flow (with subtraction of SBC), which grew 17% YoY.

The company bought back $3.7B in shares (vs. $3.6B in Q1 2026) and paid a $0.42/share dividend.

3. Key Takeaways from the Earnings Call

AI dominated the Q&A — 5 of 9 analyst questions were AI-related. Still a concern for most.

AI disintermediation: management put a number on it. Asked directly about AI’s impact on the top line, CFO Ewout Steenbergen volunteered the most concrete data point of the call: traffic arriving from large language models — paid and unpaid combined — is “still significantly below 1% of our room nights,” and “that hasn’t moved so much recently... no material change over the last few months or quarters.”

This is the first time management has quantified the AI-referral contribution, and it lands squarely in favor of the deep dive’s bull case: the disintermediation threat remains a future risk only.

A new crack worth watching: SEO pressure, mentioned for the first time. Management acknowledged declining organic search traffic: “I do believe that some of the changes that were made in the display at Google definitely put some pressure on SEO. Putting in that AI overview probably has done it. Now, here’s an important thing to keep in mind. So, our direct, our direct mid-60s, hasn’t gone down at all.”

Management is citing Q3 as the key quarter to watch: peak European summer, the first full quarter of Transformation Program savings, and the point where Middle East headwinds finally lap their own anniversary, normalizing the base effect.

4. Theses Update

4.1 Bear Thesis Signals

Room nights declining in absolute terms on core Booking.com. Not triggered. 325M, up 5% YoY, a record.

Direct booking rate below 60% and falling. Not triggered, though the sources disagree on the number. The 10-Q says mid-fifties, flat versus last year. Fogel said mid-60s on the call, also flat versus last year. Pick whichever number you trust more, but none of them is falling.

Customer acquisition cost rising specifically on accommodation. Inconclusive. No segment-level disclosure exists. Marketing spend as a share of gross bookings ticked up 10 basis points, to 4.7%. But is going from 4.6% to 4.7% significant?

An AI agent completing hotel bookings at scale outside an OTA. Not triggered. Management put a number on it this quarter for the first time: LLM referral traffic is below 1% of room nights, and that share hasn’t moved in months.

4.2 Bull Case

FCF minus SBC growing. Confirmed. $3.50B this quarter, up 17%. Pretty solid.

Operating margin expansion. Confirmed with the GAAP Operating margin at 34.0%, up 90 basis points.

Transformation Program flowing to the bottom line. Confirmed, and moving faster than expected. The savings target rose from $550M to roughly $650M this quarter. Another $100M got identified on top of that, from B2B consolidation.

Genius as an anti-disintermediation moat. Confirmed, strengthening. Level 2 and 3 members now account for a high-50s share of room nights and more than 30% of the active customer base. Both figures are up year over year.

Aggressive buybacks at depressed prices. Confirmed, unchanged pace. $3.7B this quarter. $14.5B of authorization still on the table.

Direct channel above 60% and rising. The exact level is disputed between the 10-Q and the call. It is the same argument as bear signal #2.

5. Valuation

I’m sticking with the original deep dive assumptions. I updated the current date and the EPS. The end date stays fixed, which mechanically inflates the annualized figures, since the same target price is now spread over a shorter time horizon.

The normalized P/E is 21.6x, based on normalized earnings of $7.11B, driven by 1) adding back the $457M KAYAK impairment recognized in Q3 2025, 2) removing the $581M non-cash FX gain on euro-denominated debt that artificially inflates TTM GAAP earnings, and 3) the $25M tax benefit associated with these two adjustments.

CAGR calculated from $204.1 to June 5, 2028 (1.83 years), dividends included.

How to read this table: in the central scenario at PE 30x, the price target reaches 358$ by December 31, 2028, a total return of +75% from current price, or a CAGR of 36.0% including dividends. It is the floor of my conviction, since 30 is still a quite conservative PE ratio for a business of this quality.

I’d consider selling above 35x.

Subscribe now

Disclaimer

All content published by JB Peter on this platform is strictly for educational and informational purposes. It does not constitute investment, financial, legal, or tax advice, nor does it represent a personal recommendation or solicitation to buy or sell securities. This research is operated by ORIACON (SASU) and reflects independent corporate analysis. Every reader must conduct their own independent research (Due Diligence) or consult a licensed professional before making any financial decision, as financial markets involve a high risk of capital loss. At the time of writing, ORIACON or the author HOLD shares in the company analyzed in this article. Following this publication, ORIACON and the author reserve the right to buy, sell, or modify positions in any security mentioned at any time, without prior notice to readers or subscribers.

The central point of this quarter is the guidance raise for 2026, with adjusted sales now expected to grow from flat to -6% at CER, vs. -4% to -12% previously. Management gave us two reasons to adjust sales: one good, a provision reversal, and one semi-good, with impairment of terminated drugs.

The main growth is made by the obesity segment, and inside it, by the Wegovy pill (almost none by the injectable Wegovy). I expect the worldwide launch of Wegovy HD to provide additional revenue and operating profit.

Verdict: The volume effect, which is positive, is now outweighing the price effect. In front of us, I see the upside; behind us is the downside.

Subscribe now

My reaction

I published a note yesterday about how much I dislike the word adjusted. Here I don’t see any partial presentation of the truth. The ‘adjusted’ framing asks readers to do extra work to assess the current situation. That’s part of what I do here, and it’s normal for me to provide it to you. Management is right to give us this presentation, as it delivers the most precise picture of the business health.

With Novo Nordisk, being contrarian feels like a long path of patience, and any bottom-sell could cost you a lot, just as being too early could. It was a market darling, and then a fallen angel.

The only things that could resolve its situation are time and work (from the company). They are on track: the downside is certainly behind us, and the upside is coming.

The numbers, the bear thesis status, and the bull thesis are here to validate these points.

As a personal view, somebody from my family is taking a GLP-1 drug. It turns out she was prescribed Zepbound, presented to her as ‘more effective.’ I have to admit I didn’t take that news particularly well, as a Novo shareholder. I’m fairly convinced that going forward, Novo’s prescriptions will pick up again with Wegovy HD.

2. The Numbers

2.1 At a glance

This table summarizes the key metrics from Novo Nordisk’s Q2 FY2026 earnings release: https://www.novonordisk.com/content/nncorp/global/en/news-and-media/news-and-ir-materials/news-details.html?id=916590.

2.2 Commentary

Adjustments

Management provided adjusted numbers relative to 2 one-time events.

• Impairment of DKK 6B, with DKK 4B to the sole monlunabant drug. Even though it is a non-cash expense, it carries a real cost. Additional impairment could occur in the coming quarters, depending on the results (and terminations) of the pipeline. I see it as something you should be prepared for as a pharmaceutical company shareholder.

• Rebate provision reversal: Under the U.S. 340B Drug Pricing Program, manufacturers must give discounts on drugs sold to hospitals and clinics serving low-income patients. Novo booked a provision in ‘25, which turned out, this quarter, to have been too conservative. The release of this provision gives a one-time boost to reported sales and profit. The underlying performance of the company is best read without this adjustment.

In my opinion, the best number for profit analysis is without rebate provision reversal and with impairment, as it is a “recurring” expense in the life of such a pharmaceutical company.

Revenue

Q2 2026 reported sales increased by 3% at CER (Constant Exchange Rate), and adjusted sales increased by 7% at CER. The volume effect, which is positive, is now outweighing the price effect, which had been reduced due to U.S. government decisions.

As expected, Obesity care accelerated, with +16% CER YoY, vs 3% for the diabetes segment. I wrote in the deep dive that this second segment is mature and commoditized, so there is no surprise here.

Inside Obesity:

• The injectable Wegovy is only growing 1% CER.

• The heavy lifting is done by the Wegovy pill.

• With the worldwide launch of Wegovy HD, I expect clear traction to return for the injectable (for Q3 in the EU). As said previously, the drug efficacy gap with Zepbound resolved with Wegovy HD.

• International Operations (unadjusted) sales grew 10% CER, whereas US Operations (unadjusted) sales shrank 2% CER.

Margins

The adjusted gross margin is shrinking from 82.7% in Q2 2025 to 78.2% in Q2 2026, due to “lower realised prices, one-time costs of around DKK 3 billion related to right-sizing of manufacturing capacity agreements, as well as a negative currency impact. This was partly countered by productivity gains and a positive product mix.“ (information from Novo Nordisk).

The adjusted operating margin is improving from 41.8% to 42.5% YoY, as Novo reduced its workforce and made efforts on cost to maintain its profitability.

Any additional price drop could lower the margin. As shareholders, we hope that the US government will move to another topic.

Without any change on this subject, we can expect the margin to be near its bottom with volume increase, price stability and additional variable cost reductions.

Net income and free cash-flow

Financial charges increased from DKK 1.95bn to DKK 2,76bn, with a higher interest rate on the debt. Worth mentioning, but nothing alarming here.

Net income fell 21% due to the adjustments, and free cash-flow increased by 57% YoY since the capex fell 14% CER YoY due to a higher base effect in 25. Investments are still strong as the future needs way more GLP1 drugs.

3. Key Takeaways from the Earnings Call

Novo’s management stays conservative in its assertions, as Novo Holdings Foundation controls 77% of the voting rights, so management has no structural incentive to overpromise to put the stock price higher. It is reinforced by a danish corporate culture of humility and trust. That’s my opinion and I’m not Danish, but I come from Central Europe and people tend to behave like that.What is a good point for trust is a bad point for rerating.

As the thesis is focused on obesity, I won’t elaborate on the pipeline.

Guidance. Framed as “a better than expected start to the year.” Management confirmed the outlook stays conservative on the Medicare Bridge program, even after a direct push from an analyst noting a competitor has been more bullish on volume inflection. This is consistent with the “guidance was sandbagged” read.

Access. “From an access perspective, quality of access for obesity GLP-1s remains poor and is a key focus area. While we see uptake for the pill in the reimbursed commercial channel, the majority of total prescriptions are self-pay.”

Manufacturing capacity. On the new API facilities: “validated the first product in the first facility... very low utilization from these facilities... bullish in terms of being able to supply significant volumes in the years to come.” Early innings on the capacity ramp. The moat is intact. It’s not yet monetized.

Cost savings, ahead of plan. The Q3 2025 restructuring is tracking ahead of its DKK 8 billion savings target. Headcount down to 66,700. That’s a cut of roughly 12,000 people, 15%, over the past year. Harder than the 9,000 figure originally announced.

Wegovy pill, international. UK: 300,000 patients in the first three weeks, versus 48,000 in the US over a comparable window. UAE: Novo’s obesity market share jumped from 30% to 45% after the pill launch. The pill already holds roughly 50% of the oral segment there, despite a competitor’s oral drug launching about a month earlier (surely Eli Lilly…). Germany launches in September. Around 80% of pill patients are GLP-1-treatment-naive. Limited cannibalization of injectable Wegovy.

What’s next. Capital Markets Day, September 2026. Formal 2027 guidance waits until February. The CFO already flagged two swing factors: the CagriSema US regulatory decision, expected by year-end, and the “mim8” decision, in H2 2026.

4. Theses Update

Anchored to the deep dive published 26/05/2026 (”Novo Nordisk: A GLP-1 Banger”).

4.1 Bear Thesis Signals

• No guidance beyond 2026. Nothing in this report addresses 2027 and beyond. The raise only narrows the 2026 range, from -4% to -12% CER down to 0% to -6% CER.

• Lilly winning the GLP-1 market (57% share). Lilly reports Q2 2026 the same day as Novo. One data point does stand out: Wegovy remains the U.S. market leader by new patient starts among branded obesity drugs, even as Zepbound leads on total prescriptions. I will provide content about Lilly vs Novo later.

• Structural margin compression (MFN). Confirmed, but smaller than the headline number suggests. Adjusted gross margin fell, indeed, but Novo discloses that roughly DKK 3 billion of that hit is a one-time cost tied to right-sizing manufacturing capacity agreements, layered on top of lower realised prices and a negative FX effect. Strip out the one-off and the clean MFN-driven compression is well under the headline -4.5pp.

• New this quarter: pipeline disappointment. Monlunabant’s development was terminated, “due to portfolio considerations,” triggering a DKK 4.0 billion impairment. ZEUS (ziltivekimab, cardiovascular) missed its primary endpoint. Neither touches the core obesity thesis directly. Both are new data points on non-GLP-1 pipeline execution that didn’t exist in May.

• New this quarter: quality of growth. Trade receivables grew 25.5% since year-end, DKK 70.9 billion to 88.9 billion, roughly twice the pace of H1 revenue growth. Plausibly just the ~60-country Wegovy rollout. Still, a working-capital line to watch next quarter.

4.2 Bull Case Prerequisites

• Efficacy gap with Zepbound closes, via Wegovy HD. In progress. Launched in the US in April, in the UK in June alongside the single-dose pen. No comparative prescription data yet.

• Valuation gap vs. Lilly (PE 10x vs. 36x) is an anomaly. Still wide open. At ~308.7 DKK, the stock prices closer to the Bear and low-Central scenarios than the Bull case. See the updated scorecard in section 5. The market is rewarding today’s execution without re-rating the multiple.

• Prices fell, volume hasn’t followed. Yet. This is the prerequisite that finally shows up in the numbers. Obesity care accelerated to +16% CER. Wegovy pill scripts kept climbing past 265,000 weekly by mid-July. Volume is now outweighing price at the net-sales level: +7% adjusted CER, despite ongoing MFN cuts.

• Guidance was sandbagged. Confirmed again. It is the second consecutive raise under Doustdar.

• Production moat: 14 fill-finish sites, 5 to 7 years to replicate. Reinforced by the same capex and net-debt evidence above. The moat is largely built and starting to convert into cash rather than absorbing it.

• Market stays a duopoly through 2028. No new information either way this quarter.

• CagriSema delivers in Phase 3. Mixed. REDEFINE 9, the lower-dose trial, showed statistically superior weight loss vs. placebo. It also lands in the same quarter as the monlunabant termination and the ZEUS failure.

5. Valuation

I’m sticking with the original deep dive assumptions. The only update is the current price. The end date stays fixed, which mechanically inflates the annualized figures, since the same target price is now spread over a shorter time horizon.

CAGR calculated from 308.7 DKK to December 31, 2028 (2.4 years), dividends included.

How to read this table: in the central scenario at PE 22x, the price target reaches 572 DKK by December 31, 2028, a total return of +95% from current price, or a CAGR of 32.0% including dividends. It is the floor of my conviction.

The assumptions depending on the PE:

• PE 15x — The market sees a mature pharma under pressure, losing market share to Lilly, facing patent cliff in 2031. Just a dividend stock.

• PE 22x — The market recognizes a quality compounder, solid second in a structural duopoly, with a manufacturing moat. GARP territory.

• PE 30x — The market prices leadership potential. Wegovy HD closes the efficacy gap, the oral pill scales, CagriSema delivers. Novo challenges for pole position.

Subscribe now

Disclaimer

All content published by JB Peter on this platform is strictly for educational and informational purposes. It does not constitute investment, financial, legal, or tax advice, nor does it represent a personal recommendation or solicitation to buy or sell securities. This research is operated by ORIACON (SASU) and reflects independent corporate analysis. Every reader must conduct their own independent research (Due Diligence) or consult a licensed professional before making any financial decision, as financial markets involve a high risk of capital loss. At the time of writing, ORIACON or the author HOLD shares in the company analyzed in this article. Following this publication, ORIACON and the author reserve the right to buy, sell, or modify positions in any security mentioned at any time, without prior notice to readers or subscribers.

EverQuote delivered on revenue, growing 25% year-over-year. Operating margin expanded from 9% to 12%, and operating income grew 65%. EPS growth was muted by a tax headwind, but the underlying results are solid.

Nothing here shows any sign of disruption to the business today.

EverQuote delivered another quarter of growth and profitability.

The investment thesis has always rested on a simple observation: a platform like EverQuote already exists and works in Europe — MoneySuperMarket and GoCompare in the UK, CHECK24 in Germany, LesFurets in France — and the US is simply following the same path with a lag. This is a genuinely good business that the market has beaten down.

With a PE of 11.4 and a PFCF-SBC of 12.5 on a price of $24 (premarket), EverQuote is a quality small-cap that should re-rate toward double its current price

Verdict: Another quarter, another fear of disruption that is still not showing up. Re-rating will come.

Subscribe now

My Reaction

Management is talking extensively about its EBITDA Adjusted.

To focus on this point, I dislike the usage of EBITDA, as it doesn’t reflect the true health of the company.

But Adjusted EBITDA? It is something I strongly dislike:

• The term isn’t standardized and its definition can be adjusted to whatever management wants to show you.

• If the GAAP term were good, they would present it.

• It gives me a seductive vision from management, showing you the parts of the truth they are willing to and not the actual situation. I cannot feel trust in such conditions.

In this kind of situation, I usually check for financial shenanigans, to see if the partial vision provided by management is also distorted in the financials.

Management defines it as: “net income (loss), excluding the impact of stock-based compensation expense; depreciation and amortization expense; restructuring and other charges; acquisition-related costs; legal settlement expense; interest income; and income taxes. “

Luckily, the numbers for EverQuote are still strong and the net cash position tells you the company generates enough cash to fund itself and still return some of it to shareholders

But still, I won’t comment on any Adjusted EBITDA metrics.

I have held EverQuote since April 30, 2026 and I’m up 60% at the time of writing, a pretty solid result.

The company is heavily profitable, and priced like it’s dying. I still don’t see any disruption. Management is able to take advantage of the undervaluation to buy back as many shares as possible, a good point for the long term. I can clearly see the company trading around $100 within a few quarters

When? The main question is when the market will consider that EverQuote is a winner, and won’t be disrupted anytime soon.

2. The Numbers

2.1 At a Glance

This table summarizes the key metrics from EverQuote’s Q2 FY2026 earnings release: https://investors.everquote.com/news/news-details/2026/EverQuote-Announces-Second-Quarter-2026-Financial-Results/default.aspx

2.2 Commentary

Revenue: 15% year-over-year in Q1, 25% in Q2. Those are good numbers, as they usually increase the operating leverage. But it cannot be found yet in the increase of earnings. This is the reason, in my opinion, why management is talking about EBITDA Adjusted.

The main activity is in the Auto segment (+23.3%, $172.1M), but the Home & Renters segment is gaining more traction +35.2% ($23.0M). A good signal, as it isn’t their core activity and they take market share from their competitors - online: MediaAlpha and offline: TV, display advertising, physical agents (i.e., in-person/traditional insurance agents).

Operating Margin : 12% in Q2 vs 12.3% in Q1, vs 9% in Q2 ‘25. The operating income grew 65% YoY for just 25% growth of revenue. The leverage is the key to the rerating and earning growth I am looking for. It went from a small basis, though.

Net income and Free cash flow: Free cash flow - Stock-based compensation dipped sequentially, $22.8 million versus $28.1 million in Q1. The dip reflects a working-capital timing effect.

Income before income taxes is up from $15.1M to $24.5M (+62% YoY), but the tax rate went from 2% to 21%, due to a one-time tax credit last year. It was a one-time event that won’t occur next quarter.

Cash position : $192.3 million in cash, zero debt, $50 million bought back since last August. Net cash as a share of market cap fell from 32.1% in Q1 to 23.0% in Q2. This number is pretty satisfying, as this “lowers” the effective PE ratio by 23%, and management will be able to buy shares back and push the share price higher.

