There is a particular kind of stock that makes professional investors nervous in a way that has nothing to do with the business and everything to do with career risk. SK hynix is that stock right now.

Consider the setup. Here is a company that just reported a quarter with ₩79.3tn in revenue and ₩60.5tn in operating profit — a 76% operating margin. Its shares trade at roughly seven times trailing earnings and something closer to four times what the street expects for 2027. It sits at the physical bottleneck of the entire AI build-out. And it is down roughly 47% from its June high.

Every value screen in the world is flashing. And almost nobody wants to touch it, because everyone knows what memory is: the graveyard where capital goes to be destroyed at the top of a cycle. I’ve been running money long enough to have watched three of these cycles turn. I know the pattern. Buy at a high P/E when earnings have collapsed, sell at a low P/E when earnings are peaking.

I think that rule is about to be tested harder than it has been in twenty years — and I own the stock. Not because I believe the cycle has been abolished. I don’t. But because the market is now pricing SK hynix as though the cycle has already turned, while the physical evidence says supply cannot arrive before 2029. My case rests on one uncomfortable idea: you do not need earnings to grow from here to make excellent money. You only need them not to collapse, and the multiple to stop insulting the business.

That’s the whole thesis. A fair P/E of 10 on flat earnings gets you to roughly a double. Let me walk through why I think that’s the right number.

1. Product, Business Model, Brand and Moat

SK hynix makes memory. Two kinds. DRAM — the fast, volatile working memory that a processor thinks with — and NAND flash, the slower, permanent storage. Roughly 60–70% of revenue comes from DRAM and 30–35% from NAND. Morningstar

The business model is brutally simple: build a fab costing tens of billions, run it as close to full utilisation as physics allows, and sell bits. Historically that made memory the worst business in semiconductors — undifferentiated output, four players, and a price war every time someone added a fab.

What changed is HBM. High Bandwidth Memory takes DRAM dies, stacks them vertically, and bonds them next to the GPU. It is the reason an Nvidia accelerator can feed itself data fast enough to be useful. And critically, it is not a commodity. It is co-engineered with the customer, qualified die by die, generation by generation. HBM4 doubles the interface width to 2,048 bits and takes per-stack bandwidth to 1.5–2 TB/s, up from 896 GB/s–1.28 TB/s on HBM3E — this is a genuine architectural step, not a shrink.

SK hynix got here first, and the brand consequence matters more than people appreciate. In this industry “brand” is not advertising; it is being the supplier a hyperscaler will bet a $50bn data centre programme on. SK hynix was the leading global DRAM supplier in 2025 with 34.8% revenue share, and held 63.2% of the HBM market by revenue. It was the first company to mass-produce multiple HBM generations. When Nvidia allocates, that history is the collateral.

The moat is threefold. Capital: a leading-edge DRAM fab is a decade-long, hundred-billion-dollar commitment. Process: the 1c-node and hybrid-bonding know-how behind HBM4 is not licensable. Qualification: once a stack is designed into an accelerator, switching suppliers mid-generation is close to impossible.

And here is the structural point I keep coming back to. If a customer dislikes Nvidia’s chip design, TSMC will happily fabricate someone else’s. Memory does not work that way. There is no memory foundry. The design, the process and the fab are the same asset. That vertical integration is precisely what makes the moat harder — and what makes the downside, when it comes, harder too.

2. The Market

The memory market has stopped behaving like a memory market.

The mechanism is worth understanding because it explains everything else. HBM requires roughly three to four times the wafer area to produce the same number of bits as conventional DDR5, and because all three suppliers make HBM and commodity DRAM in the same facilities, every wafer moved to HBM removes three to four gigabytes of standard memory from the market. AI demand doesn’t just add demand — it cannibalises supply. That is why DDR5 module prices went vertical while server DRAM was already sold out.

The numbers are hard to internalise. Conventional DRAM contract prices rose 93–98% quarter-over-quarter in Q1 2026, taking global DRAM industry revenue up 81% sequentially to $97bn. By Q3 2026 TrendForce still saw the market as extremely tight, with contract prices moderating only to 13–18% QoQ and NAND to 10–15%. Moderating, not falling. BigGo FinanceTrendForce

On the demand side, Omdia projects both DRAM and NAND demand to compound at 19% annually through 2030. On supply, IDC put 2026 DRAM supply growth at just 16% and NAND at 17%, both well below the 20–30% historical norm, with new fab capacity not arriving in volume before late 2027 and meaningful relief pushed into 2028 or 2029.

Cyclicality has not been repealed. But the shape has changed. Three suppliers plus a fourth in China, demand growing high-teens, supply capped by cleanroom construction lead times, and — this is new — the majority of output locked into multi-year contracts rather than sold quarterly on spot. SK hynix now has long-term agreements with around ten key customers. That caps the upside in a squeeze. It also puts a floor under the downside, which is the part the market is currently ignoring.

Political exposure is real but asymmetric in SK hynix’s favour. Korea carries a North Korea tail risk that never fully goes away. It does not carry Taiwan risk. In a world genuinely worried about the Strait, a Korean memory maker with a growing US footprint is one of the few AI-chain assets that would arguably benefit.

3. Culture and Management

SK hynix’s culture is best understood through one decision. A decade ago, HBM was a niche, expensive, low-volume product that most of the industry regarded as an engineering curiosity. SK hynix kept funding it through the 2023 downturn, when the company was posting operating losses. Samsung — a conglomerate with a hundred competing priorities — did not commit with the same focus. That single allocation choice is worth several hundred billion dollars today.

CEO Kwak Noh-jung is unusually blunt for a Korean chaebol executive. Speaking to Reuters on the day of the Nasdaq listing, he said 2027 will be “the worst year in the industry’s history from the supply perspective,” and that customer demand will exceed the company’s supply capacity even beyond 2030. Management is telling you the shortage outlasts the decade. Yahoo Finance

More importantly, they are spending accordingly — and with discipline. On 7 August the board approved ₩54tn across two domestic fabs: ₩35.2tn for Yongin Y2 and ₩19.1tn for Cheongju M17, executing a master plan of ₩600tn for the Yongin cluster and ₩100tn for Cheongju. The company frames this explicitly as a structural transformation rather than a temporary supercycle — memory moving from component to core AI infrastructure. QuartzYahoo Finance

That’s the entrepreneurial part. The shareholder-alignment part came on 19 August. The board approved a ₩40tn repurchase and full cancellation of treasury shares — the largest such move in the history of Korean listed companies — explicitly stating that intrinsic value is underrepresented in the current share price, and raised the 2025–2027 shareholder return target from “within 50% of cumulative free cash flow” to “over 50%.” That’s about 24.07 million shares, roughly 3.3% of shares outstanding, running from 20 August to 19 November. SK hynixQuartz

4. Financials, Margins and Recent Developments

The 2023 loss year is instructive. SK hynix has been profitable in essentially every year of its modern history bar that one — and then came back to record 2025 operating profit of ₩47tn, double 2024. Then 2026 happened. Yahoo Finance

Q2 2026 (reported 29 July):

Revenue ₩79.3tn, up 51% QoQ and 257% YoY. Operating income ₩60.54tn, up 61% QoQ and 557% YoY, at a record 76% operating margin. Net profit ₩93.92tn — a net margin of 118%, lifted by large non-operating gains. Cash rose to ₩88tn, net cash expanded to ₩69.4tn, and borrowings fell to ₩18.6tn. Quartr

Two things need unpacking. First, that 118% net margin is not real operating performance: ₩63.3tn came from investment asset gains, largely the closing of the Kioxia stake sale. Second — and this is the reason the stock fell 9.6% on record numbers — it missed. Consensus was ₩84tn revenue and ₩64tn operating profit, and analysts attributed the shortfall to HBM4 shipments coming in below expectations, pushing revenue recognition into later periods. QuartzQuartz

Underneath, the mix shift is doing exactly what management promised. DRAM ASPs rose about 30% and NAND in the mid-50% range. Enterprise SSD sales doubled sequentially and Solidigm’s high-capacity eSSD revenue more than tripled. SOCAMM2 sales grew significantly and 1c-node shipments began in earnest. First-half revenue crossed ₩100tn for the first time. Investing.comQuartz

Guidance was orderly rather than heroic: DRAM shipments up around 10% QoQ in Q3, NAND up low single digits, with 2026 capex targeted at the high ₩40tn range — against ₩30.2tn spent in 2025. Investing.comQuartz

Return on equity, given ₩60tn of quarterly operating profit against an equity base that was a fraction of that eighteen months ago, is running at levels that have no useful comparison in the company’s history. Which is exactly why nobody trusts it.

5. Valuation: What a Fair Multiple Actually Looks Like

Here is where I depart from both the bulls and the bears.

The bulls model 2027 consensus and get a preposterous number. FnGuide has 2027 operating profit consensus at ₩391.82tn and 2028 at ₩399.13tn. On those figures the stock trades under four times earnings, and the average analyst target of ₩3,213,393 against a ₩1,596,000 share price implies over 100% upside. I don’t want to underwrite that. Peak-cycle consensus is the least reliable number in finance. The Herald BusinessInvesting.com

The bears apply the classic rule — low P/E on peak earnings means sell — and stop thinking.

My approach is deliberately dull. I assume earnings do not grow at all from the 2026 level.

Take 2026 operating profit of roughly ₩290tn. Tax it at a normal Korean rate and strip out the Kioxia gain entirely, and you land near ₩225tn of clean net income. Across roughly 730 million shares that is about ₩300,000 of EPS. Against Friday’s ₩1,596,000 close and a ₩1,166tn market cap, that is 5.3 times earnings. Google Finance

Now the only judgement call that matters: what is a fair multiple for this business?

Not 20. This is still a capital-intensive manufacturer in a cyclical industry with a rising Chinese competitor. Not 4, either — that is a multiple for a business in structural decline, and one that ignores a net cash balance sheet, an oligopoly structure, and a demand curve compounding at 19%.

I use 10. Ten times is what the market pays for a decent industrial with modest growth and no moat. Given the HBM franchise, I regard it as conservative rather than generous.

Ten times ₩300,000 is ₩3,000,000 per share — roughly 88% above today’s price. Realised over about eighteen months, with the 3.3% share cancellation adding a mechanical lift to per-share earnings, that is an annualised return in the region of 54%.

Note what is not in that number: no 2027 earnings growth, no HBM4E, no continuation of price increases, no NAND recovery, no Indiana. The entire return comes from the multiple normalising toward something merely unexciting. Every operational tailwind is upside to the case.

Note also what the model does not survive: a genuine earnings collapse. If 2027 EPS halves, fair value at 10x is ₩1,500,000 and you have lost money. That is the actual risk, and it is not small.

6. Risks: Why the Opportunity Exists

A stock does not fall 47% from its high without a reason. Several, in this case.