3. Key Takeaways from the Earnings Call

Management spent a lot of the call on AI — a new bidding tool called Smart Campaigns, internal AI agents, early work on capturing traffic from AI search and what they called “agentic commerce.” It’s a real shift in framing: AI used to come up as a threat to comparison platforms like this one, and now it’s being pitched as EverQuote’s own weapon.

Management follows a metric, the Variable Marketing Dollars (VMD), which is Revenue minus the expense for traffic acquisition from Meta, Google, etc. It tracks the true value the company adds. The ratio of this metric is the Variable Marketing Margin (VMM).

The most-asked question on the call, more than once, was whether the margin would compress as the company invests in AI. CFO Joseph Sanborn kept steering it back to dollars, saying the company does “not solve for VMM margin, we solve for VMD,” and reiterated a target in the high 20s. That’s a defensible way to run the business, and keeps a focus on operational leverage.

The AI story moved from a defensive talking point to an offensive one. For the moment, only the story changed: it is still to be found in the VMM.

Combined ratios stayed healthy — “in the 80s,” per management — which is the clearest evidence available that the 2023 collapse was cyclical, not structural.

On capital allocation, management laid out three priorities in order: organic growth, buybacks ($50 million cumulative), and M&A as a lever toward the $1 billion target. No sign of a deal happening soon.

4. Where the Investment Thesis Stands

I had no time to formalize a deep dive yet, although I meticulously checked the company on a checklist I have.

4.1 Bear Thesis Signals — Triggered or Not

- AI disintermediation (Google/Meta build native insurance comparison tools) → not triggered. Management is positioning EverQuote as an AI infrastructure partner instead, though that’s a claim only and not a measured outcome yet.

- Traffic dependency (rising Google/Meta click costs compress margin) → not triggered. VMM held at 29.2%, in line with last year.

- Rising carrier combined ratios cutting marketing budgets → not triggered. Combined ratios described as healthy, in the 80s.

- Stock-based compensation dilution → partially active. SBC is real ($10.8 million year-to-date) but partly offset by $29 million in buybacks over the same period.

- Product concentration in Auto insurance → still structurally true (88% of Q2 revenue), though Home is growing faster and slowly diluting it.

4.2 Bull Case Prerequisites — Status

- Recovery of the auto insurance cycle → confirmed. Healthy combined ratios, accelerating Auto growth.

- Operating leverage from AI and scale → confirmed and strengthening. Operating income growth nearly tripled revenue growth.

- Disciplined capital return → confirmed. $50 million bought back, consistent pace.

- Diversification into Home insurance → in progress. Home grew 35.2% against Auto’s 23.3%.

- Path to $1 billion in revenue → reaffirmed by management this quarter.

- VMM expansion above 35% → not observed. Still at 29.2%.

Subscribe now

Disclaimer

All content published by JB Peter on this platform is strictly for educational and informational purposes. It does not constitute investment, financial, legal, or tax advice, nor does it represent a personal recommendation or solicitation to buy or sell securities. This research is operated by ORIACON (SASU) and reflects independent corporate analysis. Every reader must conduct their own independent research (Due Diligence) or consult a licensed professional before making any financial decision, as financial markets involve a high risk of capital loss. At the time of writing, ORIACON or the author HOLD shares in the company analyzed in this article. Following this publication, ORIACON and the author reserve the right to buy, sell, or modify positions in any security mentioned at any time, without prior notice to readers or subscribers.

1.1 Synthesis

Atos is coming back from a near-bankruptcy: at the end of 2024, a commercial court ordered the restructuring of its balance sheet. Today, the firm is progressing toward its 2028 objectives:

• €9-10 billion in revenue

• 10% operating margin

• An investment-grade credit rating profile

The first chapter is the cost-reduction phase of the Genesis plan, delivered ahead of schedule. The initial target was achieved within a year and has since been raised to €800M.

The second chapter is the improvement in operating margin: €190M in H1, up 43% year-on-year, from 3.7% in H1 2025 to 5.7% in H1 2026.

Management indicates that the actions behind these two chapters are complete and will continue to deliver results over the coming quarters.

The next chapter is commercial expansion, where management’s focus is now shifting.

The balance sheet is expected to improve significantly by year-end, supported by continued progress on debt refinancing.

None of the bear thesis’s warning signs have been triggered, and the bull thesis is gradually playing out.

Verdict: The company is progressing through its turnaround. The market still applies a discount, and management is advancing relentlessly toward resolving it. The price does not yet reflect the positive developments of the first half.

Subscribe now

1.2 My Reaction

Management has been heavily focused on cost-cutting measures and will continue acting to protect the margin of the business. I’m still very bullish on Atos, and I think the market is still pricing it as though the company will remain a broken IT services business.

They have succeeded on the cost-cutting front, are well on their way to improving operating margin, and are removing black contracts wherever they can — the ones that generate extra revenue but kill the margin.

They are now focusing on revenue: the commercial pipeline is growing significantly, and the 91% book-to-bill ratio, while it looks disappointing, doesn’t include framework agreements. Some of their clients were forced to end their contracts because of Atos’s previous sub-investment-grade status, and are keen to come back as soon as they can.

I’m a bit puzzled by the -7% move on results day. Turning around a company like Atos isn’t easy, and the bull thesis is playing out in the right direction — just not as fast as the market seems to expect.

I’m still confident the stock will move upward over the coming quarters. One day, management could raise guidance, and the stock will re-rate. And I’ll be there. I don’t mind waiting two years for a 5-20x return, because I’m convinced Atos will turn around.

2. This Quarter’s Numbers

2.1 At a Glance

This table summarizes the key metrics from Atos’s Q2 FY2026 earnings release: https://www.atosgroup.com/en/investors/financial-results-reports/h1-2026-results

2.2 Commentary

Numbers are those of ’26 vs. ’25.

Revenue (go-forward perimeter): €3,304m vs €3,626m and operating margin: 5.7% vs 3.7%

Management is “cost-killing” and removing black contracts with two effects: the revenue goes south and the margin goes north. Revenue will improve in the next quarter, as will the operating margin. The revenue decline is decelerating every quarter since ’25.

Organic growth: -8.9% vs n/a
H1 total organic growth rate, calculated by Atos at constant scope and FX. Not applicable to H1 2025 since it’s the comparison base.

Underlying EBIT: €190m vs €133m
+43% (+€57m) year-on-year. Every improvement in revenue will drastically increase EBIT.

GAAP net income: -€504m vs -€695m
Loss narrowing by €191m year-on-year, but still heavily weighed down by one-off refinancing costs (call premium, accelerated IFRS9 depreciation) and restructuring charges — not a clean read on underlying profitability yet. Management targets a “clean” P&L by year-end 2026.

Operating cash flow: -€120m vs +€28m (published, not comparable basis)
The -€120m figure includes €127m of restructuring cash-out — strip that out and underlying cash generation is close to flat to slightly positive. The H1 2025 published figure isn’t calculated at the same current perimeter, so it’s not a clean comparison; treat both cautiously.

Total liquidity: €1,805m vs N/D (30 June 2025)
Comfortably above the €650m covenant floor. Next payments are in 2030, giving management the time it needs to turn the company around.

Book-to-bill (global): 89% H1 / 91% Q2 vs N/D (Q2 2025: 84%)
Below 100% at face value, but methodologically conservative — framework agreements (the €187m European public sector deal, the Dutch police contract) aren’t counted. The +7pt year-on-year improvement, and >100% book-to-bill in France, UK and Eviden specifically, tell a stronger story than the headline number.

Renewal rate: 94% vs 91%
Confirms the “back to normal” narrative management pushed on the call — this is roughly where renewal rates sat before the 2024 restructuring shock hit client confidence.

3. Earnings Call Key Takeaways

Genesis ahead of schedule: Phase 1 (€650M) completed in one year, by Q1 2026. Phase 2 launched, horizon 2027-2028, total target raised to >€800M. Total restructuring cost envelope confirmed at ~€700M (~€200M spent in 2026).

Margin/revenue mechanism openly acknowledged by management:

Philippe Salle, CEO:

• “Whatever happen on the top line, we will deliver the bottom line”

• “I have just adjusted Genesis to protect the margin”

A nice way to say that he will reduce workforce whether or not it’s needed.

Official book-to-bill is conservative: 91% (+7 pts YoY), but framework agreements (the €187M European public sector contract, Dutch police, European Patent Office) aren’t counted. France, UK and Eviden all above 100% in Q2.

Qualified pipeline: +€900M in Q1, +€760M in Q2 = roughly +€1.7bn cumulative over H1.

No net client losses in H1 2026 — a contrast with 2024-early 2025.

Management posture shift: “refocus the mind of the management to the top line” — an explicit pivot from survival to growth.

Management is now explicitly focusing on revenue expansion. It is the last chapter of the turnaround story. I’m bullish that management will make it happen.

No price deflation on renewals: margin on signed contracts close to 25%, presented as evidence against broad-based pricing pressure.

Black accounts: down from several problematic contracts to just two today; one of the two loses ~€10M/year, with a maximum remaining duration of 2 years.

Debt structure: no maturity before the 1.5L tranche (2029-2030) following the May refinancing. Debt buybacks already underway: €109M on the open market + €38M linked to the South America disposal.

4. Where the Investment Thesis Stands

The objective of this section is to confront the quarter / half against the deep dive bear and bull thesis.

4.1 Warning Signs of the Bear Thesis — None Triggered

The three signals that would have invalidated the thesis (original deep dive, “Points of failure” section):

• FCF vs. reported EBIT divergence: Not triggered — the gap is almost entirely explained by the restructuring cash-out

• Book-to-bill sustained below 90%: Not triggered — 91% official, likely understated (frameworks excluded)

• Flat or rising Genesis charges: Not triggered — declining trajectory confirmed for H2 (€127M in H1 → guidance of ~€70M max in H2)

4.2 Bull Case Prerequisites — Status

As laid out in the original deep dive, the bull case rests on fourteen independent arguments. Most haven’t moved this quarter. Below are only the ones that have evolved since the deep dive:

• Argument 2 (contract rationalization → hidden operating leverage) — confirmed: EBIT +43% on declining revenue across two consecutive half-years.

• Argument 3 (judicial shield + 2031 refinancing) — confirmed. No debt maturity before 2029-2030 following the completed refinancing steps. The existential risk this argument addressed is now closed.

• Argument 4 (counterparty-risk unlocking) — in progress, strengthening. Zero net client losses in H1 2026, a clear contrast with 2024-early 2025. North America book-to-bill at 115%. Management’s own posture shift — explicitly refocusing on the top line.

• Argument 5 (deleveraging flywheel) — in progress, strengthening. Debt buybacks already executed (€109M open market + €38M from the South America disposal), and oversubscription of the refinancing from the bond market in May. I see it as the main point playing out.

  

• Arguments 7-8 (AI doesn’t destroy the IT services model / regulated sectors need Atos) — qualitatively reinforced. New contracts with sovereignty-sensitive institutional clients (European public sector agency, Dutch police, European Patent Office) support the thesis, though this remains anecdotal.

5. Valuation

The following images provide the different scenarii embedded in the deep dive

The deep dive laid out a nine-scenario scorecard, I don’t provide a unique valuation, merely scenarii . The market is still pricing Atos close to S2, the market-crash scenario, at a current price of ~€33 against an S2 target of €28. That’s roughly where the implied EV/EBIT of ~5.6-6x has sat since the deep dive was written — essentially no re-rating despite the quarter’s execution.

Out of this quarter, the price should reasonably start moving to reflect:

• margin convergence over two consecutive half-years

• the book-to-bill once framework agreements are counted

• black-account stabilization (down to two, from several)

• the absence of net client losses in H1 2026

• the improving commercial pipeline (+€1.7bn qualified pipeline over H1)

• any AI/sovereign optionality (MogwAI, Sovereign Cloud, AI Studios)

The rerating will certainly take place once management increases the guidance and when the market realizes that IT services won’t be disrupted by AI any time soon. It is a core assumption of the thesis that should be emphasized.

Subscribe now

Important Disclosure & Disclaimer

All content published by JB Peter on this platform is strictly for educational and informational purposes. It does not constitute investment, financial, legal, or tax advice, nor does it represent a personal recommendation or solicitation to buy or sell securities. This research is operated by ORIACON (SASU) and reflects independent corporate analysis. Every reader must conduct their own independent research (Due Diligence) or consult a licensed professional before making any financial decision, as financial markets involve a high risk of capital loss. At the time of writing, ORIACON or the author HOLD shares in the company analyzed in this article. Following this publication, ORIACON and the author reserve the right to buy, sell, or modify positions in any security mentioned at any time, without prior notice to readers or subscribers.

1. Quarterly Verdict

1.1 The Engines Are Running, The Bill Is Getting Heavier

Nothing in the published numbers contradicts Alphabet’s growth trajectory. Search is up 17%, YouTube up 13%, and above all Cloud is accelerating to 82% with an operating margin that has nearly doubled year over year (20.7% → 35.6%). Twelfth consecutive quarter of double-digit growth for the group.

What has changed, on the other hand, is the scale of the investment effort management is now committing to. Leadership raised its annual capital expenditure guidance from $180-190B to $195-205B — a jump of more than $15B decided mid-year, justified by a faster pace of AI capacity delivery than previously planned, in the face of demand that still outstrips supply. The immediate consequence: negative free cash flow of -$5.9B for the quarter, a first in several years.

It’s this last point, more than growth itself, that drove the stock down after the release. The market read “cash pressure” where management is presenting a calculated bet on the future return of this investment, notably through TPU system sales, the bulk of which won’t be recognized as revenue until 2027.

Bottom line. Nothing in this quarter undermines the underlying thesis on Alphabet, but the market remains hesitant about the profitability of the investments being made. How long is it willing to fund this phase before demanding concrete proof of return? Q3, with the first effects of the shift toward third-party capacity, will give a first indication.

Subscribe now

1.2 My Reaction With a 10% Alphabet Portfolio

I remain very bullish on Alphabet given the quality of their world-class products, and their impressive growth given their size. Search, YouTube and Cloud are each, taken individually, dominant positions in their category — few companies can claim three engines of this scale at once, let alone growing them simultaneously at this pace (17%, 13%, 82%) with a revenue base already above $400B annualized. It’s this combination — product quality, addressable market size, and speed of growth despite the size already reached — that in my view justifies staying invested despite short-term volatility in the stock.

The TPU business looks like a solid growth driver. The first deliveries started being recognized this quarter, but remain deliberately marginal: management is clear that the bulk of revenue from these contracts won’t arrive until 2027, which gives visibility into a growth engine that hasn’t even started running at full capacity yet. Management frames this as an expansion of the addressable market rather than cannibalization of Cloud, which is consistent with the “full stack” strategy: the more layers Alphabet controls (chips, models, infrastructure, distribution), the more value it captures at each layer.

Investing in hyperscalers seemingly requires trusting management on the profitability of these investments — which has held true so far. As a minority shareholder, you can’t audit the return on every dollar of capex poured into a data center yourself; you can only watch whether the indirect signals point the right way (Cloud margin expanding, backlog growing, a consistent narrative quarter after quarter) and note that, so far, these investments have translated into margin expansion rather than dilution. That’s no guarantee going forward, but it’s a track record that deserves credit rather than default suspicion.

If profitability were to decline, they can halt investments, sell off assets, and the income statement and cash flows would mechanically improve. That’s a safety net you don’t necessarily get with other types of businesses: Alphabet’s capex is largely discretionary and spread over time, without being locked in irreversibly for decades. If AI capacity demand were to soften, management could slow server orders, pause new data center builds, or even offload assets to third parties (as it’s already partly doing with leased capacity) — and free cash flow, negative today, would turn positive again fairly quickly.

2. This Quarter’s Numbers

2.1 At a Glance

This table summarizes the key metrics from Alphabet’s Q2 FY2026 earnings release: https://s206.q4cdn.com/479360582/files/doc_financials/2026/q2/2026q2-alphabet-earnings-release.pdf

2.2 Commentary

Revenue. At $119.8B, the quarter beats consensus by nearly 3%. The 24% growth marks a clear acceleration versus Q2 2025 (+14%) — no sign of fatigue in the core advertising business.

Off-balance-sheet purchase commitments and other contractual obligations — $332.4B total (of which $138.0B short-term). This figure comes from the Q1 2026 10-Q — the Q2 10-Q hasn’t been filed yet, so this is included here as a tracking reference rather than a Q2 data point. It’s an important number to flag: it sits entirely off the balance sheet (these are future commitments, not booked liabilities), yet it dwarfs the quarterly capex figures dominating the headlines and gives a sense of just how much additional spending is already contractually locked in beyond what shows up in the P&L or cash flow statement each quarter.

— of which long-term supply agreements (>1 year), including energy and content licenses: $232.7B. Also from the Q1 10-Q. This is the bulk of the $332.4B total and covers technical infrastructure and inventory components, energy take-or-pay contracts (with payments running through 2047), and content licensing — in other words, the multi-year backbone of Alphabet’s AI buildout that sits alongside, and arguably dwarfs, the headline capex guidance figures.

Cloud, the main engine. This is the segment driving most of the positive surprise: revenue up 82%, operating margin jumping from 20.7% to 35.6%, backlog up more than $50B in the quarter alone. The first TPU system sales to customers have started being recognized, but remain marginal — management explicitly states that the majority of revenue tied to these contracts won’t show up until 2027.

Operating margin. At 34.0% versus 32.4% a year earlier, it improves year over year despite rising depreciation charges linked to capex. Viewed sequentially, the picture is different: margin declines versus Q1 2026’s 36.1%. Part of this decline is seasonal — Alphabet structurally posts a higher margin in Q1 than in Q2, as it already did in 2025 (-1.5 pt on the same transition). But the scale of the decline this year (-2.1 pt) exceeds this pure base effect: other cost of revenues (technical depreciation, YouTube content) is rising as a share of revenue, a structural effect tied to the ramp-up in capex, compounded by one-off G&A charges (compensation, litigation) that inflate that line by +50% for the quarter. Management also warns of additional pressure in Q3, tied to increased use of third-party leased compute capacity while internal infrastructure ramps up.

Free cash flow, the sensitive point. Negative for the quarter for the first time in a long while, entirely attributable to the capex level. On a trailing-twelve-month basis it remains positive at $53.3B but is down 20% year over year. This line, more than the income statement, is what dominated analyst attention on the call.

Financing — the opposite of a buyback. Rather than reducing its share count, Alphabet did the opposite this quarter: raised $49.6B in equity and convertible notes, made no repurchases (versus $13.2B a year earlier), and diluted shares outstanding rose from 12,198M to 12,309M year over year. That’s a clean reversal from the stock’s usual dynamic, entirely dedicated to funding capex.

Quality of net income. Operating income gives a faithful picture of the quarter’s real performance; reported net income, on the other hand, is massively inflated by an accounting revaluation of equity stakes, largely illiquid and unverifiable by the market.

Operating income — the part reflecting the company’s actual business — grew 30% to $40.8B. That’s a solid signal, consistent with revenue growth (24%) and Cloud margin expansion: nothing to flag there.