The share price story. SK hynix hit an all-time high of ₩2,987,000 on 25 June 2026. What followed was violent. The KOSPI fell 29% in a month, with memory names losing a third to a half of their value. On one session SK hynix closed 14.65% lower and Samsung fell more than 13%; another day’s selling was heavy enough to trigger a 20-minute trading halt on the KOSPI. The Q2 miss on 29 July added a further leg down. 000660 Stock Price and Chart — KRX:000660 — TradingView +3

Competitive erosion — the risk I take most seriously. Counterpoint puts Samsung’s Q2 HBM revenue share at 33%, up 12 percentage points in a single quarter, narrowing the gap to SK hynix’s 50% from 37 points to 17. In overall DRAM, Samsung led with 38% while SK hynix fell to 25%, down 14 points year on year, with Micron at 24% and CXMT at 10%. Samsung has begun mass-production shipments of HBM4. The monopoly premium is being competed away in real time. My thesis does not require SK hynix to hold 60% share. It does require the company to stay technically first, and that is now genuinely contested. Seoul Economic DailySeoul Economic Daily

China. CXMT has passed 10% of global DRAM and plans to expand monthly capacity to as much as 600,000 wafers by 2028, approaching SK hynix’s roughly 590,000, funded by a $9.8bn Shanghai offering that sent SK hynix and Micron shares tumbling on the day. The nuance most bears miss: CXMT is not undercutting on price — its 64GB server DDR5 module now costs more than Samsung’s equivalent, and when Apple tried to qualify CXMT as a fourth supplier, the effort collapsed on 5 August after CXMT quoted at or above oligopoly rates. Chinese capacity is currently reinforcing pricing, not breaking it. That will not be true forever. China’s CXMT Tops 10% of DRAM Market, Nears SK hynix Capacity - Seoul Economic Daily +3

Capex at the top. ₩54tn of new fabs plus a high-₩40tn capex year is exactly the behaviour that has ended every previous memory cycle. Management’s discipline language is reassuring; the cheques are what count.

Peak timing. The consensus has migrated toward a 2027 revenue peak. When brokers start agreeing about a top, positioning is already crowded — Michael Burry established a put position against Micron near $1,051.87 on 1 July 2026, shortly after it approached record highs following a roughly 700% rally. Yahoo Finance

Governance and geography. Chaebol structure, minority-shareholder history, currency translation, and the standing Korean peninsula risk. These are why Korean assets carry a permanent discount, and part of why a fair P/E here is 10 rather than 15.

7. Latest Developments

Three things have happened since the quarter that change the shape of the investment.

The Nasdaq listing. SK hynix completed its US debut on 10 July, raising $26.5bn in the largest-ever IPO by a foreign company in the United States, more than seven times oversubscribed. This matters beyond the capital. It broadens the shareholder base from a Korean market that structurally under-values its own champions into a US market that pays up for AI infrastructure. That is precisely the mechanism by which a re-rating happens. EciksFast Company

The buyback. Backed by ₩69tn of net cash, SK hynix moved first with the largest cancellation programme ever announced by a Korean-listed company, to be completed by November. Cancelling shares also lifts parent SK Square’s ownership ratio without further purchases, and offsets dilution from the 17.79 million new shares issued for the ADR offering — meaning the controlling shareholder’s incentives are aligned with buying back stock at these levels. Further returns are expected to be detailed alongside third-quarter earnings. Dividends or buybacks: Samsung, SK hynix divide on AI windfall - The Korea Herald +2

Indiana. On 27 August SK hynix broke ground on a $3.87bn advanced packaging project at Purdue Research Park in West Lafayette — the first HBM made in the United States. Cleanroom completion is targeted for October 2028 with mass production in Q3 2029, supported by up to $458m in CHIPS Act funding and $500m in loans, alongside an R&D partnership with Purdue and a stated ambition to make Indiana a key HBM production base by 2030. Korean wafers, American packaging, tariff insulation. The Korea HeraldEE Times

Q3 results are due on 27 October. Investing.com

8. Conclusion

The bull case does not depend on the AI boom accelerating. It depends on four things, in descending order of confidence.

One: supply physics. New cleanrooms take three years. Yongin Y2 comes online in June 2029. Whatever happens to demand, meaningful new supply cannot arrive before 2029. That is concrete and steel, not a forecast.

Two: bit growth offsets price decline. This is the part the cyclical bears get wrong. When ASPs eventually fall, they fall against a volume base that is expanding at high-teens rates, with HBM consuming three to four times the wafer area per bit. Revenue does not have to collapse when prices normalise — that is the arithmetic that broke every previous memory cycle and does not apply the same way here.

Three: mix, not price. The strategic thrust is away from commodity toward custom HBM, high-capacity enterprise SSD and AI-specific DRAM. Industry expectations put the B2B share of revenue at roughly 70% by 2027, far above prior cycles. Margin expansion from mix is durable in a way margin expansion from shortage is not. And the demand vectors keep multiplying — inference workloads, agentic systems, and physical AI in robotics, where every autonomous system needs local memory that doesn’t exist yet.

Four: the buyback compounds the discount. A company retiring stock at four times forward earnings converts market pessimism directly into per-share value. Every month the market stays sceptical, shareholders get a better deal.

Against that, the honest counter-case: Samsung is genuinely back, CXMT is genuinely scaling, capex is genuinely aggressive, and if the AI capex cycle breaks, a 76% operating margin has an extraordinarily long way to fall. Anyone telling you this is a low-risk position is not being straight with you.

My position sizing reflects that. This is not a core holding I would size like a compounder. It is a calculated small allocation where I think the market has mispriced a probability — pricing an imminent collapse into a business whose customers have contracted supply into 2027 and whose new capacity cannot arrive until 2029.

At five times earnings, with net cash, in an oligopoly, you are not required to be right about the boom. You are only required to be right that the bust is not next quarter. That, to me, is the cheapest genuine AI exposure available in public markets today.

Risk Disclaimer

This article represents my personal opinion and is intended for informational and educational purposes only. It does not constitute investment advice, a recommendation, or an offer to buy or sell any security. SK hynix is a position in the portfolio of the cost-efficient Haas Invest4 Innovation Fund (invest4.net), and I therefore have a financial interest in the security discussed. This constitutes a potential conflict of interest.

Share prices, valuation multiples and financial data cited reflect information available as of 4 September 2026 and may change without notice. Semiconductor memory is among the most cyclical industries in public markets; the figures cited reflect exceptional and possibly unsustainable conditions. Forward-looking statements are estimates and may prove materially wrong. Currency fluctuations between the Korean won and the euro or US dollar can significantly affect returns for foreign investors. Past performance is not indicative of future results. Every investment carries the risk of total loss. Please conduct your own research and consult a qualified financial adviser before making investment decisions.

investresearch.net · 3 September 2026

Introduction: A Business Model I Wish Existed in Frankfurt

Every so often I look at a Japanese small cap and think: why on earth does this not exist in Europe? Digital Grid is one of those cases.

Picture the German or Spanish power market. On one side you have a farmer with a 5 MW solar park, a municipal utility with a wind farm, a corporate rooftop installation. On the other side you have a mid-sized manufacturer who wants clean electricity, wants it cheaper than the incumbent utility offers, and increasingly wants to prove to auditors exactly which electron came from where. Between them sits an oligopoly of large integrated utilities taking a spread that nobody can really see.

Digital Grid built the thing that removes that middleman. The Digital Grid Platform, or DGP, is a marketplace where generators and corporate consumers transact directly. Not a power producer. Not a retailer in the traditional sense. A matching engine, a settlement layer, and a risk-management brain sitting on top of Japan’s liberalised electricity market — and it earns a fee on every kilowatt-hour that flows across it.

The concept itself came out of Professor Rikiya Abe’s laboratory at the University of Tokyo in 2008, where the idea was that electricity could be digitally “colour-coded” and traded freely, like packets on the internet. In 2008 that sounded like science fiction, because Japan’s grid was a one-way street from large thermal and nuclear plants. Nearly two decades later, with rooftop solar everywhere and corporate decarbonisation reporting now a board-level obligation, the idea has become a real business with real cash flows. Abe is Chairman and CTO. The company was incorporated in 2017 and listed on the TSE Growth Market in April 2025.

Here is why I am writing about it today. The stock IPO’d at ¥4,520 (¥904 split-adjusted), roughly doubled into August 2025, and has since fallen more than 60% over twelve months, touching an all-time low of ¥666 on 11 June 2026. It closed on 2 September at ¥770, giving a market capitalisation of ¥32.2bn — around €174m or $199m. That is a smallercompany than it was at IPO, despite revenue having grown from ¥1.2bn in FY2022 to a guided ¥6.6bn this year and operating margins above 40%.

The stock trades at 15.9x this year’s company-guided earnings. I think the fair multiple for this asset is 24x. Combine that with the earnings growth the business is capable of delivering into fiscal 2029, and I get an annualised return potential of roughly 34%.

The catch — and there always is one — is that Digital Grid is having an awkward year. Volume is growing nicely; unit prices are not. That transition from hypergrowth to digestion is exactly what has broken the share price, and it is exactly why the opportunity exists.

1. Product, Business Model, Brand and Moat

What they actually do

The group reports three segments, and it is worth separating them properly because they are very different animals.

Electricity Platform (Power PF). The core. Corporates come to DGP to procure electricity; generators come to sell it. Digital Grid matches them, handles the wholesale market interface, and takes a usage fee tied to the volume transacted and the contracted capacity. Crucially, the company has built what it calls a tailor-made procurement model: a customer can blend a fixed-price tranche with a market-linked tranche to suit its own risk appetite, rather than being forced into one or the other. This is the fee engine and the majority of group profit.

Renewable Energy Platform. The higher-growth sibling. This covers renewable power traded across DGP plus the environmental-attribute layer: Econohashi, an agency service for buying non-fossil certificates and environmental value, and RE Bridge, a virtual PPA auction site where a corporate can effectively post a request for the kind of renewable generation it wants and let generators bid. In the first nine months of the current fiscal year this segment grew revenue 54.6% and more than doubled segment profit, up 113.5%.

Other — balancing power and education. Two things live here. The balancing-power (adjustment) business, where Digital Grid acts as an aggregator optimising grid-scale batteries, earning recurring fees; and GX Navi, a decarbonisation training service for companies starting their green-transformation journey. This segment turned profitable during the current year.

How the money is made

Fee-based, volume-linked, asset-light — with one important qualification. Because Digital Grid stands between the wholesale exchange and the end customer, it carries balancing-group responsibility: it must match planned and actual half-hourly volumes, and the difference settles as an imbalance charge or credit. That is a real P&L line, it moves around, and it is also the single biggest reason a competitor cannot simply clone the front-end and win.

Gross margin runs at roughly 79%. Contracted capacity stood at 1,034 MW at the end of fiscal 2025, up 29.4% year on year. The whole operation is run by around 79 employees. That combination — near-800 megawatts of contracted load per hundred staff — is what a genuine platform looks like.

Brand

Let me be honest about scale. This is not a consumer brand and it never will be. Digital Grid is known inside a specific Japanese ecosystem: corporate energy managers, sustainability officers, independent generators, and the GX policy community. Within that world the reputation is strong and getting stronger, helped by two things. First, Toshiba is the largest shareholder with roughly 12% — a strategic anchor that opens doors in an industry where a startup would otherwise be treated as an unserious counterparty. Second, the company has positioned itself early on the reporting side: when the GHG Protocol’s Scope 2 guidance moved towards hourly matching, Digital Grid was out in front with hourly visibility of renewable procurement. In a market where the buyer’s real problem is increasingly proving what they bought, that is brand-building of the most useful kind.

The retail investor community in Japan knows the name too, though for less flattering reasons this year — the discussion boards and the Japanese finance corner of X have been a mix of frustrated IPO buyers and value hunters, with sentiment recently tilting bullish again as the price stabilised in the high ¥700s.

Moat

I would rate this a genuine, narrow, deepening moat. Four planks:

1. Two-sided network effects. More generators means better prices and more choice for buyers, which pulls in more buyers, which makes the platform the obvious place for a generator to list. Contracted capacity compounding at ~30% is the evidence.

2. Imbalance risk management. Anyone can build a matching website. Very few can forecast demand and generation accurately enough, half-hour by half-hour, to avoid being bled dry by settlement charges. Digital Grid has pushed AI-based forecasting into production specifically to shrink this variance. This is the operational barrier.

3. Regulatory and operational plumbing. Licences, JEPX membership, credit lines, balancing-group status, system integrations. Slow and expensive to replicate.