The portion of income coming from gains on equity securities, on the other hand, is a largely accounting artifact disconnected from operations: $98.0B in unrealized gains, which explain most of the gap between operating income (+30%) and reported net income (+298%). The detail available for Q1 2026 (the Q2 release only publishes the aggregate total) shows that this type of gain comes almost entirely from non-marketable securities — stakes in private companies, valued not at market price but revalued only occasionally upon observable transactions (typically new funding rounds). In Q1, the net gain on non-marketable securities was $36.3B versus just $105M on marketable securities — a ratio of more than 300 to 1. Of the $101.3B carrying value of non-marketable securities as of March 31, 2026, $73.6B had been remeasured during that single quarter.

For Q2, Alphabet does not break down the composition of its $98.0B in securities gains, but financial media (Fortune, among others) attribute it to two stakes in particular: Anthropic, which remained private, whose valuation rose from $350B to $965B over the quarter following a $65B funding round, and SpaceX, which went public in early June 2026 at a $1.77T valuation (versus roughly $400B at its last private valuation), and is therefore now sitting on the marketable side. This specific attribution is not confirmed by Alphabet itself — the 10-Q never names its stakes — but it’s consistent with the pattern observed in Q1: most of the gain comes from a private-stake revaluation, not a market move on listed shares.

3. Earnings Call Key Takeaways

Opening / growth. Twelfth consecutive quarter of double-digit growth; Pichai says Alphabet is still in the “very early innings” of an AI-driven secular shift. Cloud backlog at $514B (+$50B for the quarter), Gemini Enterprise adopted by ~90% of the Fortune 100. Model APIs now processing 22 billion tokens per minute (up from 16B last quarter), Gemini app at 950 million MAU.

Models / product roadmap. Gemini 3.6 Flash and 3.5 Flash-Lite launched just ahead of the call; Gemini 3.5 Pro in testing. Gemini 4 is in training — described as the “most ambitious effort yet,” needed to stay at the frontier. Management is targeting a near-monthly release cadence going forward. Coding/agentic coding flagged as a current weak spot, being actively closed (3.6 Flash: +10 points on DeepSuite vs. 3.5 Flash).

TPUs (the most-probed topic in Q&A). Allocation priority: frontier model development (AGI) first, then Search/YouTube/Cloud, particularly Vertex AI and Gemini Enterprise. Third-party capacity deals (e.g. SpaceX, Blackstone) act as a pressure valve to absorb demand without diverting internal TPUs. The CFO explicitly declined to disclose TPU sale margins specifically — only saying it “expands the TAM” rather than cannibalizing Cloud. Revenue from TPU sales to customer data centers will mostly land in 2027, not 2026.

Capex and financial structure. The guidance increase (to $195-205B) is framed as an acceleration in capacity delivery, not a change in philosophy. Funding waterfall: operating cash flow first, then debt (up from ~$16B to ~$100B in a year), then equity — the recent equity raise isn’t expected to repeat (aside from the ATM program for SBC-related taxes). Anat Ashkenazi: “we will continue to invest as long as we see an attractive return.”

Competition / moat. Pichai emphasizes the “full stack” approach (chips + models + cloud + distribution) as the differentiator rather than any single model. On the “model war” question, explicit positioning across the full price spectrum (Flash-Lite through frontier), not just the high end.

Other Bets. Waymo launched a new vehicle (”Oasis”); Pichai stayed evasive on a possible spin-off, prioritizing execution instead. Isomorphic Labs raised over $2B for its drug discovery pipeline.

Search/YouTube monetization (Philipp Schindler). Retail and finance were the largest contributors to Search growth. AI Max (out of beta): +15% average conversions for advertisers who adopt it. YouTube: 1.7B unique World Cup viewers, new “Buy with Google Pay” format on CTV.

Overall tone of the analyst Q&A: nearly every exchange came back to capex, TPUs, and ROI — relatively few questions on advertising or core Search, a sign that capex/ROI is what the market is trying to understand right now.

4. Where the Investment Thesis Stands

This kind of analysis is usually built on a full deep dive — easy to sketch in broad strokes, but very time-consuming to do properly (segment-level modeling, scenarios, precise invalidation criteria). I haven’t yet had the time to do one for Alphabet. In the meantime, here’s the thesis in summary.

I opened the position in May 2025, at a time when Alphabet, despite world-class products, was trading in “Value” territory — the market was pricing the risk of disruption from generative AI (ChatGPT and the like threatening Search) more than the underlying quality of the business. The original idea: given its financial resources and the quality of its teams, Alphabet remained one of tech’s juggernauts, capable both of reclaiming AI market share and of staying the leader in Search. The quarters since then, and this one in particular, have leaned in that direction — but absent a formalized deep dive, the points below remain a coarse-grained assessment rather than a rigorous checklist.

4.1 Warning Signs — None Triggered

• Structural slowdown of Search below 10% growth → not observed (17%)

• Stagnation or decline in Cloud margin → not observed (35.6%, expanding sharply)

• Slowdown in Cloud backlog → not observed (+$50B for the quarter)

4.2 Bull Case Prerequisites — Status

• Search and YouTube resilience against external generative AI competition → confirmed, record engagement during the World Cup

• Google Cloud market share gains → confirmed, growth well above the sector’s

• Ramp-up of proprietary TPU monetization → in progress, effect expected mainly in 2027

• Return on capex as the business scales → to be watched closely, this is precisely what the guidance increase raises

4.3 What Changes This Quarter

The new element isn’t the strategy — the integrated chips/models/cloud approach was already known — but its pace of execution. By raising its capex guidance by more than $15B in one go, Alphabet is accepting negative free cash flow in the short term to secure capacity faster, in the face of demand it describes as durably exceeding supply. The coherence of the strategy isn’t in question; it’s the speed of execution and its immediate cash cost that are the variable to watch. In short, Alphabet is borrowing against its future cash flows to grow its present.

5. Valuation: A Gap to Watch Rather Than Resolve

The stock closed the regular session at $341.91 before giving up ground after hours to around $327. This move responds to a specific announcement — the capex increase — not to a reassessment of the quarter’s commercial momentum. This type of reaction, where the market penalizes a spending commitment before seeing its results, is typical during heavy investment phases and doesn’t prejudge the outcome.

Over ten years, Alphabet’s PE has generally ranged between 17 and 33, depending on investor sentiment through the cycle. At the time of writing, the PE sits around 26 — nothing particular to flag at this level, neither expensive nor cheap relative to history. Given the quality of the business, I’d be tempted to sell if the PE were to climb above 40.

The quarter confirms the operational trajectory. The market, for its part, chose to focus on the cost of the strategy rather than on results already visible. The next concrete checkpoint will be the materialization, over the course of 2026 and 2027, of revenue tied to the Cloud backlog and already-signed TPU contracts.

Subscribe now

Important Disclosure & Disclaimer

All content published by JB Peter on this platform is strictly for educational and informational purposes. It does not constitute investment, financial, legal, or tax advice, nor does it represent a personal recommendation or solicitation to buy or sell securities. This research is operated by ORIACON (SASU) and reflects independent corporate analysis. Every reader must conduct their own independent research (Due Diligence) or consult a licensed professional before making any financial decision, as financial markets involve a high risk of capital loss. At the time of writing, ORIACON or the author HOLD shares in the company analyzed in this article. Following this publication, ORIACON and the author reserve the right to buy, sell, or modify positions in any security mentioned at any time, without prior notice to readers or subscribers.

Subscribe now

1. What the Nebius release says

Nebius ($NBIS) is launching a new model: “infrastructure partners.” The split of responsibilities is clear-cut.

The partner finances and owns 100% of the infrastructure — the building, the GPUs, the hardware. It sources powered land, handles construction (mechanical, electrical, plumbing), imports the hardware (customs included), operates the site, secures it physically, and obtains regulatory and export approvals.

Nebius provides the reference design (NVIDIA-validated, from HGX H200 to GB300 NVL72), supply-chain access, a certified assembly partner, deploys and validates the full cloud platform, keeps the SLA, and sells the partner’s capacity to its own global customer base.

The partner owns the hardware. Nebius owns the software running on top of it and the customer relationship.

The precise economics of the deal (revenue share, licensing fees, commissions) are only disclosed under NDA, after an initial qualification call. No partner has been named publicly — the release refers to “initial arrangements” already signed, without detail on who or how much.

Here is the link: https://nebius.com/infrastructure-partners

2. Interpretation

Nebius continues its massive direct capex in parallel — agreements with Meta and Microsoft, a commitment with NVIDIA to deploy more than 5 gigawatts of capacity by 2030. The partner model adds a new growth driver on top of that trajectory.

The underlying logic: Nebius saw hardware commoditization coming before the rest of the sector. Any player with capital and access to the power grid can now build a GPU facility. What doesn’t commoditize is the customer contract, the brand, the software layer, and the SLA that commits to a consistent quality of service. Nebius concentrates its resources on that layer, and delegates land, power, and local regulation to partners better positioned to handle them market by market.

The useful image here is the car manufacturer: it doesn’t make its own tires or seats, but owns the brand and the design, remains accountable for the finished product’s quality to the customer, and has its parts built by suppliers — without diluting its ability to guarantee the product. Nebius’s software layer plays that role: it ensures a customer gets “a single, consistent service” whether running on Nebius’s own infrastructure or a partner’s.

The key question for judging how solid the model is comes down to the quality of that software. If it’s genuinely better than what AWS, Azure, or GCP can offer in this segment, substitution by hyperscalers becomes difficult: they can copy the franchise concept, but they can’t instantly match a software stack built to run on heterogeneous, third-party hardware. That’s where the real defensibility of the model lies — not in the contractual structure itself, which is copyable, but in the underlying technical edge.

Each partner also brings a sector-specific advantage tied to its local position: compute stays close to the powered land and the regional demand the partner understands better than Nebius does. That proximity advantage — grid access, local relationships, regulatory knowledge of the market — is something Nebius couldn’t efficiently replicate on its own in every jurisdiction.

In a commoditized market, value concentrates on what stays scarce: the contract, the brand, and — if the technical edge is real — the software.

3. Opportunities and risks

Opportunities

• Capacity growth without weighing down Nebius’s balance sheet — compute demand already exceeds supply by their own account, so every gigawatt of incremental partner capacity finds a near-immediate outlet.

• Structurally higher margins than conventional wholesale bare-metal contracts, per the CEO’s own terms — to be confirmed once the first deals become public.

• A lock-in effect that builds fast: the more partners run on the Nebius design, the more costly switching platforms becomes for them (operational and software lock-in).

• Access to regional markets (AI sovereignty, local capital) without Nebius having to manage local regulation itself.

• If the software edge is real, the barrier to entry for hyperscalers is higher than it looks — they would need to build a stack designed for third-party hardware, not simply adapt their existing proprietary infrastructure.

Risks

• Zero named partners so far, zero verifiable margin figures beyond the CEO’s own statements. The thesis rests entirely on execution yet to be proven.

• Dependency on a given partner: local powered-land access is often hard to replace if a partner fails to deliver.

• Regulatory and export-control risk on GPUs: the partner bears this contractually, but a failure here reflects back on Nebius’s reputation through the unified SLA it guarantees to the end customer.

• The software edge has to stay real over time. If Nebius’s stack only has a limited head start, hyperscalers will eventually replicate an equivalent offering with their own capital and sales network.

• The scarcity paradox: by accelerating the arrival of new capacity on the market through third parties, Nebius contributes to closing the compute shortage that currently supports the sector’s high prices — itself included.

The metric to watch: customer contract renewal rates, and Nebius’s ability to keep its software layer’s price stable even if spot GPU prices collapse. If both hold, the “contract + brand + software” thesis is empirically validated. If customers start negotiating directly with partners around Nebius, the software edge was overstated.

Subscribe now

Important Disclosure & Disclaimer:

All content published by JB Peter on this platform is strictly for educational and informational purposes. It does not constitute investment, financial, legal, or tax advice, nor does it represent a personal recommendation or solicitation to buy or sell securities. This research is operated by ORIACON (SASU) and reflects independent corporate analysis. Every reader must conduct their own independent research (Due Diligence) or consult a licensed professional before making any financial decision, as financial markets involve a high risk of capital loss. At the time of writing, ORIACON or the author HOLD shares in the company analyzed in this article. Following this publication, ORIACON and the author reserve the right to buy, sell, or modify positions in any security mentioned at any time, without prior notice to readers or subscribers.

Faced with a mismatch between supply and demand, the adjustment always happens through one of two levers: volume, or price. When available volume can’t keep up — because production capacity is structurally limited, or because it takes years to build — price is what absorbs the imbalance. It’s the same mechanism as oil when available quantities tighten, or any other supply squeeze: a rise in demand follows exactly the same logic as a drop in supply, the effect on price is identical. That’s the mechanism that drove up the price of memory in the previous article, and it’s the one we’re trying to spot here in the reducer and linear-motion space.

The trigger is never the product itself: it’s the change in status, from a commodity bought at the best price to a strategic asset whose access gets secured at almost any reasonable price. The question this article asks is where HDS and THK stand on that path — and the answer, as we’ll see, isn’t the same for both.

1. The short version

Two Japanese companies each carry the flag for their segment: Harmonic Drive Systems (HDS) for the reducer, THK for linear motion. My valuation view on both is simple to state, even though it’s hard to pin down precisely — there’s no consensus on which assumptions to use, and I’m setting some myself for lack of anything better: broadly, these companies are already too expensive in the absence of a bottleneck, and undervalued in the presence of one. As I am very bullish on the robotics and the probability of a bottleneck on these segments, I’m staring a position on HDS at 2% of my portfolio and 1% on THK.

Subscribe now

2. Why I think there is a bottleneck in these segments

I came across a table from Goldman Sachs listing the possible bottleneck segments in humanoid robotics. I’ve reclassified it into three tiers below: real scarcity, real leverage but fragmented, or essential with no humanoid-specific alpha.

The ten categories from the Goldman table reclassified into three tiers: real scarcity, real but fragmented leverage, or essential with no humanoid-specific alpha.

Three items in the left-hand column deserve a fuller explanation, because the whole article rests on them.

The harmonic reduction gear is the part that converts a motor’s fast, weak rotation into slow, precise, high-torque motion — without the mechanical play (”backlash”) found in a conventional gear. In practice, a motor on its own spins far too fast and pushes far too weakly to move a robotic arm; the harmonic reducer does this conversion work inside the joint itself, in a compact housing. Its manufacture relies on a thin special-steel part (the “flexspline”) that flexes millions of times without fatiguing — a metallurgical know-how that’s hard to replicate quickly, which explains the scarcity of suppliers.

The planetary roller screw does the same job as the harmonic reducer, but for linear rather than rotary joints: it converts rotation into straight-line displacement, with load capacity and precision far superior to a standard ball screw. It’s the part typically found in the hips or knees of a humanoid robot, where you need to push hard in a straight line rather than pivot a joint.

The dexterous hand module brings together all the miniature motors, sensors and transmission mechanisms that let a robotic hand grip and manipulate objects with the finesse of a human hand. It’s a concentrate of every other component in the body (motors, reducers, sensors) miniaturized down to finger scale — which mechanically makes it the most complex and most expensive part of the robot.

3. The central debate: bull vs bear

The bear case common to both names

• Price competition from China (for HDS) and from China/Taiwan via Hiwin (for THK) — an established fact for both companies, not a deduction.

• Both stocks have already run up a lot: HDS is up nearly 3x in a year (roughly +200%), THK has doubled (roughly +100%, including +79% since the start of the year alone). Part of the margin of safety that existed a year ago has therefore already been consumed — entering today isn’t entering at the same price as whoever identified the thesis in 2025.

• Strong cyclicality in both businesses, each dependent on semiconductor capex and industrial-automation cycles — not a guaranteed linear growth path.

The bull case common to both

• A real demographic tailwind: the labor shortage is a structural driver of automation demand for both companies — not just a research-desk estimate.

• Operational recovery already visible in recent numbers, not just in the narrative: HDS orders +16.2%, THK’s cost-to-sales ratio down 4.2 points in Q1 2026.

• Real barriers to entry on both sides: long qualification cycles at robotics OEMs once a component is “designed-in,” and metallurgical know-how that’s hard to replicate quickly.

• The fragmentation of the world into blocs: each bloc (Western, and within the West, the EU, the US and Japan taken separately) will want to keep strategic suppliers tied to its own bloc rather than depend entirely on an outside supplier. This isn’t just a price or volume argument: it’s a supply-security logic that structurally leaves room for a “trusted” bloc supplier — even at a higher price, even with less volume than the cheapest Chinese option. Both HDS and THK benefit from this mechanism, independent of their pure price competitiveness.

• Momentum itself is a bull argument, not just a caution flag: the rally already realized in both stocks draws attention, flows, and media/analyst coverage, which tends to be self-reinforcing over the short and medium term. A stock that’s rallying hard attracts the next marginal buyer more easily than one that’s flat — that’s no guarantee of continuation, but it’s a genuine technical factor in favor of it continuing, to be weighed against the already-consumed margin of safety noted above.

Companies ruled out

a. China

LeaderDrive (Suzhou Green Harmonic, 688017.SH) is clearly a serious competitor to HDS on the harmonic reducer, at roughly half HDS’s product cost, and is gaining ground quickly (its share of Chinese robotics production has gone from near-zero in 2018 to over 30-35% in 2024-2025). I’m not investing in this segment for geographic reasons already laid out at the start of the series (Chinese A-shares, access, governance). That choice has a cost I’d rather name than stay silent about: it’s possible that price competition turns out to be tougher on the Chinese side than this thesis captures, if LeaderDrive and others keep gaining ground at the same pace.

b. Nabtesco

This is a company where the reducer (Component Solutions segment) accounts for only about a quarter of its revenue (~26% of FY2025 revenue, ¥307.9bn).

c. Schaeffler

A more classic case of diversification: the bulk of its revenue comes from automotive (bearings and transmission systems for combustion and electric vehicles), and its humanoid exposure is buried inside a much larger whole.

4. Harmonic Drive Systems

4.1 What the company does

HDS makes and sells precision reducers and mechatronic products, split into two families: harmonic reducers (77.8% of FY2026 revenue, ¥46.3bn, +9.5% YoY) — the “strain wave gearing” technology the company invented — and mechatronic products (22.2% of revenue, ¥13.2bn, -0.9% YoY), actuators integrating motor, reducer and sensors. Strong product concentration: 77.8% of revenue rests on a single line.

Customers: no names are broken out, and no single customer exceeds 10% of revenue in FY2026 (Nissan Motor, the only customer above 10% in FY2025 at ¥5.7bn, has fallen below that threshold) — good customer diversification. But end applications remain concentrated in three uses: industrial robots (the dominant application), semiconductor manufacturing equipment (growing fast, driven by generative AI and data centers), and automotive applications (declining) — solid customer diversification, more limited diversification by end use. The real end customers are industrial and collaborative robot OEMs (Fanuc, Yaskawa, ABB, KUKA, Universal Robots and Chinese manufacturers), semiconductor equipment makers, and medical/aerospace integrators — a “designed-in” component, very hard to substitute once built into a customer’s design.

Suppliers: no names identified. Gross margin depends heavily on the cost of raw materials (special steel alloys for the flexspline, which must withstand cyclical deformation without fatigue) and on labor — material and labor costs remain elevated, with an internal cost-reduction program underway.