4. Switching costs. A corporate that has built its Scope 2 reporting workflow around DGP data does not casually move.

What it is not is unassailable. The large incumbent utilities could compete on price if they chose to, and other aggregators exist. But they would be cannibalising their own retail margins to do it, which is the classic innovator’s dilemma that keeps a company like this alive long enough to matter.

2. Market: Big, Structurally Growing, Politically Exposed, Moderately Cyclical

Japan liberalised retail electricity in 2016. The entire national market is in the order of ¥20 trillion. Digital Grid’s revenue of ¥6.6bn against that number tells you the runway is essentially unlimited in mathematical terms — penetration is a rounding error.

Three structural drivers make me comfortable this is not a fad:

Decarbonisation is now mandatory, not optional. Japan’s latest energy plan pushes renewables towards 40–50% of the power mix by 2040, and the GX framework has made corporate emissions disclosure a normal part of doing business. Every large Japanese corporate now needs a renewable procurement strategy, and most of them do not want to build one themselves.

Distributed generation needs a distributed marketplace. As generation shifts from a handful of large plants to thousands of small ones, the coordination problem grows combinatorially. That is a software problem, not a turbine problem.

Intermittency creates a storage and balancing market from nothing. Grid-scale batteries in Japan are moving from pilot to build-out, and someone has to optimise them. Digital Grid is positioning as that someone.

On political interference I have to be candid: this is an infrastructure-adjacent, regulated market. Capacity market rules, feed-in structures, non-fossil certificate mechanics and grid tariffs all change by administrative decision. That is a permanent feature, not a passing risk. The offsetting point is that the direction of policy travel is strongly in Digital Grid’s favour — liberalisation and decarbonisation both increase the value of a neutral marketplace.

On cyclicality: less than you would fear, more than the company’s fee model suggests. Industrial electricity demand is reasonably defensive. But wholesale power prices are volatile, and — as this year has demonstrated — the unit priceDigital Grid earns per transaction moves with them. There is also clear intra-year seasonality: the August–October first quarter carries peak summer demand and is the strongest of the year, while the February–April third quarter is the seasonal trough. Management stated at the last quarterly Q&A that geopolitical tension and a fuel price spike would not derail the current outlook, since the model is largely insulated from directional wholesale price risk. I believe them at the operating level. I do not believe it protects the share price from sentiment.

3. Culture and Management: Founder Energy, Wall Street Discipline

CEO Yusuke Toyoda is 39. He started at Goldman Sachs Japan in 2012, moved to the private equity firm Integral in 2016, joined Digital Grid in February 2018 and took the CEO seat in July 2019 at the age of 32. He holds roughly 5.2%of the company.

That biography matters more than it usually does. This is a business whose central operational challenge is managing a book of forward power positions and imbalance exposure while extending working capital to customers — genuinely a trading and risk job wrapped in a software skin. Having a former investment banker with buyout-firm experience running it is a better fit than a pure engineer would be. And Abe, the academic who invented the concept, remains as Chairman and CTO, so the technical vision has not been diluted.

The stated mission is the “democratisation of energy” — connecting a world without energy constraints to the next generation. The internal values are, charmingly, Be on the Edge, Far Together and Stay Gold. Make of that what you will; I read it as a startup that has not yet been institutionalised into blandness.

On strategy, the medium-term plan through fiscal 2028 is unusually specific for a Japanese small cap and unusually shareholder-oriented:

• ROE sustained above 20%

• Operating margin sustained above 40%

• CAGR in total electricity handled above 30%

• ¥10bn invested in storage batteries over three years, targeting 40 MW by FY2028

Toyoda has also publicly stated an aspiration to reach a ¥100bn market capitalisation. Against today’s ¥32bn, that is a triple, and I generally like managers who name a number in public and then have to live with it.

The one thing I would flag on capital allocation: the battery build-out is a deliberate move away from asset-light. Management is spending real capex to own balancing assets rather than only aggregate third-party ones. I think this is defensible — it secures a position in a market that is forming right now — but it changes the character of the balance sheet, and it deserves monitoring rather than applause.

4. Financials, Margins and Recent Developments

The growth record

Five times revenue in four years, with operating profit going from essentially zero to ¥2.7bn. That is the track record the market fell in love with in 2025.

Then look at the last column and you see the problem. Growth has decelerated sharply. This is the single most important fact about the stock right now, and it deserves an honest explanation rather than a dismissal.

What is actually happening under the surface

The deceleration is a price effect, not a demand effect. In the fourth quarter of FY2025 the per-transaction usage fee dropped 17.7% year on year, and management guided the current year on the assumption that average unit prices would keep falling even as handled volume rose more than 20%. The original FY2026 guidance therefore called for a 13.8% decline in operating profit — which is precisely the moment the market stopped paying a growth multiple.

The nine-month results published on 11 June 2026 told a better story: revenue of ¥5,107m (+6.6%) and operating profit of ¥2,447m (+3.1%), with the renewable platform up 54.6% and the other segment swinging into profit. Good enough that management raised full-year guidance on the same day:

• Revenue: ¥6,281m → ¥6,595m (+5.0% vs prior guidance)

• Operating profit: ¥2,363m → ¥2,836m (+20.0%)

• Ordinary profit: ¥2,128m → ¥2,660m (+25.0%)

• Net profit: ¥1,476m → ¥1,919m (+30.0%)

So the year that was supposed to see profits fall 14% will now see them rise. That is a materially different picture from the one priced in.

These are software margins on an energy revenue line. A 43% operating margin on a business that intermediates physical electricity is remarkable, and it is the clearest evidence that the fee model is genuinely platform economics rather than a repackaged utility.

The balance sheet holds more cash than debt, which is important given the structural working capital drag: Digital Grid pays JEPX before it collects from customers, so growth consumes cash. Management has arranged committed credit facilities with multiple banks specifically to handle this. Those facilities carry net-asset covenants, which is worth knowing.