Competitors: Nabtesco (the most significant, but positioned on RV reducers — a competing rather than directly substitutable technology, ~60% share of that sub-segment), and Sumitomo Drive Technologies. Harmonic Drive LLC (US) is not a competitor but HDS’s own local distribution/production subsidiary — not to be double-counted in the competitive analysis. The most significant competitor is actually Chinese: LeaderDrive (Suzhou Green Harmonic, 688017.SH), which produces at roughly half HDS’s unit cost and is gaining ground quickly — its share of Chinese robotics production has reportedly gone from near-zero in 2018 to over 30-35% in 2024-2025, alongside Beijing CTKM Harmonic Drive, Zhongda Leader and Zhenkang. It’s specifically this name, not a Western competitor, that represents the real competitive risk to the thesis — but it stays outside my investable universe (Chinese A-shares, per the geographic filter set out at the start of the series). In the overall precision-reducer market (RV plus harmonic combined), HDS remains a follower; on the harmonic reducer segment specifically, HDS is the category definer of its own technology, not the leader of the total market.

Moat: decades of patents and real-world reliability data, metallurgical know-how on the flexspline that’s hard to replicate quickly, and above all a high switching cost once the component is “designed-in” to a customer’s robotic arm architecture (1-2 year qualification cycles before series integration).

Balance sheet (gearing): gross financial debt of ¥13,219m against equity of ¥80,390m, a debt-to-equity ratio of 16.4% (22.5% including lease liabilities of ¥4,846m). Debt/EBITDA of roughly 1.3x (1.8x with leases) — comfortably manageable. HDS is in a net cash position (~¥6bn excluding leases).

Geographic revenue split (FY2026, by customer location): Japan ¥24,388m (41.0%), Europe ¥16,777m (28.2%, of which Germany alone 10.1%), North America ¥12,108m (20.3%, of which the US alone 18.0%), China ¥4,047m (6.8%), other regions (Korea, Taiwan, Oceania) ¥2,239m (3.8%). 59.1% of revenue is generated outside Japan — genuine geographic diversification, with a notable dependence on Germany within Europe. Like any company that exports the majority of its revenue, HDS remains mechanically sensitive to currency swings (USD/JPY, EUR/JPY): a weak yen mechanically inflates consolidated results in yen terms regardless of actual operating performance, and vice versa.

4.2 HDS valuation

The scorecard below crosses a flat 50%/year revenue growth assumption over four years (FY2026 → FY2030) with three net-margin scenarios (Bear 5%, Central 15%, Bull 30%), the same method as in the Memory article. Useful reference point: HDS’s current net margin is 2.7% (FY2026) — below the Bear scenario (5%) itself, which shows how far there is to go even to reach the bottom of the table.

HDS is clearly too expensive in the absence of a bottleneck: under the Bear scenario (5% net margin, the scarcity thesis doesn’t pan out and Chinese competition captures pricing power), the stock is overvalued by 40-80% depending on the multiple used. It is, on the other hand, clearly a great price if the bottleneck is confirmed: under the Bull scenario (30% net margin, the harmonic reducer’s scarcity genuinely translates into pricing power), the stock is undervalued by 19% to 258%. The whole thesis comes down to that bet.

A word of context, so as not to dismiss the Bull scenario as unrealistic out of hand. In the Memory article, Micron and SanDisk genuinely saw a net-margin expansion of several dozen percentage points, driven by revenue growth well above 50% per year during the period when the memory bottleneck was biting — so the idea that a real bottleneck can produce this kind of trajectory isn’t a made-up scenario, it’s a verified precedent within this same series of articles. The asymmetry is therefore real. But Micron and SanDisk delivered that margin expansion with numbers already booked at the time of purchase, whereas HDS’s current guidance (FY2027 revenue barely +14.2%) is well short of a 50%/year pace. A bearish scenario therefore remains very much possible for HDS — the favorable asymmetry exists, but for now it’s only a possibility, not a precedent already confirmed on this stock.

4.3 Sizing

I’m keeping the series’ overall envelope: 5-10% of the total portfolio for all bottleneck bets combined, not on a single name, not all at once. I plan to add other companies in other bottleneck segments (optics, energy, power semiconductors in particular) in upcoming installments of the series — reallocation between names already in the portfolio and these newcomers can happen based on projections and how each one’s price-to-value ratio evolves, not on a fixed basis.

On Harmonic Drive Systems: I’m buying in, with 20% of the bottleneck sub-total allocation, i.e. 2% of the total portfolio. This isn’t a full position: the scorecard shows the current price only makes sense if the bottleneck is genuinely confirmed (Bull scenario), and stays frankly expensive otherwise (Bear scenario). The remaining allocation stays open for a top-up if the stock moves closer to the Bear/Central scenarios in the table, or if the China segment stabilizes over at least two consecutive quarters.

5. THK

5.1 What the company does

THK makes precision linear-motion components: the LM Guide (its founding product, invented in 1972), ball screws, linear actuators, cross roller rings, and seismic isolation/damping products as a diversification. Strong technological concentration: a single technology family (”rolling motion”) underlies the whole portfolio — a shock to that base technology would hit the entire group. Application diversification, on the other hand, is broad: machine tools, semiconductors, robotics, medical, aerospace. The automotive & transportation business (steering, suspension, and braking components) is being divested to AP87/Advantage Partners (closing expected June 1, 2026) and reclassified as a discontinued operation — the FY2026 figures (¥276bn of guided revenue) therefore cover only the continuing industrial business.

Customers: no customer named and no concentration data (top 10) available. Geographic and sector diversification mechanically limits dependence on any single customer, but this isn’t a verified figure, only a structural inference. Direct sales in Japan (plus agents) and through local subsidiaries in 5 regions (Japan, Americas, Europe, China, Other — with India/ASEAN expanding).

Suppliers: a diversified global supply base, not limited to Japan. No key supplier named. Likely critical inputs (special steel, bearing balls, precision grinding equipment) aren’t disclosed publicly — not a documented vulnerability point.

Competitors: NSK and IKO/Nippon Thompson (Japan), Bosch Rexroth and Schaeffler/INA (Germany), and above all Hiwin Technologies (Taiwan) — the low-cost competitor moving upmarket, particularly in China. No precise market share is available; as an unverified indicative point, THK and NSK are generally cited in trade press as the two largest global players in LM Guides.

Moat: precision-machining know-how built up over more than 50 years, historical patents on rolling technology, long OEM qualification cycles (a “designed-in” component, hard to dislodge once specified into a machine’s design), a global service and application-engineering network, economies of scale. Low-end competitive pressure is intensifying, driven by the rise of Chinese and other emerging-market products — real barriers, but not absolute ones, particularly against Hiwin and Chinese entrants.

Balance sheet (gearing): a debt-to-equity ratio of roughly 31% — noticeably higher than HDS’s (16.4%, 22.5% with leases), but still at a moderate level for an industrial company, no warning sign in itself. Consistent with other observations: short-term borrowings doubled in Q1 FY2026 (¥33bn → ¥60bn), partly to fund the buyback and the automotive divestiture; confirmed credit lines of ¥50bn are in place; cash stayed positive throughout the 2021-2025 cycle (¥120-160bn) and operating cash flow stayed positive even at the low point of the cycle (¥15.6bn in 2021). The point to watch is therefore not the absolute level of leverage, but its recent trajectory (short-term debt doubling in a single quarter) against a backdrop of already-generous shareholder returns (record dividend plus a completed buyback).

Geographic revenue split (FY2025, consolidated basis before the automotive reclassification): Japan ¥110.9bn (30%), Americas ¥90.2bn (25%), China ¥76.0bn (21%), Europe ¥67.5bn (18%), Other ¥21.6bn (6%) — a similar split in Q1 2026. Note that this breakdown covers the historical total business (automotive included); it isn’t directly comparable to the guided FY2026 revenue of ¥276bn, which excludes the divested automotive unit. Like HDS, THK remains mechanically sensitive to currency moves across its four export regions — a weak yen inflates consolidated results in yen terms regardless of actual operating performance.

5.2 Valuation

Same assumptions as for HDS, to keep things comparable: FY2030 horizon (4 years), a flat 50%/year revenue growth rate, and the same three net-margin scenarios (Bear 5%, Central 15%, Bull 30%) applied to guided FY2026 revenue for the continuing business (¥276bn) and to 112.02 million shares outstanding (derived from guided FY2026 EPS of ¥202.64 and guided net income of ¥22.7bn). Useful reference point: THK’s current net margin comes out at roughly 17% TTM — already above the Central scenario (15%) and close to the Bull scenario (30%), the reverse situation from HDS.

Like HDS, THK is too expensive in the absence of a bottleneck and a great price if the bottleneck is confirmed — but the tipping point is different: even under the Bear scenario (5% net margin), the stock is already undervalued by 23% to 143% depending on the multiple. That mostly reflects the smaller size of the revenue base used (¥276bn, smaller than HDS’s) relative to the same 50%/year growth assumption — a signal that should be read with caution, see the caveat below.

An important methodological caveat, even more so than for HDS: looking at the past, THK’s revenue growth runs more in the range of 10-15%/year over a full cycle (consistent with its own FY2026 guidance of +14.8%), not a sustained 50%/year — over 2021-2025, its revenue went from ¥318bn to ¥223bn depending on the machine-tool/semiconductor capex cycle phase, a cyclical pattern rather than a structural trend at that pace. For the 50%/year assumption to become coherent, it would take a genuine bottleneck effect combining both a price effect (pricing power regained on linear motion) and a volume effect (humanoid demand adding to the existing industrial cycle rather than replacing it) — not just a normal cyclical rebound in semiconductor capex. The table above should therefore be read as a methodological comparability exercise with HDS, not as a reference projection for THK: on a growth assumption closer to its own history (12-15%/year), the current price would remain considerably more stretched, as the earlier calculation showed (forward PE of ~38x versus a historical 15-18x).

5.3 Sizing

The name has cleared the quality bar (a disciplined refocusing via the automotive divestiture, sound governance, aligned incentives, real tailwinds) but the price question remains entirely dependent on which growth assumption is used — on the shared 50%/year assumption, the valuation looks cheap; on a realistic assumption based on its own history (12-15%/year), it stays stretched (forward PE ~38x versus a historical 15-18x).

On THK: I’m buying in at 1% of the total portfolio (when the momentum will get better) — a more cautious size than HDS (2%), consistent with the gap between the two growth readings still being too wide to fully resolve. Adding to the position will depend on confirmation, over two consecutive quarters, that growth is moving closer to the humanoid/AI pace rather than the historical semiconductor/machine-tool pace; failing that, a price pullback toward a PE near 15-18x (THK’s normal historical range) or toward its 200-week moving average would be the alternative trigger.

Subscribe now

Sources

I have no formal obligation to disclose my sources, but I want to express that my articles are inspired by Ren’s articles and ideas, which I strongly recommend: https://substack.com/@renstocks

As well as Micron’s and SanDisk’s 10-Q and 10-K filings.

Important Disclosure & Disclaimer

All content published by JB Peter on this platform is strictly for educational and informational purposes. It does not constitute investment, financial, legal, or tax advice, nor does it represent a personal recommendation or solicitation to buy or sell securities. This research is operated by ORIACON (SASU) and reflects independent corporate analysis. Every reader must conduct their own independent research (Due Diligence) or consult a licensed professional before making any financial decision, as financial markets involve a high risk of capital loss. At the time of writing, ORIACON or the author HOLD shares in the companies analyzed in this article. Following this publication, ORIACON and the author reserve the right to buy, sell, or modify positions in any security mentioned at any time, without prior notice to readers or subscribers

The first in the series is about memory, with company proposals tied to this trade. It builds on elements already present in my portfolio, based on supply/demand asymmetries — the same logic as in my article on Strategy Inc. (MSTR): an asset whose supply is constrained, a demand that can’t route around it, and a price that eventually stops drifting and snaps instead.

This article is structured in seven parts: first the building blocks of memory, to lay down the vocabulary; then the Nvidia supercycle pattern, transposed to memory; the state of the market today; the companies where to play this thesis; the central debate over whether the cycle is over; valuation; and finally sizing. Each part builds on the previous one.

The Short Version

Supply and demand indicators say the memory cycle isn’t over. But Micron’s and SanDisk’s prices already behave as if the peak had passed. Today, Micron trades around $948.80, down more than 24% from its mid-June high of $1,255. SanDisk trades around $1,727.18, down nearly 27% from its 52-week high of $2,354, with a good chunk of that lost in just a few sessions. Both stocks remain extremely volatile day to day — these levels will move fast, and that’s not the point.

The point: it’s this gap between price and fundamentals that interests me — reinforced by an almost mechanical effect: with gross margins of 78-85%, as long as prices hold, every extra dollar of revenue falls almost entirely to the bottom line.

Subscribe now

The building blocks of memory: what it’s for, and why AI can’t do without it

Before going further, the definitions need to be laid out. Memory isn’t a single block — it’s several families of products, each playing a different role in an AI machine. The most useful distinction to keep in mind is between volatile memory (which loses its content without power) and non-volatile memory (which retains its data).

The premise of this section: being able to understand the mechanics at play in broad strokes, with even a very limited technical interest. Judging the precise relative value between products requires being a specialist in the field — that’s not what’s being asked here. The goal is to understand enough vocabulary to follow the rest of the article, not to become a memory engineer.

Volatile memory

Dynamic RAM (DRAM) is fast, volatile memory: it loses all of its content the moment power is cut. It’s the working memory used by data centers, PCs, mobile devices, and cars. It comes in several flavors depending on use: DDR5 is the workhorse of standard servers and PCs; LPDDR is the low-power variant that powers every smartphone, and increasingly AI servers where energy efficiency matters at rack scale; GDDR is the graphics variant, soldered next to GPUs in gaming cards and certain inference accelerators.

HBM (High-Bandwidth Memory) is the flagship DRAM product for AI: a 3D-stacked architecture, directly fused onto the processor. It’s the piece that answers the real problem of AI compute: a GPU starved of data to process is the most expensive and most useless machine in the data center. Compute is worthless if memory can’t feed the processor fast enough — that’s exactly HBM’s job. It’s also why only three companies in the world are capable of making it.

Non-volatile memory

NAND is rewritable non-volatile storage — the kind that keeps data even without power (SSDs, memory cards, USB drives). Historically, its role in AI was secondary to DRAM’s. That’s changing: recent inference architectures now rely on large NAND pools to offload the “KV-cache” (the context memory a model holds onto mid-generation) and store long context — a use case that barely existed a year ago and could, on its own, add substantial demand by 2027-2028.

NOR handles reliable code storage (system boot-up, embedded automotive/industrial applications) — a more minor segment, less central to this thesis.

HBF (High Bandwidth Flash) aims to combine flash density with DRAM-like bandwidth — co-developed by SanDisk and SK Hynix. Important point to keep in mind: it’s still at the design and prototype stage, not commercial production. None of this is in current revenue or in analyst models yet — it’s optionality, not a given.

One idea to take away from this section: memory isn’t a simple accessory to AI compute, it’s what decides whether compute can actually run at full capacity. That’s why it deserves its own place in a series on bottlenecks.

---

1. The Nvidia supercycle pattern

Three years ago, Nvidia followed a precise pattern. It wasn’t just “the stock went up.” It’s two engines running at the same time:

• Earnings explode, because AI compute demand far exceeds what the industry can produce.

• The multiple re-rates, because the market changes its reading grid: Nvidia is no longer a cyclical graphics card maker, it has become the mandatory infrastructure of an entire industry.

Two engines running together doesn’t produce an additive gain. It produces a squared gain: earnings ×, multiple × —> price ײ.

The trigger for the re-rating is never the product itself. It’s the change in status: from a commodity you buy at the best price, to a strategic asset whose access you secure at almost any reasonable price. That’s the status shift that has to be spotted before it gets priced.

I think we’re watching exactly this pattern replay in memory, with a two-to-three-year lag on Nvidia.

---

2. The state of the memory market today

Memory has become AI’s new wall. The “memory tax” — memory’s share of AI capex — now represents roughly a third of the total, and that share is growing.

What’s really changed is long-term contracts (LTAs / NBMs depending on the company). Historically, memory sold like any commodity: produce, stockpile, sell on a spot market and short contracts that adjusted violently. LTAs change the mechanics: multi-year agreements, prepayments, committed volumes, take-or-pay penalties. It increasingly looks like what a semiconductor foundry does — not what a commodity seller does.

The bulls’ argument: if memory pulls off the same business-model transition as foundries or hard drives after 2011, it deserves the same valuation transition.

One nuance worth keeping in mind: these contracts dampen cyclicality, they don’t eliminate it. A precedent exists and it stings: in 2017, similar commitments held exactly until they were tested by a real slowdown, then got renegotiated back toward spot prices within a few quarters.

The difference with 2017 is the demand engine. In 2017, the tension came from a classic smartphone/server upgrade cycle. Today, it’s generative AI that needs unprecedented compute power — and therefore memory — with multi-year investment budgets committed by players who historically never had to haggle over memory prices. This article rests on the hypothesis that this demand engine constitutes a genuine investment supercycle, not a classic cycle peak that will unwind at its usual pace. If that hypothesis is wrong, so is the rest of the thesis.

---

3. Segments and companies: where to play the thesis

In this supercycle, not everyone is exposed the same way. Two worlds need distinguishing: HBM, on the GPU side, serving model training at Nvidia — and “classic” high-capacity memory (DRAM, NAND, and tomorrow HBF), serving inference instead.

**Micron** is the American generalist pure player — DRAM and NAND, with a clear geopolitical positioning: it’s the only one of the three big memory players that answers to the United States, with CHIPS Act-backed fabs in Idaho, New York, and Virginia. Strategic contracts signed with hyperscalers, HBM capacity sold out in advance.

**SanDisk** is the NAND/flash pure-play. It’s developing HBF, co-developed with... SK Hynix — still at the prototype stage, as noted above. But the nuance needs to stay clear: SanDisk does one thing, flash, while SK Hynix remains structurally focused on HBM for Nvidia GPUs — on the training side, not the inference-memory side that’s the core of this thesis.

**Samsung** is deliberately not covered in detail here. It’s a generalist conglomerate — memory, foundry, consumer electronics — whose memory exposure is too diluted within a much larger empire to make it a pure vehicle for this specific narrative.

One nuance not to forget: flash itself isn’t fully sheltered. HBM, CXL-attached memory, and new non-volatile architectures could eventually shrink flash’s role in AI pipelines. SanDisk is developing HBF precisely to stay ahead of that risk — but incumbents historically don’t always see disruption coming in time.

---

4. The central debate: is the memory cycle over?

This is the question that decides everything else. And the answer doesn’t depend on opinion — it depends on two things that can be observed: supply/demand fit, and price as the normalization gauge for the shortage.

**The bear camp** says: capacity is coming. New fabs and production lines are set to come online starting 2027-2028. Historically, every memory margin peak has eventually attracted the capacity that floods the market and breaks prices. The classic mechanism, once the shortage clears: prices normalize, earnings normalize along with them, and the market doesn’t discover it all at once — the P/E drifts down slowly ahead of time, as investors anticipate the end of the cycle before it even shows up in the results. 2028 is when the music stops.

**The bull camp** answers with arithmetic rather than pattern-matching: bit demand is growing in the mid-20s percent per year. The HBM conversion ratio (it takes roughly three commodity wafers to produce the bit-equivalent of one HBM wafer) means the current capex wave isn’t excessive — it’s the bare minimum required to keep the shortage from becoming absurd.

My read: the real test arrives in 2028. I don’t calculate beyond that date — as they say in chess, long variation, bad variation. If a thesis needs a chain of assumptions stretching indefinitely into the future to hold up, that’s a sign it’s already fragile. Before 2028, the thesis of supply ruining the market has no physical grip. After that date, anything is possible, and I’d rather reassess the position at that point than pretend I can calculate it today. This is a position with a limited shelf life, not a conviction held forever.

Metrics to watch ahead of earnings

The problem with this debate is the temptation to wait for the next quarter to find out who’s right. Mistake: the market gives signals beforehand, provided you look in the right place.

Signal #1, The spot/contract spread

Memory sells on two parallel markets. The contract is where most of the volume gets negotiated — quarterly, between manufacturers and large buyers. That’s the price that actually generates companies’ revenue. Spot is the marginal market — small volumes traded day-to-day between distributors and brokers. It represents only a fraction of bits sold, but it moves instantly with sentiment, while the contract adjusts with a lag.

In a shortage, spot trades at a premium above contract, and that premium pulls the contract up the following quarter. In oversupply, it’s the reverse: spot drops below contract, buyers wait, and the contract eventually falls too. The most honest turning-point signal is exactly this crossover — spot dropping back below contract while inventories build. That’s happened before every historical peak of the memory cycle.

Spot is easy to track: public indices, widely relayed by trade press, free in broad terms — tracking its trend over time is already enough as an early warning. Contract is less immediate — three ways to approach it:

1. Monthly assessments from industry pricing services, built from surveys of market participants.

2. The average selling price each manufacturer reports on earnings calls — an implicit contract price, calculated after the fact, so available only on a quarterly cadence.

3. The anchor price written into the LTAs themselves, referenced to a given quarterly average — revealed only when deal announcements are made.

Signal #2: competing capacity announcements

Fully public — press releases, earnings calls, trade press. This isn’t theoretical: it’s precisely this kind of announcement (Samsung and SK Hynix signaling plans to expand capacity) that triggered the recent correction across the entire memory sector. The real test isn’t the announcement itself, it’s whether it translates into actual volumes that loosen the supply/demand ratio.

Signal #3: the 2027-2028 production timeline.

Known through manufacturers’ guidance and investor presentations, updated every quarter. As long as these sites aren’t producing yet, the capacity announcement remains an anticipated signal, not a consumed fact.

The other signals worth tracking — wafer reallocation between HBM and commodity DRAM, actual renewal of LTA contracts, inventory levels at manufacturers — are useful for understanding the mechanism, but in practice, they only confirm on a quarterly cadence. Better to know that than to believe everything can be tracked continuously.

**This dashboard — spot/contract spread first — is what serves as the exit or position-reduction signal, not a fixed price target.** As long as these indicators don’t move, the gap between current pricing and physical fundamentals remains the core of the thesis.

---

5. Valuation scorecard: does the price already reflect an end of cycle?

The question isn’t “is it expensive or cheap” in the abstract. The question is: what cycle scenario does the current multiple imply, and does that scenario match what’s actually observed in the supply and demand data?

• Micron trades at a single-digit multiple on next-twelve-months earnings, even as it just posted record results. A multiple that low on such elevated profits means one thing: the market is betting these profits won’t last, that this is a peak and not a new floor. Set against the data from part 4 — a shortage that structurally shouldn’t clear before 2028 — this pricing looks ahead of the facts.

• SanDisk trades at a hefty premium on this year’s earnings, but that premium collapses sharply when looking at 2027, provided long-term contracts hold. This is a different setup from Micron’s: here, the market isn’t pricing pessimism about the length of the cycle, it’s pricing optimism about the duration of the contracts — with little room for error in case of disappointment.

Scorecard — Micron (MU), current price $948.80

The table below applies 50%/year revenue growth through 2028 (a quite conservative assumption IMO) across all three scenarios — only net margin varies, between a cycle trough (10%), mid-cycle (30%), and a hold near the current peak (60%). Each cell gives the implied price, the change versus the current price, and the 2-year CAGR.

How to read this: the current price already sits between Central/PE 10 and Central/PE 20. In other words, it takes neither a Bull scenario nor a generous P/E to justify today’s price — a simple hold at mid-cycle with a modestly re-rated multiple is enough. Only the Bear scenario, at any P/E, is clearly a loss. That’s a favorable risk setup: the current price doesn’t require believing in the best-case scenario to be breakeven.

Scorecard — SanDisk (SNDK), current price $1,727.18

Same growth (50%/year) and margin (10/30/60%) assumptions as for Micron, to make the comparison direct.

How to read this: here, the current price sits between Central/PE 20 and Central/PE 30 — it already takes a rich P/E applied to a mid-cycle scenario to be breakeven, and the Bear scenario stays a loss at any P/E in this table. That’s a less favorable risk setup than Micron’s: the current price leaves much less room for error before flipping into a loss.

The point linking both tables: at the 2027-2028 horizon, it’s the same parameter — the margin assumed — that decides everything, far more than the P/E applied. But the asymmetry between the two stocks is clear once the assumptions are harmonized: Micron needs fewer things to go right, SanDisk needs more things to go right just to avoid going wrong.

The argument that makes this bet interesting despite the uncertainty: the fixed-cost structure of the memory industry means that, as long as prices stay elevated, a revenue increase converts almost entirely into a net income increase, with current gross margins of 78-85%. It’s this mechanical gap — not a market opinion — that justifies taking the bet. The risk is symmetric: that same leverage works in reverse if prices turn.

My conviction: I’m not going in for now. I’m waiting for a price pullback, ideally through a genuine sector-wide sell-off, before entering — see part 6. That pullback might never come and I might miss the move. Fine by me: in that case I’d rather focus on other bottlenecks in the series, earlier in their own re-rating cycle.

6. Sizing

I plan to allocate between 5% and 10% of my portfolio to the full set of ‘bottleneck’ stocks in this series, across upcoming articles — not all at once, and not on a single name.

For Micron and SanDisk specifically, I’m passing for now. Both scorecards in part 5 show a real thesis, but both stocks have already rallied hard from their recent lows, and I’d rather wait for a genuine price pullback before entering. By “pullback,” I mean a temporary price panic — a sentiment purge like the one already documented in part 4 (sector rotation, scary capacity announcements, profit-taking after an extreme rally) — not a signal that the cycle is genuinely over. It’s precisely because nothing in the fundamentals changed during those episodes that a price plunge becomes an entry opportunity, not a reason to run.

That pullback might never come and the price might keep climbing without me. Fine by me. In that case I’d rather stay focused on other bottlenecks in the series, earlier in their own re-rating cycle, where the risk/price-paid ratio is more favorable today. Nothing stops me from coming back to Micron or SanDisk later if the price comes to me.

The dashboard from part 4 — spot/contract spread, capacity announcements, production timeline — remains the right filter for telling a genuine cycle turn apart from a simple sentiment purge, whether now or on a future entry attempt.

Conclusion

The same pattern that made Nvidia is replaying in memory, with a lag. Supply and demand data say the shortage holds at least through 2028. The price, meanwhile, already behaves as if it were all over — on Micron through pessimism about the length of the cycle, on SanDisk through a premium that leaves no room for error. Between the two, a simple mechanism: as long as prices hold, operating leverage makes earnings explode much faster than revenue. That’s the gap worth watching, with the dashboard from part 4 as the signal.

A new article is currently in the works on another bottleneck.

Subscribe now

Sources

I have no formal obligation to disclose my sources, but I want to express that my articles are inspired by Ren’s articles and ideas, which I strongly recommend: https://substack.com/@renstocks

As well as Micron’s and SanDisk’s 10-Q and 10-K filings.

Important Disclosure & Disclaimer

All content published by JB Peter on this platform is strictly for educational and informational purposes. It does not constitute investment, financial, legal, or tax advice, nor does it represent a personal recommendation or solicitation to buy or sell securities. This research is operated by ORIACON (SASU) and reflects independent corporate analysis. Every reader must conduct their own independent research (Due Diligence) or consult a licensed professional before making any financial decision, as financial markets involve a high risk of capital loss. At the time of writing, ORIACON or the author DON’T HOLD shares in the company analyzed in this article. Following this publication, ORIACON and the author reserve the right to buy, sell, or modify positions in any security mentioned at any time, without prior notice to readers or subscribers

Non of this is happening with the stock I suggest you. A perfect niche quality stock in a defensive sector, temporarily undervalued.

Plan

This article starts with the business — what Interparfums actually does, how the licensing model works, who its clients and suppliers are, and what the brand portfolio looks like today. From there, it examines whether this model constitutes a genuine competitive advantage or a contractual dependency dressed up as one, and how Interparfums sits against both the licensed players and the integrated luxury houses that dominate the category.

Financials and management follow, with particular attention to how the business absorbed a currency- and tariff-battered 2025 without breaking stride.

The article then presents the two competing theses in full, a valuation scorecard, and portfolio considerations.

I know you’re busy, so I always start with a summary called The short version. If that’s all you have time for, that’s fine. The rest of this article explains why.

The Short Version

Interparfums is a forty-year-old licensing operator trading like its growth engine broke. It didn’t. A currency move and a tariff shock cost real money in 2025 and Q1 2026, and the stock repriced as if the damage were structural — PE compressed from a historical 30-40x range to roughly 15-16x on a normalized basis, the cheapest this business has traded in years.

Nothing in the numbers supports that read. Net cash, no leverage. A dividend maintained through the worst of the slowdown. Tariff-adjusted net income actually grew in 2025, once the one-off €7.6 million tariff cost is stripped out. The Boucheron license the market priced as lost in November 2025 was renewed in February 2026 — the clearest real-time test of the renewal track record this business has run for forty years, and it passed.

My conviction is the bull case: EPS compounding at roughly 14% a year as Lacoste matures and the 2027 brand launches (Off-White™, Annick Goutal, Longchamp) start contributing, with the multiple re-rating to 25x as the market re-reads this as a temporary air pocket rather than a structural decline. That scenario puts the stock near €58 within 2.5 years — +126% from current levels, a 38.4% CAGR. It doesn’t require heroic assumptions. It requires the licensing model to keep doing what it’s done for four decades, priced today as if it might stop.

This isn’t the loudest position in the portfolio. It’s a defensive business, temporarily priced like a declining one, sitting completely outside the AI narrative that moves everything else I hold. That’s the trade.

Subscribe now

Understanding Interparfums’ business, supply chain and environment

History

Interparfums was founded in 1982 in Paris, by Philippe Benacin and Jean Madar. From the start, the structure was split in two: an American parent company, Interparfums Inc., listed on the Nasdaq in 1988, and a French subsidiary, Interparfums SA, listed in Paris starting in 1994. That dual listing is not a footnote — it gives the group two separate access points to capital markets and two distinct shareholder bases, an architecture it still carries today.

The core business has not changed in over forty years: sign licensing agreements with fashion, leather goods, or jewelry brands, and develop fragrances on their behalf. The first agreement, in 1988, was for the Régine’s brand. Others followed — Burberry (1993, terminated in 2012), S.T. Dupont, Paul Smith, Lanvin, Van Cleef & Arpels, Jimmy Choo (2009), Montblanc and Boucheron (2010), Balmain and Repetto (2011), Karl Lagerfeld (2012), Coach (2015), Kate Spade (2019), Moncler (2020), Lacoste (2022). Some licenses end — Burberry is the clearest example. Others last for decades and get renewed repeatedly, like Montblanc, extended through 2031.

Two breaks from the pure-licensing model stand out. In 2015, Interparfums acquired the Rochas brand outright — its first owned brand. Then, starting in 2024, the group accelerated acquisitions: Off-White™ in 2024, Annick Goutal in 2025, alongside a new licensing agreement signed with Longchamp. The company is still, first and foremost, a licensing operator. But it is building, in parallel, a second pillar made of brands it actually owns.

Segments of activity

Interparfums runs one integrated business with two revenue mechanisms attached to it.

The licensing model is the historical core. A luxury house grants Interparfums the right to use its name in exchange for an annual royalty indexed on sales. Interparfums then owns the entire execution chain: fragrance creation, component sourcing, manufacturing, packaging, marketing tools, and distribution — all built in close collaboration with the licensor’s own creative and marketing teams. The relationship is long-term by design: brands are chosen for their international recognition and their “readable” identity, and each is developed through a steady cadence of launches meant to build a full product range over years, not quarters.

The owned-brand model applies to Rochas and to the fashion side of the business acquired more recently (Off-White™, and the newly launched Solférino Paris collection). Here the group runs a hybrid setup: part of the activity is sub-licensed to specialized partners who produce and distribute certain categories, while another part is operated directly by Interparfums, which then earns both the product margin and the royalty stream from its own sub-licensees.

Brand portfolio

The brand portfolio spans three tiers of maturity: long-established franchises with predictable renewal cycles (Montblanc, Jimmy Choo, Coach, Lacoste, Boucheron), the owned Rochas fragrance and fashion lines, and a newer growth layer built through recent acquisitions and licenses — Off-White™, Annick Goutal, Longchamp — whose first product launches are scheduled for 2027 and beyond. Management describes 2026 as a year of roughly fifteen line extensions on existing fragrances, alongside early development work on the newly acquired brands.

Clients

Interparfums does not sell directly to the end consumer. Distribution runs through wholly-owned or joint-venture subsidiaries, independent distribution companies, subsidiaries of large cosmetics groups, and duty-free operators, reaching close to 120 countries. International sales represented 93.6% of group revenue in 2025 — a figure that has stayed remarkably stable (between 93.6% and 94.6%) over the last five years. The United States is the single most important market and the most exposed to currency swings.

Suppliers

Interparfums owns no factories. Manufacturing is entirely outsourced to a network of roughly a hundred specialized industrial partners spanning several distinct trades: the “noses” and concentrate producers who create the fragrance oils themselves, glassmakers who produce the bottles, component manufacturers responsible for caps, pumps, and metal parts, box and packaging makers, and packaging and logistics providers based in France. The group describes this as a deliberate choice — flexibility over vertical integration — reinforced by a diversified network of production sites capable of making the same product, which limits the risk of a single subcontractor failure. Logistics is centralized through a dedicated 36,000 sqm warehouse in France, supplemented by warehouses in the United States and South Korea.

Interparfums spent close to €192 million on marketing and advertising in 2025 — more than 21% of revenue.

Competition

The prestige fragrance market breaks into two structurally different groups of players, and the annual report itself draws that line explicitly.

The integrated houses — LVMH (Christian Dior, Guerlain, Givenchy, Kenzo, Bulgari), Estée Lauder, Chanel, Puig — own the brands they sell fragrances under. They control the entire narrative: the runway shows, the boutiques, the advertising campaigns, the pricing architecture. Fragrance for them is one expression of a brand they already own outright, not a licensed activity.

The licensing operators — L’Oréal, Coty, Shiseido, Euroitalia, and Interparfums — do not own most of the brands on their shelves. They rent the name, execute the fragrance, and pay a royalty back to the house that owns the identity. Within this group, L’Oréal, Coty and Shiseido operate at a completely different scale, with fragrance divisions posting several billion euros in revenue each. Interparfums sits in the second tier alongside roughly ten other mid-size players, with revenue in the €100 million to €2 billion range.

That’s the direct competitive set. Indirectly, Interparfums also competes for licenses themselves — a fashion or jewelry house choosing between Interparfums, Coty, Euroitalia or an in-house fragrance division when a contract comes up for renewal or a new brand looks for a partner. Losing that competition doesn’t cost market share on a shelf. It costs the license outright, as happened with Burberry in 2012.

Category definers. Chanel and LVMH’s owned brands — Dior, Guerlain — are the closest thing this market has to category definers. They don’t need a licensing partner, they set the pricing ceiling, and much of what “prestige fragrance” means as a category is built on their positioning. No licensing operator, including Interparfums, occupies that role. The category is defined elsewhere, and Interparfums operates underneath it.

Market share. Interparfums holds around 4% of the French selective distribution market, and between 2% and 5% in other key countries — the United States, the United Kingdom, Mexico, China. In the US specifically, the report places Interparfums as the 7th-largest player, with a 4.5% share, and its three flagship brands (Jimmy Choo, Coach, Montblanc) sit among the top 30 fragrance franchises in the market. The global selective fragrance market is estimated at roughly $40 billion. On any of these numbers, Interparfums is a meaningful but clearly secondary player — never the largest, always present.

What the report is careful to point out is that the company doesn’t try to compete on the same terms as the leaders. The stated approach is explicit: a methodical, long-term development strategy, “not focused on volume and advertising, but rather on creation and consumer loyalty.” Whether that’s a genuine structural difference or a polite description of operating at a smaller scale is a question worth holding onto — it belongs in the competitive advantage section, not here.

Geographic revenue

North America alone accounted for €347.1 million of the €899.4 million in group revenue in 2025 — 38.6% of the total, and the only country the group is required to flag individually under IFRS 8, since the United States crosses the 10% threshold on its own. One US customer represented 12.6% of total group revenue in 2025. That’s a meaningful single point of dependency for a company whose whole model is built on distribution breadth.

The rest of the map: Western Europe €162.7 million, Asia €115.0 million, Eastern Europe €79.1 million, South America €78.7 million, France €57.9 million, Middle East €52.2 million, Africa €6.8 million. Asia actually declined year-over-year (from €125.2 million in 2024), a reversal that runs against the growth narrative used elsewhere in the report for the region.

The practical read: Interparfums is not a diversified basket of comparable-sized markets. It’s a US-led business with a long tail of smaller regions attached, and the dollar exposure that comes with it — the same currency dynamic management pointed to as the main drag on 2025 profitability.

Competitive advantage

Porter, Mauboussin, and most moat frameworks converge on the same question: why can’t a well-funded competitor simply replicate what this business does? For Interparfums, the honest answer starts with an admission — the most obvious asset in this business, the brand name on the bottle, isn’t one Interparfums owns or controls. That has to be dealt with directly before anything else.

What looks like a moat and isn’t one. A licensing contract is not a durable competitive advantage. It’s a lease. Burberry proves the point: thirteen years into the relationship, the license was terminated in 2012, and Interparfums lost the brand entirely — not gradually, just a contract that ended. Any investor tempted to treat the brand portfolio as a permanent asset base needs to hold that example in mind. The Maisons chose Interparfums once. They can choose someone else next time the contract comes up.

What might actually be durable. Three things, in order of how convincing they are.

Renewal behavior over four decades. Contracts ending is the exception, not the rule. Montblanc has just been extended to 2031. Boucheron to 2027. Coach’s latest renewal added five more years. Across roughly forty licensing relationships signed since 1988, the group has lost very few outright — Burberry is close to the only clean example in the report. That’s not proof of a moat, but it is evidence of a switching cost that runs in Interparfums’ favor: a Maison that terminates a working relationship has to rebuild an entire fragrance operation — creative process, manufacturing partners, global distribution — either in-house or with a competitor, and accept the multi-year gap in launches while that happens. Rebuilding is expensive and slow for the licensor, not just for Interparfums.