Recent developments

Three moves in the last few months tell you where this is going:

• 1 July 2026 — entry into the low-voltage corporate segment, with a dedicated subsidiary, DG Life, established to run it. This widens the addressable customer base from large industrial accounts down to smaller businesses across the whole country.

• 7 August 2026 — a new matching service for grid-scale storage assets, extending trading support to batteries from the development stage through to operation.

• June 2026 — an upgrade to RE Bridge letting corporate buyers post specifications for the renewable generation they want, plus a response to the revised GHG Protocol Scope 2 standard giving hour-by-hour visibility of renewable procurement.

Meanwhile the first grid-scale battery facility is already operating, with the bulk of storage revenue expected to arrive in fiscal 2027 and 2028. The balancing business runs on proprietary automation and a cross-manufacturer battery gateway — meaning Digital Grid can manage assets regardless of who built the battery, which is precisely the right architecture for an aggregator.

5. Valuation: A Fair PE of 24 and 34% Annually

My valuation approach here is the one I use across the portfolio: establish what the business should earn three years out, apply a defensible normalised multiple, and let the implied annual return decide whether the stock is interesting.

Where we start. At ¥770, with company-guided EPS of ¥48.41 for the year ending July 2026, the stock trades at 15.9x. Price to book is 3.06x. There is no dividend.

What the fair multiple should be. I set the fair PE at 24x. My reasoning:

• A two-sided marketplace with 79% gross margins, 43% operating margins and a high-teens-to-low-twenties ROE would attract 30x or more in almost any Western market.

• Revenue quality is high and recurring in character — contracted capacity, not one-off projects.

• Against that, I discount for genuine unit-price volatility, regulatory dependency, single-country exposure, a working-capital-hungry model, and a TSE Growth listing with essentially no sell-side coverage.

• 24x is roughly a 50% premium to the Japanese small-cap average and a meaningful discount to what a comparable Western platform would fetch. It is a multiple I would be comfortable defending in a drawdown.

The earnings bridge. To justify a 34% annual return over three years I need EPS to compound at roughly 17% per annum from ¥48.41 in FY2026 to about ¥77 in FY2029. Given that management is targeting more than 30% annual growth in electricity handled through FY2028, that the renewable platform is compounding at 50%+, that storage revenue is only scheduled to arrive from FY2027, and that the low-voltage market has just opened, 17% strikes me as the conservative case rather than the aggressive one. It assumes unit prices keep grinding lower and volume does the heavy lifting.

The result:

Two sensitivities worth holding in mind. If the market never re-rates and the stock stays permanently at 16x, I still earn the 17% earnings growth — a perfectly respectable outcome, and my downside case. And if this becomes a hype stock again — which it demonstrably can, given it traded above ¥2,000 thirteen months ago — a 30x multiple on ¥77 puts the shares near ¥2,300, or something close to 44% annually. I do not underwrite that outcome. I simply note that the distribution is skewed to the upside.

6. Risks: Why the Price Fell and Why the Opportunity Exists

The share price chart is brutal and it deserves a proper explanation, because if I cannot explain the decline I have no business claiming the market is wrong.

The IPO was priced for perfection and then some. Listing in April 2025 into a market that loved Japanese growth stories, the shares roughly doubled by mid-August 2025. At the peak, investors were paying a very large multiple for a company that had just grown revenue 75%.

The FY2025 results were the turning point. On 11 September 2025 the company reported a superb year — and simultaneously guided the next one to a 13.8% profit decline on falling unit prices. Growth investors do not tolerate that. The de-rating started that day and ran for nine months, bottoming at ¥666 on 11 June 2026. The day after the raised guidance, the stock went limit-up 14.35%. That is a shareholder register in transition, violently.

Retail leverage is still an overhang. The margin buy balance stands at roughly 2.48m shares against 41.8m outstanding, a margin ratio above 6x. Every rally meets sellers who bought higher and financed the position. This mechanically caps rallies until the balance works down.

No coverage, no institutions. The stock is formally uncovered by analysts. Roughly 71% of the register sits with retail and non-institutional holders. There is no natural buyer to arbitrage a valuation gap.

Now the fundamental risks, which are the ones that actually matter for a three-year holding period. Management’s own disclosure is refreshingly direct, and I would highlight these:

1. Unit price and churn risk. If wholesale prices spike, customers on fully market-linked plans get hurt and may renegotiate or leave. Fee income falls and retention spending rises.

2. Imbalance settlement volatility. Forecast error translates directly into P&L swings and cash flow noise. This is the operational heart of the business.

3. Working capital and funding. Payments to JEPX precede customer collections. Growth burns cash. If credit lines tighten, growth is constrained.

4. Covenant sensitivity. The committed facilities require minimum net asset levels; a bad year plus heavy battery capex could reduce financial flexibility at the worst moment.

5. Regulatory change. Capacity market design, certificate mechanics and system reform all evolve. Each change means system rework and possible margin impact.

6. Single-platform technology risk. DGP is self-developed. A serious outage, bug or cyber incident would be a direct hit to trading continuity and reputation.

7. Battery execution. ¥10bn is a lot of capex for a company this size. Delays or disappointing returns would push out the FY2027–28 revenue contribution the plan depends on.

8. Revenue concentration. The Electricity PF fee stream still dominates group profit. If it stalls, the other segments cannot yet compensate.

The honest summary: this is a good business going through a price-mix reset, owned by a shareholder base that bought it as a momentum story and is still liquidating. Those two things together produced a 60% drawdown in a company whose profits are now guided up. That gap between narrative and numbers is where the return lives.

7. Latest News and Earnings

The most recent reported numbers are the third-quarter figures from 11 June 2026 covering the nine months to 30 April: revenue ¥5,107m (+6.6%), operating profit ¥2,447m (+3.1%), with the renewable platform up 54.6% on revenue and 113.5% on segment profit, and the other segment turning positive. Full-year guidance was raised the same day to ¥6,595m revenue, ¥2,836m operating profit and ¥1,919m net profit.