Contractual minimums that work both ways. In a standard licensing deal, Interparfums pays the Maison (Montblanc, Coach, and so on) a royalty calculated as a percentage of the sales Interparfums generates from that brand’s fragrances — more sales, more royalty owed to the Maison; fewer sales, less owed. On top of that, the group discloses €308.7 million in minimum guaranteed royalty commitments: fixed floor payments that Interparfums owes its licensors even if actual sales fall short. The direction of the money is always the same — from Interparfums to the Maison — the minimum guarantee just sets a floor under it.

That floor is a liability for Interparfums: a bad year on a given brand doesn’t reduce what’s owed. But it’s also worth reading the other way. A Maison agrees to a multi-year minimum-guarantee structure, instead of a simple percentage-of-sales royalty, because fragrance licensing is a narrow market with specific, hard-to-replace expertise — a handful of operators (Interparfums, Coty, L’Oréal, a few others) who can actually run the twelve-to-eighteen-month development cycle, the manufacturing network, and the global distribution. The Maison isn’t choosing from a deep bench of interchangeable vendors it can threaten to replace next quarter. Locking in a guaranteed minimum is what a licensor does when it has already accepted that switching partners would be slow and costly on its own side too — not just a gesture of trust, but a rational response to a market with few credible alternatives.

Operational depth that’s hard to buy quickly. Forty years of relationships with fragrance houses, glass manufacturers, and packaging suppliers; a twelve-to-eighteen-month development cycle run in close coordination with each Maison’s own creative team; a deliberately diversified production network so no single subcontractor failure can halt a launch. None of this is unique in the abstract — Coty and L’Oréal have the same kind of infrastructure, at greater scale. But it is genuinely difficult for a new entrant, or for a Maison trying to bring fragrance in-house, to replicate quickly. That barrier belongs to the handful of established licensing operators as a group, not to Interparfums specifically against Coty or L’Oréal.

The honest conclusion. This is not a moat in the strict sense — no network effect, nothing that gets structurally harder to dislodge as the company gets bigger. It’s a set of relationship-based, contractually-reinforced barriers that are real and have held up for decades, but that depend on Interparfums continuing to execute well for every Maison, every renewal cycle, indefinitely. The competitive position is defended continuously, not automatically. What should reassure an investor is not the existence of a moat — there isn’t one — but the track record of a company that keeps winning the renewal anyway, cycle after cycle, for forty years.

Financials

Key metrics

Comments

On revenue growth (2021-2025) — the deceleration is the single most important line in this table. Growth ran at 26%, 13%, and 10% in the three years through 2024, then dropped to 2.1% in 2025. Management attributes the slowdown to two external, largely non-repeating factors: an unfavorable EUR/USD move that cost roughly €20 million in sales, and a 15% US tariff introduced mid-2025 that forced pricing flexibility on the group’s largest market. The brand-level picture is more mixed than that framing suggests, though — Coach and Lacoste both grew in 2025, while Jimmy Choo and Montblanc, two of the three historical pillars, actually declined. Whether 2026 confirms this as a one-year currency-and-tariff air pocket or the start of a genuine deceleration is the question the valuation section has to answer.

On operating margin — the trend peaked in 2023 at 20.7% and has declined two years running, to 19.5% in 2025. That’s a smaller move than the revenue deceleration, but it’s moving in the same direction, and it should be watched rather than dismissed as noise.

On net income — the headline number looks like a decline: €129.9m to €126.6m, -2.5%. But that’s the unadjusted figure, and it’s misleading on its own. Interparfums discloses a tariff-adjusted net income of €132.3 million for 2025 — stripping out the one-off €7.6 million cost from the 2025 US tariff introduction — which comes in +2% versus 2024. Once that single external policy shock is removed, net income didn’t shrink in 2025. It grew, modestly, in line with a business absorbing a temporary cost rather than losing structural profitability.

On cash generation — operating cash flow rose from €107.7 million in 2024 to €150.0 million in 2025, even as reported IFRS net income dipped slightly. The driver is working capital: inventory swung from building up (-€19.3m in 2024) to being drawn down (+€23.3m in 2025). Cash conversion improved in the same year earnings stagnated — either disciplined inventory management going into a slower year, or a one-off destocking that won’t repeat. Capex roughly doubled to €40.5 million in 2025 (from ~€20.5m in 2024), which limits how much of that cash flow gain shows up as free cash flow, but estimated FCF still grew from roughly €87 million to €109 million.

On ROE — calculated directly from the reported figures, group-share ROE was approximately 17.3% in 2025 and 18.6% in 2024. Down slightly, consistent with the earnings deceleration already discussed, but still well above what’s typical for a consumer staples business of this size. Some third-party sources cite a higher range, closer to 19-21% — the figure used here comes from a direct calculation on the group’s own reported net income and equity.

On the balance sheet — net cash has grown every year except 2022, and stands at €63.3 million. There is no leverage overhang to monitor here, which matters when weighing this business against the two other positions in the portfolio that do carry meaningful debt structures.

Management

The founders are still in charge

Interparfums has had the same two people at the top since it was founded in 1982. Philippe Benacin and Jean Madar co-founded the company, and forty-plus years later they still control it — together holding 44% of Interparfums Inc., the Nasdaq-listed US parent, which in turn owns 72% of Interparfums SA, the Paris-listed operating entity. That’s not a symbolic stake. It’s a controlling one, structured through two layers, held by the same two founders since day one.

Benacin holds the Chairman-CEO role combined (Président-Directeur Général) — a concentration of power worth naming directly. In most governance frameworks that combination draws scrutiny, and it should here too: there is no independent chair providing a check on the CEO’s decisions. What offsets this to some extent is tenure and track record rather than structure — Benacin has run this exact business, with the same partner, through four decades of licensing cycles, without the kind of governance blow-up (auditor turnover, restated accounts, board revolt) that would normally accompany that level of concentrated control. Philippe Santi serves as Deputy CEO (Directeur Général Délégué), providing some operational counterweight day to day, but the ultimate authority sits with Benacin.

Skin in the game

A 44% combined stake in the parent company is about as direct an alignment as a shareholder can ask for — Benacin and Madar’s personal wealth moves with the stock, full stop. This is a materially different situation from a hired-in turnaround CEO with a few million euros of purchased shares. These are the two people who built the business, still holding the majority of the equity that controls it, forty years in.

Benacin’s annual variable pay is split 50% financial criteria and 50% extra-financial (qualitative and ESG) criteria — a heavier weighting toward non-financial metrics than is typical, and one investor focused purely on capital allocation might want to interrogate.

The succession

Succession planning is not addressed. For a business this dependent on long-standing personal relationships with luxury houses — built and maintained personally by Benacin and Madar over decades — the absence of a visible succession plan is a real gap, not a formality.

The two theses

Why the stock has fallen

The decline is not a single event. It’s a sequence of three, each compounding the last.

The first hit came on November 19, 2025, when Interparfums lowered its 2025 revenue guidance to €890 million and — breaking with its usual practice — declined to give any 2026 target, citing reduced visibility given the number of favorable and unfavorable variables in play. The stock fell more than 9% that morning. Analysts read the implied fourth-quarter figures as a mid-single-digit organic decline, and one research note flagged an unfavorable 2026 comparison base tied to the Boucheron license’s scheduled expiration at the end of 2025. The sector context made it worse: L’Oréal had already signaled that its own fragrance division growth had slowed sharply in the third quarter, and at least one broker described the broader prestige fragrance segment as losing momentum industry-wide.

The second hit was slower and structural: a twelve-month slide that had, by early March 2026, taken the stock down roughly 40% year-over-year, trading near €24 and testing a key technical support level.

The third hit landed with Q1 2026 results on April 22: revenue of €215.5 million against €235.5 million a year earlier, an 8.5% decline, driven by unfavorable currency effects and geopolitical disruption in the Middle East. That report landed in the middle of a broader selloff across French luxury names — Kering and Hermès both dropped sharply the same month on sector-wide concerns.

The one piece of good news buried in this sequence is easy to miss if you only read the November headlines: the Boucheron license that analysts flagged as a comparison-base risk for 2026 was, in fact, extended — not lost. In February 2026, Boucheron and Interparfums agreed to prolong their partnership on the main existing lines through December 31, 2027. The market priced in a loss in November. The actual outcome, three months later, was a renewal. That gap between feared outcome and actual outcome is worth holding onto going into the bull case.

As of early July 2026, the stock trades around €25-26, within a 52-week range of €21.78 to €35.60 — down roughly 30% over twelve months. The PE has compressed to approximately 16-17x, against a historical range that has run 30-40x in normal years. The dividend yield sits around 4-4.5%, with a payout ratio near 70%.

Bear

The revenue deceleration is broader and deeper than management’s FX-and-tariffs framing suggests. 2025’s slowdown was explained as currency and tariffs. But Q1 2026 revenue fell 8.5%, and the explanation expanded to include Middle East geopolitical disruption — a third external factor layered onto the first two. At some point, a pattern of successive external explanations for successive quarters of decline starts to look less like bad luck and more like a business that has lost some pricing power or demand resilience it used to have. Operating margin has now declined for two consecutive years, from 20.7% in 2023 to 19.5% in 2025, and margin compression doesn’t automatically reverse when a currency does.

No structural protection, only behavioral evidence. Every brand in the portfolio is rented, not owned. Burberry is proof that a decades-long relationship can still end. The renewal track record is real — Boucheron’s extension confirms it again — but it’s a pattern, not a guarantee, and a pattern can break on any given renewal date.

Customer and geographic concentration are real risks. North America is 38.6% of group revenue. One single US customer accounts for 12.6% of total group revenue on its own. A distribution shift or a renegotiation with that customer would show up immediately and materially in the numbers.

The new brands are a cost today, a promise for tomorrow. Off-White™, Annick Goutal, and the Longchamp license don’t contribute meaningfully to revenue until 2027 at the earliest. They consume management attention and development capital while the base business is already decelerating.

A Maison could stop needing a licensee at all. The Burberry precedent already covers the risk of losing a license to a competitor. There’s a more severe version of that risk the bear case hasn’t named yet: a Maison deciding to bring fragrance production in-house entirely, exiting the licensing model rather than switching partners within it. This isn’t hypothetical — LVMH, Chanel, and Estée Lauder already run fragrance divisions internally, proving the model is viable at scale. If a house the size of Coach or Montblanc concluded it now had enough marketing and industrial infrastructure of its own to justify internalizing, Interparfums wouldn’t lose a renewal negotiation to a rival licensor. It would lose the category permanently, with no contract left to compete for next cycle.

Governance concentration compounds all of the above. A combined Chairman-CEO role, held by the same person for over forty years, with no visible succession plan and licensor relationships that are, by the report’s own emphasis, personally built.

If the bear case is fully right: margin keeps drifting down through 2026-2027, external headwinds prove sticky rather than transitory, and the market settles on a lower structural growth rate for the whole licensing model — not a collapse, but a re-rating from “quality compounder” to “mature operator in slow decline,” with the PE multiple staying compressed near current levels rather than reverting toward its historical range.

Bull

The market has already priced in the worst version of this story, and the worst version keeps not happening. November 2025 priced in a lost Boucheron license. Boucheron was renewed in February. The PE has compressed from a historical 30-40x range to roughly 16-17x (unadjusted) — a compression larger than the actual deterioration in the fundamentals justifies, given that revenue is still growing, tariff-adjusted net income actually grew 2% in a year with two simultaneous external shocks, and operating cash flow improved sharply.

Forty years of renewal behavior, tested again in real time. Montblanc extended to 2031. Boucheron just extended to 2027, after the market had priced its loss. Coach added five more years. That consistency, sustained through the exact kind of pressure the market is currently pricing as terminal, is the strongest evidence available that the licensing relationships are more durable than a rented-brand framework implies.

The balance sheet gives this business time the market isn’t pricing. Net cash of €63.3 million, zero structural leverage, a dividend maintained through the slowdown, and a share buyback program approved at the April 2026 shareholder meeting — none of this is the profile of a business under financial stress. It’s the profile of a business absorbing a demand-side air pocket from a position of strength.

The next growth layer hasn’t started yet. Lacoste is still ramping. Off-White™, Annick Goutal, and Longchamp don’t show up in the numbers until 2027. A market pricing the stock on trailing growth is, by definition, not pricing what happens when that layer starts contributing.

What would need to be true for this to work: the Middle East and FX headwinds normalize rather than persist, the Boucheron pattern — feared loss, actual renewal — holds for the next major license coming up for renewal, and the 2027 brand launches deliver enough incremental revenue to restore double-digit growth. None of these are heroic assumptions. They’re a continuation of what this business has done for forty years, priced today as if it might stop.

Blind spots

Q1 2026’s 8.5% revenue decline is one data point, not a trend line. Whether it stabilizes, worsens, or reverses in H1 2026 will only be visible once that report is published. Everything in the bear and bull cases about the trajectory of the current slowdown is, for now, an extrapolation from a single quarter.

Succession planning is absent, even though the founders’ age makes it a near-term question, not a distant one. Benacin (67) and Madar (65) have run Interparfums together since 1982. No succession plan is disclosed. It will certainly be someone currently in the organization.

Valuation

The metric

TTM EPS based on IFRS net income, with one adjustment I’m making explicit rather than silently baking in: 2025 group net income of €126.6 million already absorbs a one-off €7.6 million cost from the 2025 US tariff introduction. Management’s own disclosure isolates this, giving a tariff-adjusted net income of €132.3 million (+2% versus 2024) — a genuinely comparable, recurring-basis figure rather than a number distorted by a single external policy shock. I’m using that adjusted figure as the base for the scorecard below, on roughly 80 million shares outstanding, for an adjusted EPS TTM of approximately €1.65. The unadjusted IFRS EPS, for reference, is closer to €1.58.

At a current price of roughly €25.50, the adjusted PE TTM stands at approximately 15.5x — against a historical range that ran 30-40x in normal years, and the lowest level the stock has traded at in years.

The scorecard

The table below uses adjusted EPS TTM of €1.65, a 2.5-year horizon to 2028, and three growth scenarios. Bear case assumes 3% annual EPS growth — consistent with the current slowdown persisting largely unresolved. Central case assumes 8%, roughly in line with what the licensing model has delivered across past cycles once currency and tariff effects normalize. Bull case assumes 14%, reflecting a return to something closer to the growth this business posted before 2025. Combinations that are internally contradictory — a bear earnings trajectory paired with a bull-level multiple, or vice versa — are marked Irrelevant.

One important caveat: all figures in the scorecard are pre-tax and pre-fees. The actual return in your hands will depend on your tax situation, the investment vehicle you use and any transaction costs. Run the numbers for your own situation before drawing conclusions.

How to read this. The only scenarios producing a loss combine PE 12x with either bear-case or central-case earnings — a multiple below where the stock trades even today, applied to a growth rate that’s already the pessimistic case. That’s a narrow downside window, not a broad one.

The central case at PE 20x — €40, +58%, +20.0% CAGR — requires no heroic assumption. It prices the business at barely above two-thirds of its own historical multiple, with EPS growth roughly in line with what this licensing model has delivered across past cycles. That’s what a partial re-rating looks like when a sentiment-driven discount unwinds without needing the bull case to be right.

My conviction is the Bull case with a 25-30 PE ratio, happening in less than 2,5 years.

The asymmetry

At today’s 15.5x, the market is pricing something close to the bear case already. Getting back to even a partial re-rating — not the historical 30-40x premium, just a multiple more consistent with a stable licensing operator putting a rough year behind it — produces a return profile that doesn’t require the bull case to be right, only for the current slowdown not to be permanent.

Portfolio considerations

Sizing

The safer the profile, the more room it can occupy in the portfolio

On my portfolio, it is a defensive stock with great repricing opportunities so I could allow a significant portion of my portfolio, up to 15% in the initial position. But I’m lacking of cash, and i’m confortable with my actual quality positions (booking, novo, adobe, etc.). I don’t want to reallocate massive chunks of my multibagger stocks (Nebius, Atos, Strategy) as they will provide most of my returns. So I will trim Nebius as soon as the stock goes back to its ATH, around 10%. I’m not comfortable with the current Nebius price, down because of fears of the meta competition.

Correlation

As said in the introduction, what I like in this stock is the decorrelation to the AI narrative. No bottleneck, no risk of circular funding. You can sleep on it for 2 years, just waiting for the catalysts to happen. Interparfums is a great stock in case of tension with the tech narrative: as offensive stocks will be beaten by the market, I can reallocate with a potential revaluation.

Subscribe now

Conclusion

Interparfums is not a hard business to understand. Sign a license, make a fragrance, sell it in 120 countries, collect a royalty or a margin. Forty years of doing exactly that, without a scandal, without a debt crisis, without a governance blowup.

2025 and Q1 2026 made it look harder than it is. A currency move, a tariff, a guidance cut that broke with the company’s own habit of quiet reliability. The market did what markets do with a broken habit — it repriced the stock as if the habit itself was gone, not just interrupted. The PE compressed from a historical 30-40x range to roughly 15-16x on a normalized basis. That’s not a mild discount. That’s a business priced as if the growth engine were structurally damaged.

Nothing in the numbers supports that read. Net cash. A dividend held through the worst of it. A tariff-adjusted net income that actually grew in 2025, once the one-off is stripped out. A Boucheron license the market priced as lost in November, renewed in February. Two of the three historical brand pillars declined this year, but the newer ones — Coach, Lacoste — picked up the slack, which is closer to a portfolio rotating than a portfolio failing.

The real risk isn’t a moat question — there isn’t one, and there never was. It’s a renewal question, license by license, forever. Forty years of renewals is not a guarantee. It’s a track record. The bear case only wins if that track record breaks, and nothing in this report says it has.

What this position offers isn’t the asymmetry of a turnaround or the disruption-fear unwind of a mispriced platform. It’s something plainer: a defensive business, temporarily priced like a declining one, sitting completely outside the narrative that moves the rest of this portfolio. That’s not the loudest reason to own something. It might be the most useful one.

Image credits

Product and marketing visuals sourced from the investor relations section of Interparfums. Financial charts sourced from Fiscal.ai.

Important Disclosure & Disclaimer

All content published by JB Peter on this platform is strictly for educational and informational purposes. It does not constitute investment, financial, legal, or tax advice, nor does it represent a personal recommendation or solicitation to buy or sell securities. This research is operated by ORIACON (SASU) and reflects independent corporate analysis. Every reader must conduct their own independent research (Due Diligence) or consult a licensed professional before making any financial decision, as financial markets involve a high risk of capital loss. At the time of writing, ORIACON or the author HOLD shares in the company analyzed in this article. Following this publication, ORIACON and the author reserve the right to buy, sell, or modify positions in any security mentioned at any time, without prior notice to readers or subscribers.

This question becomes considerably more interesting inside a tax-free wrapper, where the full magnitude of a re-rating compounds without erosion.

Microsoft is currently trading at a PE of 22x — its lowest valuation in a decade, last seen briefly during two narrow windows in the last ten years. The business is not showing any of the symptoms that usually accompany a multiple this depressed. Azure has accelerated for five consecutive quarters. Management is publicly admitting it cannot install GPUs fast enough — not because demand is missing, but because there isn’t enough electricity to plug them in.

And yet the stock has fallen hard. The market has a name for what’s worrying it: OpenAI. A partner Microsoft owns 27% of, has committed $13 billion to, and increasingly depends on for the AI growth story that justifies a $190 billion capital expenditure plan. No audited financial statements exist for OpenAI, and none will until an IPO whose timing remains genuinely uncertain.

This article starts with the business itself — what Microsoft actually does, how it makes money, who its customers and suppliers are, and what makes it structurally difficult to displace. From there, it examines the two competing theses: why the market is afraid, and whether that fear holds up under scrutiny. The valuation section translates those arguments into a scorecard with concrete scenarios. The article closes with portfolio considerations and a conclusion.

I know you’re busy, so I always start with a summary called The short version. If that’s all you have time for, that’s fine. The rest of this article explains why.

Subscribe now

The Short Version

Microsoft is the world’s dominant enterprise software and cloud platform — hundreds of millions of users locked into Windows, Office, and Azure not by preference but by switching costs that compound the longer they’re deployed. The stock has fallen roughly 25% from its 2025 peak. The market has one explanation: a $190 billion AI capital expenditure plan whose return isn’t yet visible in free cash flow, and a growing dependence on OpenAI — a partner Microsoft owns 27% of, has committed $13 billion to, and cannot audit, because no audited OpenAI financial statements exist.

My view: the market is pricing one real, bounded risk as though it were a structural one. Azure has accelerated for five consecutive quarters, with management openly admitting the constraint is electricity to power GPUs, not demand for them. Operating margin has expanded every year since FY2023. Two independent hyperscalers — Alphabet and Amazon — are reporting the same pattern: AI capex translating into expanding margins, not eroding them. None of that depends on OpenAI’s solvency.

EPS has compounded at roughly 18-20% a year for three years. The stock has compounded more slowly. That gap is the trade.

At a PE of 25x — still below the ten-year average of roughly 32x — the central scenario in the valuation scorecard implies a price near $795 within 3.5 years, a return of +113%. The only scenario in the scorecard that produces a loss combines a multiple below the broader software sector’s median with EPS growth cut roughly in half from its recent pace, sustained for years. That is not a base case.

Understanding Microsoft’s business, supply chain and environment

History

Microsoft was founded in 1975 and built its first fortune on a near-monopoly in desktop operating systems and office productivity software. For two decades, that dominance was nearly unchallengeable — and nearly fatal to the company’s long-term relevance. The early 2010s saw Microsoft miss the mobile transition almost entirely, dismissed by much of the market as a legacy software vendor in terminal decline.

The turnaround began in 2014, when Satya Nadella became CEO and redirected the company toward cloud computing and a cultural shift away from internal competition toward what he termed a “growth mindset.” Azure, a minor product line at the time, became the centerpiece of a decade-long transformation. By the mid-2020s, Microsoft had gone from a company written off as irrelevant to one of the most valuable companies in the world.

Two recent decisions define the current chapter. In October 2023, Microsoft completed the acquisition of Activision Blizzard for $75.4 billion, its largest deal in history, consolidating its position in gaming content. And since 2019, it has built an increasingly deep partnership with OpenAI — an investment of $13 billion in exchange for a 27% equity stake and a privileged commercial relationship that has become the centerpiece of the current market debate. That partnership was substantially restructured on April 27, 2026, a change significant enough to warrant its own section later in this article.

The engine: three segments, one platform

Microsoft organizes its business into three reportable segments, each at a different stage of maturity.

Productivity and Business Processes is the historical core: Microsoft 365 Commercial, Microsoft 365 Consumer, LinkedIn, and Dynamics 365. This segment generated $35.0 billion in revenue in the quarter ended March 2026, up 17%, with Microsoft 365 Commercial cloud revenue up 19% and LinkedIn up 12%.

Intelligent Cloud is the growth engine: Azure, server products, GitHub, and enterprise services. Revenue reached $34.7 billion in the same quarter, up 30%, with Azure itself growing 40% — its fifth consecutive quarter of acceleration, a streak that has persisted despite management repeatedly acknowledging that supply, not demand, is the binding constraint.

More Personal Computing is the most mature and most heterogeneous segment: Windows OEM and devices, gaming (Xbox, Game Pass, Activision Blizzard content), and search advertising. It was the only segment to decline in the most recent quarter, down 1%, dragged by a 2% fall in Windows OEM revenue and a 5% decline in Xbox content and services.

Layered across all three segments is Copilot — Microsoft’s attempt to embed AI monetization horizontally rather than concentrate it in a single product line. GitHub Copilot has demonstrated clear, measured value (developers documented as 55% faster on well-scoped coding tasks) and has scaled to 4.7 million paid subscribers. Microsoft 365 Copilot has grown faster in absolute seat count — over 20 million paid seats, up from 15 million the prior quarter — but penetration of the installed M365 base remains low, at roughly 4.4%, and retention appears weaker than GitHub Copilot’s. Management has already begun pivoting the product toward agentic capabilities rather than a generic assistant, an implicit acknowledgment that the first iteration underdelivered.

Customers

Microsoft’s customer base spans individual consumers, small and medium businesses, large global enterprises, public-sector institutions, service providers, application developers, and OEMs. Roughly 80% of the Fortune 500 use Azure AI Foundry. LinkedIn counts 1.2 billion members. Gaming reaches 500 million monthly active users.

One nuance worth naming explicitly, because it complicates the popular “European digital sovereignty” narrative: public-sector deviations from Microsoft are currently concentrated almost entirely in desktop software — France’s planned migration of 2.5 million government workstations to Linux, Germany’s Schleswig-Holstein region, Italy’s Ministry of Defense — not in identity infrastructure or cloud. Active Directory, Entra ID, and Azure consumption remain firmly within Microsoft’s grip even in administrations actively reducing their Windows footprint, most plausibly because the cost-benefit calculation for migrating cost-driven desktop licensing is straightforward, while migrating identity and PaaS-layer infrastructure is not.

Suppliers

Microsoft’s most consequential supplier relationship is also, structurally, its most ambivalent: Nvidia. GPU capacity is the binding constraint on the company’s entire AI growth narrative, and Nvidia is by far the dominant supplier of that capacity — a dependency Microsoft is actively working to reduce through internal silicon development (Maia, Cobalt), though that effort remains early-stage.

The second critical dependency is OpenAI itself, which occupies an unusual dual role: simultaneously a technology partner whose models power Copilot, and the single largest identified consumer of Azure’s own AI compute capacity. Independent analysis of Microsoft’s most recent 10-Q estimates that consumption tied to OpenAI represents the largest single line within Microsoft’s $37 billion AI revenue run rate — though Microsoft itself has not published an official breakdown, and this figure should be treated as a well-reasoned estimate rather than a confirmed fact.

Beyond these two, the 10-K explicitly flags concentration risk in hardware components: “there are few qualified suppliers for certain components of our servers and devices.” Azure AI Foundry also hosts models from third parties — OpenAI, Cohere, DeepSeek, Meta, Mistral, xAI — giving Microsoft a degree of model-layer diversification that reduces, without eliminating, the company’s dependency on any single AI lab.

Competition

Cloud infrastructure. Azure competes directly with Amazon Web Services and Google Cloud Platform in what is effectively a three-player oligopoly. Market share estimates converge around AWS at 30-33%, Azure at 23-28%, and Google Cloud at 11-13% — the three together capturing roughly two-thirds of the global market, with the remainder fragmented among smaller players. Azure has narrowed the gap with AWS over recent quarters rather than widened it, growing 40% against AWS’s 28% in the most recently reported periods.

Productivity software. Google Workspace is Microsoft’s only meaningful direct competitor here, and the two together control the overwhelming majority of the market — though the precise leader depends heavily on methodology. By paid enterprise seats, Microsoft 365 leads decisively, with over 450 million paid seats and roughly 75% of the Fortune 500. By raw domain or account count, Google Workspace is often cited as larger, buoyed by small businesses and education accounts where Google is free or near-free. This is a genuine duopoly with no credible third entrant.

Gaming. Sony’s PlayStation remains the dominant console platform, with Microsoft a consistent second, though Activision Blizzard content and Game Pass subscription economics have strengthened Microsoft’s position on content and recurring revenue even where hardware share lags.

Search and consumer AI. Google Search retains over 90% global market share; Bing and Copilot remain minor players in pure search terms. On the consumer AI chatbot front specifically, ChatGPT’s share of monthly active usage has declined from roughly 85% to around 45% over the past two years as Gemini, Claude, DeepSeek, and others have gained ground — though this metric measures consumer app usage, not enterprise revenue, where Anthropic’s Claude reportedly overtook OpenAI as early as mid-2025.

None of Microsoft’s direct competitors qualifies as a category-definer in the strict sense of having created an uncontested category. The closer candidate for that label sits one layer up the stack: Nvidia, whose dominance in AI compute underwrites the infrastructure all three hyperscalers depend on, and OpenAI, whose generative AI breakthrough effectively created the category Microsoft is now monetizing through partnership rather than head-on competition — a strategic choice that distinguishes Microsoft’s approach from Google’s, which is building Gemini in-house and therefore competes more directly with its own cloud customers in the AI lab space.

Competitive advantage

Most competitive-moat frameworks converge on a single question: why can’t a well-funded competitor simply replicate what this business does? For Microsoft, there are three distinct, mutually reinforcing answers — a fourth, the deepest one, is developed separately in the bull thesis below.

Economies of scale. The 10-K states plainly that the company’s datacenters “deploy computational resources at significantly lower cost per unit than smaller ones” — a direct structural barrier to entry that no new entrant can replicate without comparable capital intensity. The 2026 capital expenditure guidance of roughly $190 billion is itself a barrier: few companies on earth can fund infrastructure at this scale.

Network effects. LinkedIn’s 1.2 billion members and GitHub’s developer ecosystem both compound in the classic two-sided sense — more users make the platform more valuable to every other user.

A self-reinforcing, multi-layer lock-in system. Unlike a single-axis moat, Microsoft’s advantage is a system in which the operating system layer, the line-of-business software layer, the PaaS layer, and now the AI layer all reinforce one another rather than standing as four separate, independently defensible barriers. This is the single most important point of this analysis, and it deserves more than a paragraph — it’s developed in full in the bull thesis below.

Financials

Key financial metrics

Comments on the numbers above

On the free cash flow decline. FCF fell to $71.6B in FY2025 from $74.1B in FY2024 — not because operating cash generation weakened (it grew 15% to $136.2B) but because capital expenditure grew faster, up 45% to $64.6B. This is a capex story, not an earnings-quality story. FCF minus stock-based compensation tells the same story more starkly: $63.4B in FY2024 was the peak of the period, falling to $59.6B in FY2025, and the pressure has only intensified since — Q3 FY2026 alone saw FCF-SBC fall 26.7% year over year, the steepest decline in the series. Worth noting: this isn’t purely a recent, AI-driven phenomenon. FCF-SBC was already down 13.4% in FY2023, well before the current capex cycle began in earnest — the tension between operating cash generation and capital intensity has a longer history than the AI narrative alone would suggest.

On the volatility inside net income tied to OpenAI. Microsoft’s equity-method stake in OpenAI has produced wildly different quarterly impacts: a $4.1 billion pre-tax loss in Q1 FY2026, a $7.6 billion gain in Q2 FY2026 — driven by a revaluation following the April 2026 restructuring, not by any improvement in OpenAI’s underlying profitability — and a near-neutral $14 million loss in Q3 FY2026, down sharply from a $583 million loss in the same quarter a year earlier. This swings on Microsoft’s use of the hypothetical-liquidation-at-book-value (HLBV) accounting method, which can produce results that diverge meaningfully from a simple pro-rata share of OpenAI’s reported net income. None of this reflects a change in OpenAI’s actual cash burn, which independent estimates continue to place in the tens of billions annually.

On Azure’s acceleration despite a binding supply constraint. Azure growth has climbed in every period shown — 29% (FY2023), 30% (FY2024), 34% (FY2025), 40% (Q3 FY2026) — five consecutive quarters of acceleration, each beating prior guidance. That is the opposite of what you’d expect if AI demand were overstated. CFO Amy Hood has stated explicitly across multiple quarters that supply, not demand, remains the limiting factor, adding in the Q3 FY2026 call that the company expects to “remain constrained at least through 2026” — an admission that costs the company growth in the short term and therefore carries more credibility than a typical demand-side sales pitch.

On margin compression in the cloud segment. Microsoft Cloud gross margin fell to 66% in Q3 FY2026, which the company attributes directly to the cost of scaling AI infrastructure, partly offset by efficiency gains in Azure and Microsoft 365 Commercial cloud. Operating margin at the company level has nonetheless continued to expand throughout the period (41.5% → 44.6% → 45.6% → 46.3%), helped by declining total headcount — discipline below the gross-margin line is absorbing some of the pressure from infrastructure scaling rather than letting it flow straight through to the bottom line.

On FY2023 as the trough year. FY2023 stands out as the one period in this table where net income (-0.4%) and diluted EPS (+0.3%) were essentially flat — a pause following the post-pandemic deceleration, before the cloud and AI reacceleration that has driven double-digit growth in every subsequent year shown.

ROIC trend — a longer pattern than the AI capex story alone

The current ROIC of 19.9% is not an isolated trough created by the AI investment cycle — it is the continuation of a decline that began five years earlier. According to Fiscal.ai data spanning fiscal years 2017 through the trailing twelve months, Microsoft’s ROIC peaked at 31.4% in FY2021, held near that level through FY2022 (30.6%), then declined in successive years: 25.5% (FY2023), 24% (FY2024), 21.8% (FY2025), 19.9% (LTM). Over the same nine-year window, diluted EPS rose from $3.25 to $16.80 — a 20.0% CAGR, against a roughly flat ROIC CAGR of -0.8% over the same period.

The two-phase nature of this decline matters for how the AI capex debate should be framed. The first phase, FY2021 to FY2023, predates the current AI infrastructure buildout and likely reflects post-pandemic normalization combined with digesting the integration of the Activision Blizzard acquisition. The second phase, FY2024 onward, coincides with and likely reflects the accelerating capital intensity discussed throughout this section. Treating the entire nine-year decline as evidence that “AI investment is destroying returns on capital” overstates the case — roughly half the decline happened before the AI capex cycle began.

What the trend does confirm is the mechanism already described above: Microsoft is generating substantially more absolute profit (EPS more than doubling since FY2021) while doing so on a capital base that is growing even faster — precisely the signature of a company in a heavy investment phase, where marginal returns on newly deployed capital dilute the average ROIC even as total earnings power continues to expand.

Geographic revenue

Microsoft discloses a simple geographic split: the United States accounted for $144.5 billion, or 51.3%, of fiscal 2025 revenue, with the remaining 48.7% generated elsewhere. The company notes that no individual country outside the US exceeds 10% of revenue, so no further granularity is published.

Capital expenditure — pace and intent

Capex has compounded at a rate that materially outpaces revenue growth: $28.1B (FY2023) → $44.5B (+58%, FY2024) → $64.6B (+45%, FY2025) → roughly $190B guided for calendar 2026 (+61% on a comparable basis). Sector-wide, the four largest AI spenders (Microsoft, Alphabet, Amazon, Meta) are projected to spend a combined $505 billion in 2026, up from roughly $366 billion in 2025 — Microsoft’s acceleration is not an outlier, it’s part of an industry-wide pattern.

Roughly two-thirds of this spend goes to short-lived assets (GPUs and CPUs, depreciated over roughly two years), and one-third to long-lived assets (buildings, power infrastructure, land, increasingly financed through leases rather than direct construction). Of the $190B guided for 2026, an estimated $25B reflects component price inflation rather than incremental capacity — spend that doesn’t buy more compute, just buys the same compute at a higher price.

Capacity allocated this way is not earmarked for a single segment in Microsoft’s reporting — it is corporate spend that flows simultaneously into Azure consumption, into Microsoft’s own first-party AI products (Copilot, GitHub Copilot), and into internal R&D acceleration. Management has explicitly cautioned analysts against drawing a mechanical one-dollar-to-one-dollar link between capex and Azure revenue specifically.

Balance sheet and leverage

Total debt: $43.15B. Stockholders’ equity: $343.5B. Debt-to-equity: 0.13. Cash and short-term investments: $94.6B — exceeding total debt, putting the company in a net cash position. EBITDA (operating income plus D&A): approximately $162.7B. At this scale, total debt represents roughly 0.34 years of operating income — about four months — making leverage a non-issue for Microsoft.

A genuine off-balance-sheet item worth flagging: $92.7 billion in datacenter lease commitments not yet commenced.

Accounting notes worth flagging

The auditor is Deloitte & Touche LLP, with an unqualified opinion on both the financial statements and internal controls. Two Critical Audit Matters were identified: revenue recognition complexity (standard for a business combining licensing and cloud contracts) and uncertain tax positions (linked directly to the IRS dispute below). Neither suggests irregularity.

The IRS has assessed approximately $28.9 billion in additional tax related to transfer pricing for tax years 2004-2013, which Microsoft is contesting and has not fully provisioned. Separately, $541 million in litigation accruals are recorded, with an estimated additional $600 million in reasonably possible losses beyond what’s already booked — modest figures relative to a $619 billion balance sheet.

Management

Satya Nadella — CEO since 2014

Nadella inherited a company widely regarded as past its prime and rebuilt its growth trajectory around cloud and, more recently, AI. His public communication style is notably more measured than the typical “platform shift” rhetoric he himself favors — quarterly calls remain anchored in specific operational metrics rather than visionary abstraction, a contrast worth noting given how aggressively the company is spending on an unproven return.

The clearest data point on his character as an operator: when cybersecurity incidents drew public criticism in 2024, Nadella voluntarily reduced his own cash bonus by roughly 50% — from $10.66 million to $5.2 million. The board has since added security as a standalone compensation criterion for senior executives for the first time. This is the same instinct Microsoft’s M365 Copilot pivot reflects at the product level: naming an underperforming bet rather than quietly burying it.

Skin in the game

Nadella’s FY2025 compensation totaled $96.5 million — but the figure that matters more than the total is its composition. 87% of it, $84.2 million, came in the form of stock awards rather than cash, vesting over three to four years and tied to performance metrics including ROIC and relative shareholder return. This is a distinct, individually negotiated package — separate from the broader $12 billion in company-wide stock-based compensation expensed across the workforce — but it illustrates the same principle the board has built into its evaluation framework: pay tracks long-term capital returns and shareholder outcomes, not just short-term financial results. The board’s own scoring of his FY2025 performance reflects that blend — 117% of target on financial metrics, 151.67% on operational assessment — with security added as a standalone criterion for the first time following the cybersecurity incidents discussed above.

Capital allocation discipline

Across the criteria that typically separate disciplined capital allocators from undisciplined ones: capital is unambiguously directed toward the highest-return opportunity (Azure/AI capex dwarfs spend on flatter segments), there has been no pattern of frequent, poorly integrated acquisitions (Activision Blizzard remains the only major deal in the recent period, and it is performing), buybacks have been modest relative to free cash flow ($13.0B in FY2025 against $71.6B in FCF, with $57.3B of a $60B authorization still undeployed), and the board has demonstrably updated its own oversight criteria in response to operational failures rather than leaving them static.

The one mark against the company on this dimension: communication around the OpenAI relationship was, by external accounts, opaque in its early stages — the word “primarily” carried a great deal of unstated weight in describing the source of equity-method losses before Microsoft began isolating the figure explicitly from Q1 FY2026 onward. Transparency improved under pressure rather than arriving proactively.