The accompanying investor Q&A was, to my eye, the most reassuring document the company has published since listing. Management addressed the geopolitical and fuel-price question head on and argued that the model is largely insulated from directional wholesale price risk. They reported surging interest in the tailor-made procurement structure that blends fixed and market-linked supply — which is exactly the product you would want to be selling into a volatile price environment, because it converts customer anxiety into a reason to use the platform rather than a reason to leave it.

They also confirmed the storage roadmap: ¥10bn towards 40 MW by fiscal 2028, first grid-scale facility already live, meaningful revenue from fiscal 2027–2028, and a balancing business built on proprietary automation plus a cross-manufacturer gateway.

Since then, the flow of announcements has been steady rather than dramatic: the low-voltage corporate launch and the DG Life subsidiary from 1 July, the grid-scale storage matching service on 7 August, the RE Bridge functionality upgrade, and the Scope 2 hourly-matching response.

The near-term catalyst is one week away. Full-year results for the year ending July 2026 are scheduled for 10 September 2026. Two things matter in that release: whether the raised guidance was met or beaten, and — far more important — what management guides for fiscal 2027. Last year’s guidance day was the moment the stock broke. This year’s is the moment it could stop being broken, particularly if the FY2027 outlook reflects the low-voltage expansion and the first real storage contribution. I would rather own this into that print than after it.

The shares closed at ¥770 on 2 September, down 3.4% on the day and drifting on modest volume, with the year-to-date range running from ¥666 to ¥1,050.

8. Conclusion: What Has to Go Right

Digital Grid is the cleanest expression I have found of an idea I keep returning to — that the energy transition will be won as much by software that coordinates electrons as by hardware that generates them. It is a marketplace with network effects, 79% gross margins, 43% operating margins, a founder-adjacent CEO with real skin in the game, a strategic anchor shareholder in Toshiba, and a three-year plan with numbers attached to it.

It trades at 15.9x earnings because it had one difficult year on pricing and because its post-IPO shareholder register has been unwinding for twelve months. Those are real problems. Neither of them is a problem with the business.

I see three ways this works:

The base case — the deceleration was temporary. Volume compounds at the 20–30% management is targeting, unit prices stabilise, and earnings grow around 17% annually. The multiple drifts back toward 24x as the growth-scare narrative fades. That is roughly 34% per year and it is the case I underwrite.

The storage case — a second engine appears. The ¥10bn battery programme and the aggregation business deliver from fiscal 2027, adding a recurring-fee revenue stream that is structurally less exposed to platform unit-price pressure. Earnings growth reaccelerates above my assumption and the multiple expands because the business looks more durable. This is the case where the fair PE I have used turns out to be too low.

The hype case — a small cap in a hot theme. Japanese grid-scale storage and AI-driven power demand are exactly the kind of themes that produce violent re-ratings in a 41-million-share float with a 6x margin ratio. The stock traded above ¥2,000 in August 2025 on worse fundamentals than it has today. I do not model this and I would be selling into it, but it is a real feature of the risk-reward.

The bear case is equally clear: unit prices keep compressing faster than volume grows, imbalance losses widen in a volatile market, the battery capex disappoints, and this settles as a 15x business growing at single digits. In that scenario I lose time rather than a great deal of capital — the balance sheet is sound, cash exceeds debt, and the company is profitable throughout.

What has to go right is not complicated. Volume has to keep compounding, and management has to stop the price line falling faster than the volume line rises. The nine-month numbers and the June guidance raise say they are winning that fight. The share price says nobody has noticed. On 10 September we find out who is right.

Risk Disclaimer

This article reflects my personal opinion and is intended for information and educational purposes only. It is not investment advice, not a recommendation to buy or sell any security, and not an offer or solicitation of any kind. Nothing here takes account of your individual circumstances, objectives, financial situation or risk tolerance.

Equity investing involves substantial risk, including the total loss of capital. Small caps listed on the TSE Growth Market are illiquid, volatile and can move violently on single news items. Digital Grid in particular has fallen more than 60% over the past twelve months and carries elevated business risk from power price volatility, imbalance settlement exposure, working capital intensity, financial covenants, regulatory change, technology dependency, execution risk on its battery investment programme and revenue concentration in a single segment. Investors outside Japan additionally bear yen currency risk and higher transaction costs.

All figures are drawn from company disclosure and publicly available market data as of 3 September 2026 and may contain errors or become outdated without notice. Forward-looking statements, projections, target prices and return estimates are assumptions, not forecasts and certainly not promises. Past performance is no indicator of future results.

Conflict of interest disclosure: Digital Grid Corporation (TSE: 350A) is a holding in the cost-efficient Haas Invest4 Innovation Fund (invest4.net), which I manage. I therefore have a financial interest in the security discussed and may buy or sell it at any time without notice or updating this article. Please form your own view and consult a licensed adviser before making any investment decision.

Philipp Haas · investresearch.net

I have been following Amazon for a long time. I covered the stock as an analyst at a large German fund house, where the position sizes were big enough that being wrong was not an academic problem. My wife worked at Amazon, so consider that disclosed up front. And I have written about the company before, at a point when the honest conclusion was: this is probably the best business on the planet, and you cannot value it with any conventional tool.

That second part is no longer true, and that is the whole story of this article.

For roughly two decades Amazon was a growth machine that refused to show earnings. You valued it on revenue, on gross merchandise value, on a sum of the parts, on faith. Today Amazon closed at $254.92, up from a 52-week low of $196 and about 11% below the all-time high of $284 set on 3 August. The company just reported a quarter with $200.6 billion in revenue, $27.5 billion in operating income, and a cloud division growing at 36.7% while carrying a 39.4% operating margin. You can now build a normal earnings model on Amazon. That is a genuine regime change, and most investors have not repriced their mental model to match.