Two thesis

Bear thesis

Why the stock has fallen

The decline is not explained by a deterioration in fundamentals. Revenue grew 18% in the most recent quarter, Azure has accelerated for five consecutive quarters, and operating margin has expanded every year shown in the table above. The stock fell because the market repriced the risk of a capital expenditure cycle whose return is not yet fully visible in free cash flow.

Three factors combined:

• First, capital expenditure guidance that came in well above consensus — $190 billion for calendar 2026 against a Visible Alpha estimate of $154.6 billion — with management attributing roughly $25 billion of that to component price inflation alone, not incremental capacity.

• Second, a broader market debate about whether AI infrastructure spending across the hyperscaler cohort will generate returns commensurate with its scale.

• Third, and most specifically to Microsoft, a growing dependence on OpenAI — a partner the company owns 27% of, has committed $13 billion to, and whose contractual Azure commitments now account for the majority of the deceleration gap between headline cloud growth and growth excluding that single relationship.

The OpenAI exposure gave the bears a concrete, recurring data point. It is the one argument in this section that does not unwind cleanly under scrutiny.

The bear arguments

Dependence on an unaudited, cash-burning partner. This is the central, legitimate concern, and it deserves to be unpacked as two distinct scenarios rather than one vague risk, because each implies a different magnitude of consequence for Microsoft.

OpenAI is not a public company. It has never filed audited financial statements, and won’t until an IPO whose timing remains uncertain — a confidential S-1 was filed in June 2026, but estimates of when a listing might actually happen range from late 2026 to sometime in 2027, with most analysts leaning toward the later end. In the absence of audited numbers, what’s available is a series of internal projections leaked to the press, and those projections have moved dramatically and repeatedly: cumulative revenue estimates for 2030 have been revised from roughly $85 billion to as low as $39 billion, then back up to roughly $280 billion, within the span of a few months — a swing of nearly sevenfold between the low and high figures, none of it audited, none of it confirmed by OpenAI itself. That volatility in OpenAI’s own internal forecasting is the backdrop against which both scenarios below need to be read.

In the bankruptcy scenario, the direct accounting impact is bounded — a write-down of at most the $13 billion invested, absorbable in a single quarter against net income exceeding $100 billion annually. The more consequential impact is operational: a meaningful share of the $37 billion AI revenue run rate and the 99% year-over-year growth in remaining performance obligations is tied to OpenAI’s own Azure consumption — strip that out, and RPO growth falls to roughly 26%, still healthy but a different growth story than the headline figure implies. Physical capacity would not be destroyed — the binding constraint today is electricity, not demand, meaning a queue of other customers exists to absorb freed capacity — but reallocation would not be instantaneous, and the narrative hit to Azure’s growth rate would be immediate.

In the continuous financing scenario — OpenAI survives but requires large-scale, repeated financing for years — the relevant question shifts from solvency to financing risk. OpenAI’s path to the IPO it has confidentially filed for remains genuinely uncertain in timing, and even a successful listing raising upward of $60 billion would not, on its own, close a cumulative cash burn gap that internal projections place in the hundreds of billions through 2030. Microsoft would likely continue benefiting from this relationship through royalties and Azure consumption regardless of OpenAI’s path to profitability — but the market’s confidence in the durability of that arrangement depends on financial visibility that, as things stand, does not exist.

Margin compression from infrastructure scaling. Microsoft Cloud gross margin has fallen to 66%, and the ratio of capital expenditure to operating cash flow has crossed 50% — a level that concerns analysts more than the absolute dollar figure. The bear case treats this as evidence that the capital intensity of AI-driven growth is structurally different, and structurally worse, than the software-centric model that built Microsoft’s historical margin profile.

European public-sector erosion. Several European governments — France, Germany’s Schleswig-Holstein region, Italy’s Ministry of Defense — have announced migrations away from Microsoft desktop software, framed publicly around digital sovereignty.

What if the bear thesis is really true?

If every element above plays out simultaneously — Azure’s market share gains stall, cloud margins continue eroding under the weight of infrastructure scaling, OpenAI is forced to materially scale back its Azure commitments for lack of financing, and European public-sector erosion spreads beyond the desktop into infrastructure layers — the investment thesis would shift fundamentally. Microsoft would no longer be a quality compounder temporarily mispriced; it would be a mature company in a prolonged overinvestment phase, carrying fixed costs (the $92.7 billion in datacenter leases not yet commenced) against demand that failed to materialize as planned. In that scenario, a PE of 22x would not be cheap — it would still be too expensive for a business facing structural margin erosion and a sharply slower growth trajectory than the one currently priced in.

None of the first three signals — Azure share, cloud margins, European desktop migration — currently shows evidence of this trajectory: Azure has accelerated in every one of the last five quarters, not decelerated; operating margin has expanded every year since FY2023; and European public-sector deviations remain almost entirely confined to desktop software, with identity infrastructure (Active Directory, Entra ID) and Azure consumption remaining intact even in administrations actively reducing their Windows footprint — most plausibly because the cost-benefit calculation for migrating cost-driven desktop licensing is straightforward, while migrating PaaS-layer infrastructure is not. The fourth signal, OpenAI’s financing trajectory, remains genuinely unresolved and is the one component of this bear case that current data cannot rule out.

Bull thesis

The market is pricing one fear and ignoring three independent confirmations that AI capital expenditure is profitable

CFO Amy Hood stated directly in the Q3 FY2026 earnings call, in response to an analyst question comparing AI margins to historical cloud margins, that margins on AI products and tools are already better than cloud margins were at a comparable stage of scale — a claim that is verifiable over time rather than a vague promise, and one the company has every incentive not to make if it weren’t broadly true given how closely analysts now track this exact metric.

That claim does not stand alone. Alphabet was the only one of the three major hyperscalers to “convince investors” in the most recent earnings cycle, according to financial press coverage, on the strength of a clear acceleration in Google Cloud growth that the market read as direct evidence of capex translating into results. Amazon corroborates the pattern with harder numbers: group operating margin reached a record 13.1% in the same period it announced its most aggressive capital expenditure plan in company history — $200 billion for 2026, $344 billion cumulative through 2027 — while AWS itself accelerated to 28% growth. An independent calculation based on AWS’s current margin and growth rate puts the incremental ROIC on AI capital expenditure above 25-30%, comfortably above any reasonable estimate of the sector’s cost of capital.

Three independent companies, three different competitive positions, three confirmations of the same underlying pattern: AI infrastructure investment is not destroying returns on capital at the margin — it is being deployed against demand that is currently outstripping supply, evidenced most directly by management at all three companies admitting, repeatedly and on the record, that capacity rather than demand is the binding constraint on growth. That kind of admission costs a company near-term growth and credibility with the market if it overspends on unrealized demand; it is not the kind of claim a management team makes lightly, and it is the opposite of what you’d expect to hear if the spending were not paying off.

A lock-in system that compounds across every layer of the stack — and gets stronger, not weaker, the deeper AI is embedded

Microsoft’s most durable advantage is not technological superiority at any given moment — it is the cost of leaving. That cost is best understood through specific, lived examples rather than abstraction. A notary running case-management software that only runs on Windows Server cannot migrate without rewriting mission-critical, regulator-adjacent software they did not write and cannot easily replace — the alternative isn’t switching platforms, it’s a multi-year engineering project with legal exposure attached. A systems administrator managing several thousand endpoints does not consider Linux a realistic alternative not because Linux is technically inferior, but because the tooling, the talent pool, and years of institutional muscle memory are built entirely around Active Directory — propose removing it, and the practical response is closer to resignation than migration planning. Independent software vendors who built their products on .NET and SQL Server, on-premise, propagate that same lock-in to every one of their own customers in turn, multiplying its reach far beyond what market-share statistics for Windows or Azure alone would suggest.

This switching-cost moat has a second layer that matters specifically for the AI debate. Infrastructure-as-a-Service workloads — raw virtual machines — are relatively portable between cloud providers; a VM migrates from Azure to AWS with moderate effort. Platform-as-a-Service workloads — applications built directly against proprietary managed services like Azure SQL Database, Azure Functions, or Entra ID — are not; migrating them means rewriting application code, not simply redeploying infrastructure elsewhere. Enterprise AI today is being built almost exclusively on this PaaS layer — Copilot integrations wired directly into Entra ID and Microsoft Graph, agentic workflows built on Azure AI Foundry — which means each new AI deployment deepens the same lock-in rather than creating an independent, swappable product. An organization already entrenched in Windows and Active Directory is also the organization most likely to have built its cloud workloads on Azure PaaS for integration convenience in the first place — the layers reinforce each other rather than standing as separate, independently defensible barriers. The European public-sector migrations discussed in the bear case make this point empirically: even governments with explicit political mandates to reduce dependence on Microsoft have, so far, only managed to substitute the desktop layer, leaving identity and cloud infrastructure untouched.

A demonstrated competence in structuring contracts, not just building products

The April 2026 restructuring of the OpenAI partnership is a recent, concrete illustration of this. Microsoft gave up cloud exclusivity — a real concession, made under pressure from a partner seeking to diversify its infrastructure risk — in exchange for a technology license extended to 2032, royalties capped at 20% of OpenAI’s revenue through 2030 regardless of whether OpenAI achieves AGI, a right of first refusal on capacity, and the $250 billion Azure spending commitment discussed above. The same instinct shows up at the product level: Microsoft 365 Copilot has grown to over 20 million paid seats despite penetration of the installed base remaining low and usage retention appearing weaker than GitHub Copilot’s — the company is successfully monetizing a product whose standalone value proposition remains debated, by bundling it into existing enterprise contracts rather than relying on the product winning on merit alone. This is the same mechanism, applied at different scales: Microsoft’s economic engine has historically depended less on having the best product at any given moment than on making the cost of leaving — or the cost of not adopting what’s already embedded in an existing contract — higher than the cost of staying.

The valuation gap. EPS has compounded at roughly 18-20% annually over the period shown in the financials table. The stock, over a comparable multi-year window, has compounded at a meaningfully slower rate — the PE has compressed from a ten-year average of roughly 31-33x to 22x today, the lowest level in a decade.

Catalysts

Earnings. A quarter where the capex-to-free-cash-flow ratio stabilizes or improves — even without a major beat on revenue — would likely matter more to sentiment than another quarter of headline growth, since the market’s current concern is specifically about cash conversion, not top-line demand.

Structural. Stabilization or further clarification of OpenAI’s financing path — whether through a successful IPO process or continued evidence that its Azure commitments are being honored at scale — would directly address the one bear argument that current data cannot resolve.

Macro. A shift toward lower interest rates would mechanically benefit the valuation of long-duration growth compounders as a class, Microsoft included — though the 2022 precedent, when Microsoft fell further than the broader index during the rate-hiking cycle, is a reminder that this sensitivity cuts in both directions.

What would invalidate the bull thesis

Four signals worth monitoring every quarter:

• The capex-to-operating-cash-flow ratio continuing to deteriorate without stabilization beyond 2026

• Azure growth excluding OpenAI’s contractual commitments (currently estimated around 26%) beginning to decelerate meaningfully

• Continued, unresolved deterioration in OpenAI’s equity-method losses without a new restructuring in Microsoft’s favor

• Commercial RPO converting to recognized revenue at a materially lower rate than the roughly 30% management has signaled for the next twelve months

None of these have appeared as of the most recent quarter reported. When one does, the thesis deserves reassessment.

Blind spots

The exact breakdown of the $37 billion AI revenue run rate by product and by customer is not published by Microsoft. Independent analysis of the most recent 10-Q estimates that OpenAI-linked Azure consumption represents the largest single component, but this remains a reasoned inference rather than a confirmed figure — Microsoft has not isolated it, and this should be revisited once the company provides more granular disclosure, if it ever does.

The precise terms governing newly signed datacenter lease commitments — early termination clauses, penalty structures — are not public. What is known is that roughly 200 MW of pre-commencement capacity has been cancelled with at least two operators, while market-wide lease terms have hardened in landlords’ favor on new agreements, meaning the flexibility demonstrated on cancelled commitments may not extend to leases signed going forward.

The degree to which line-of-business software lock-in protects Microsoft’s enterprise base cannot be precisely quantified from public data — no reliable figure exists for the share of enterprise applications built on the Windows/.NET/SQL Server stack versus portable alternatives. The argument rests on structural logic and specific, verifiable examples rather than an aggregate statistic, and should be read accordingly.

Valuation

This section is the output of everything above — not a standalone recommendation. The numbers only make sense if you’ve read the two thesis section. A valuation without a thesis is just a spreadsheet.

The metric

PE TTM, based on GAAP net income as published — no normalization. This is a deliberate departure from how this kind of section is sometimes built. The volatility inside Microsoft’s reported earnings tied to OpenAI (a $4.1 billion loss one quarter, a $7.6 billion gain the next, a near-neutral $14 million the quarter after that) is recurring and structural rather than a one-time, isolable charge — stripping it out trimester by trimester would introduce more judgment and arbitrariness into the number than it would remove. The published GAAP figure is taken as-is; its composition is discussed in the financials section above rather than adjusted away here.

At a current price of roughly $373, the PE TTM stands at approximately 22x — against a ten-year average of roughly 31-33x, and the lowest level the stock has traded at in a decade.

The scorecard

The table below uses EPS TTM of $16.80, a 3.5-year horizon to January 1, 2030, and three growth scenarios. Bear case assumes 10% annual EPS growth — roughly half the rate Microsoft has delivered over the past three years. Central case assumes 20%, in line with recent history. Bull case assumes 30%, reflecting continued AI-driven acceleration. Combinations that are internally contradictory — a bear earnings trajectory paired with a bull-level multiple, or vice versa — are marked irrelevant.

One important caveat: all figures in the scorecard are pre-tax and pre-fees. The actual return in your hands will depend on your tax situation, the investment vehicle you use, and any transaction costs. A gain of +113% in a taxable account is not the same as +113% in a tax-sheltered envelope. Run the numbers for your own situation before drawing conclusions.

One scenario this scorecard does not attempt to price: a broad market selloff unrelated to Microsoft’s own fundamentals — a recession, a credit event, a systemic AI-sector repricing. Any of these could push the stock below the bear case shown here, temporarily, independent of the company’s underlying performance. The scorecard maps where the stock could trade if the thesis plays out roughly as described; it says nothing about what the market does in between.

How to read this

The only scenario that produces a loss is a PE of 12x combined with EPS growth slowing to 10% a year — a multiple below the median of the broader software sector and a growth rate roughly half of what Microsoft has delivered over the past three years, sustained through 2030. That is not a base case. It requires the AI capex cycle to fail to generate meaningful returns and the multiple to compress well past anything seen even during the past decade’s lowest points.

The central case at PE 25x — $795, +113%, 24.1% CAGR — requires no heroic assumptions. It prices the business at a multiple still below its own ten-year average, with earnings growing at a rate close to what the company has already delivered. That is what a re-rating looks like when a sentiment-driven discount unwinds without requiring the bull case to be right.

The asymmetry

The realistic downside is contained to scenarios that combine both a growth slowdown and a multiple well below anything the stock has historically traded at, sustained over multiple years. The realistic upside does not require the AI investment cycle to be a triumph — it requires it not to be a failure. That gap, not any single price target, is the investment case.

Portfolio considerations

Sizing

This is structured as an initial position — sized to put real capital behind the conviction expressed in this article, not as a maximum allocation. The reasoning follows directly from the bear thesis section above: three of the four risks identified there don’t hold up under scrutiny with current data, but the fourth — OpenAI’s financing trajectory — remains genuinely unresolved. A position sized for full conviction would be premature when one material part of the thesis depends on visibility that doesn’t yet exist.

The stop is not a price. It is a thesis: if the capex-to-operating-cash-flow ratio continues deteriorating without stabilization beyond 2026, if Azure growth excluding OpenAI’s contractual commitments begins decelerating meaningfully, or if OpenAI’s equity-method losses resume worsening without a new restructuring in Microsoft’s favor, the investment case deserves reassessment — regardless of where the stock is trading at the time. As long as the thesis holds, further weakness is an opportunity to add, not a signal to reconsider.

On the exit side, the central-case scorecard above — a PE around 30x on continued 20% EPS growth — is the closest thing to a defined target this analysis produces, without being a fixed price commitment. The honest answer is that the precise exit discipline (a specific multiple, a return threshold, or a combination of both) will depend on how the thesis evolves over the coming quarters, and particularly on how the OpenAI financing question resolves. What’s fixed is the criterion for reassessment described above; what remains open is the criterion for taking profits once the re-rating, if it happens, is underway.

Correlation

This position adds limited diversification at the factor level. Alphabet, already held in the portfolio, shares the same underlying exposure — hyperscaler capital expenditure, AI monetization timing, and sensitivity to the same sector-wide sentiment swings discussed throughout this article. Both stocks fell and would likely recover together on the same catalysts. The case for Microsoft here is not that it diversifies away from that exposure, but that it offers a better entry price on it: Alphabet trades close to its own historical average multiple, while Microsoft trades at the bottom of its ten-year range. This is a reallocation within a factor, not a rotation across factors.

Currency

Microsoft is dollar-denominated, with no natural currency hedge embedded in the business structure comparable to a foreign-currency debt offset. Currency exposure here is real and unmitigated at the position level.

Defensive / offensive

The position sits between the two. The downside case rests on a defensive foundation — Windows, Office, and Azure outside its OpenAI exposure remain fully verifiable and intact even in the bear scenario, with a balance sheet carrying more cash than debt. The upside case is more offensive in nature, requiring continued execution on an AI capital expenditure cycle whose full return is not yet visible in the numbers. The position is being sized accordingly: large enough to matter if the offensive case plays out, small enough that the defensive floor is what determines the outcome if it doesn’t.

Conclusion

Microsoft is not a broken business. It is a business where the market’s confidence in one specific, bounded relationship has not kept pace with everything else the company has demonstrated.

The core — Windows, Office, Azure outside its OpenAI exposure, a balance sheet with more cash than debt, operating margins that have expanded in every year shown in this analysis — is fully intact and independently verifiable. The part of the thesis that isn’t independently verifiable is narrow and identifiable: the financing trajectory of a single partner whose audited financial statements don’t yet exist.

EPS has compounded at roughly 18-20% a year. The stock has compounded more slowly. That gap closed the multiple to its lowest level in a decade, alongside Alphabet and Amazon — both of which corroborate, with their own numbers, that AI capital expenditure across the sector is being met by demand rather than outrunning it.

I don’t know when the OpenAI financing question resolves, or in which direction. What I know is that the price being asked today does not require it to resolve favorably — only for it not to be a catastrophe. That distinction is the entire trade.

Subscribe now

Image credits

Financial charts sourced from Fiscal.ai.

Important Disclosure & Disclaimer

All content published by JB Peter on this platform is strictly for educational and informational purposes. It does not constitute investment, financial, legal, or tax advice, nor does it represent a personal recommendation or solicitation to buy or sell securities. This research is operated by ORIACON (SASU) and reflects independent corporate analysis. Every reader must conduct their own independent research (Due Diligence) or consult a licensed professional before making any financial decision, as financial markets involve a high risk of capital loss. At the time of writing, ORIACON or the author HOLD shares in the company analyzed in this article. Following this publication, ORIACON and the author reserve the right to buy, sell, or modify positions in any security mentioned at any time, without prior notice to readers or subscribers.