The investment case in one paragraph: Amazon has converted itself from a low-margin retailer with an attached cloud business into an infrastructure and advertising company with an attached retailer. AWS is reaccelerating for the fifth consecutive quarter on the back of AI demand, advertising is compounding in the mid-twenties with software-like incremental margins, and the company now designs its own AI silicon at a $25 billion run rate. Against roughly $8 of normalised 2026 earnings power, the stock trades near 31 times. On a fair P/E of 30 and a mid-teens earnings CAGR, I get to roughly 13% annualised return over five years. That is not a bargain. It is a fair price for a compounding asset with several options attached that nobody is currently paying for. I hold it, and I have used the recent weakness to stay at full weight rather than trim.

1. Product, Business Model, Brand and Moat

Most people still describe Amazon as “the online shop.” That description was already outdated ten years ago and is now close to useless.

Retail and marketplace. The first-party retail business is the one everybody knows and the one I like least. You buy inventory, you carry the markdown risk, you eat the returns, and the margins are thin. The interesting move was turning that shop into a marketplace. Third-party sellers now carry the inventory risk, Amazon takes a cut of the transaction plus fulfilment fees, and the assortment gets broader without a single euro of additional working capital. Same customer experience, radically better economics, far less capital intensity. North America revenue grew 16% to $116.2 billion last quarter and international grew 15% to $42.2 billion, which is a remarkable number for a business this mature.

AWS. This is where the value sits. Amazon built cloud infrastructure for its own retail needs and then sold the spare capacity, which is one of the great accidental strategic moves in corporate history. Microsoft had the enterprise relationships already and simply moved existing customers into the cloud. Amazon had to win every account from scratch and still ended up the largest provider. AWS did $42.2 billion in the quarter, a $169 billion annualised run rate, with a backlog of $496 billion that grew by $130 billion in three months.

Advertising. The segment most retail investors still underweight in their models. When you search on Amazon, the sponsored placements at the top are an auction, and that revenue lands close to the operating line. Advertising grew 26% to $19.8 billion in the quarter. Amazon is now the third-largest digital advertising platform in the world behind Google and Meta. It is the single highest-margin dollar in the business.

Prime. The subscription is the flywheel. Once you pay the annual fee, ordering anywhere else feels like leaving money on the table. Prime Video, delivery, music and photo storage all raise the switching cost, and the ad-supported Video tier turned a cost centre into a second advertising surface.

The moat. Ask the replacement question honestly. Who replaces AWS for an enterprise that has spent six years building on its primitives? Who replaces a fulfilment network of that density? Who replaces a search bar that a hundred million shoppers treat as the default starting point for buying anything? The moat is not one wall. It is scale economics, switching costs, network effects between buyers and sellers, and a brand that survives price comparison rather than depending on it. In Western e-commerce there is simply no second global platform. Competition exists inside verticals such as fashion and furniture, and Alibaba and PDD dominate at home in China, but there is no global alternative.

2. Market

Amazon has the largest addressable market of any company I follow. There is an old joke among analysts that I still find accurate: if Amazon is in your business, that is bad, because Amazon is in your business. If Amazon is not in your business, that is also bad, because it means your business is not worth entering.

Cloud is the biggest of the pools. Global cloud infrastructure spending is compounding in the thirties, AI-related workloads have grown from roughly 8% of cloud spend in 2023 to around 19% now, and the market as a whole is on a path toward a trillion dollars. AWS is no longer twice the size of Azure. Depending on which tracker you use, AWS sits at roughly 28% to 31% of global cloud infrastructure, Azure at 21% to 25%, and Google Cloud at 11% to 14%. And Google is not fading. Google Cloud grew 82% year over year last quarter and Azure has printed roughly 40% for several quarters running. AWS is the leader by revenue and by profit, but the growth-rate leadership has genuinely moved.

Advertising is a structurally attractive pool because Amazon owns the transaction, not just the impression. Advertisers can see conversion directly. That is why budgets keep migrating.

Retail remains the most cyclical and most politically exposed piece. Tariffs, consumer sentiment and freight costs all pass through it. Cloud and advertising are far less cyclical but far more politically exposed in a different way, through antitrust and consumer-protection enforcement. That combination matters: this is not a business that is free from political intervention, and pretending otherwise would be sloppy.

3. Culture and Management

Soft factors do more of the work in long-term returns than most models admit, and this is the part of the Amazon story that has changed the most since I last wrote about the company.

Jeff Bezos handed the CEO role to Andy Jassy in 2021 and is now Executive Chairman. Anyone still writing about Bezos as Amazon’s operating leader is working from a stale script. Jassy built AWS from nothing into the most profitable division in the company, which means the person running Amazon today is the person who built its best business. That is about as good a succession outcome as a large-cap can get.

The culture is codified in 16 Leadership Principles, not the twelve of the earlier era. What matters is not the count but that these are operationally live. Candidates are interviewed against them. Employees are reviewed against them. Frugality, hiring people better than yourself, disagreeing and committing, being right a lot, ownership. These are not agency-produced posters. Most companies write values and ignore them; Amazon writes values and grades people on them.

The entrepreneurial question has a harder edge in 2026. Jassy has cut aggressively, with more than 57,000 corporate roles removed since 2022, roughly 16% of the corporate workforce, framed explicitly as removing bureaucratic layers so the company can operate like “the world’s largest startup.” He has also said plainly that AI efficiency gains will keep shrinking corporate headcount. Reporting suggests morale has taken a hit, and I do not want to gloss over that. But the strategic logic is consistent: reallocate operating expense into capital expenditure, and spend the capital on compute.

That is the thing to understand about this management team. They are willing to depress reported free cash flow for years to build something structural. They did it with fulfilment centres. They did it with AWS. They are doing it now with data centres and silicon. A management team that will absorb short-term pain for long-term position is exactly what a long-term holder wants, and it is exactly what makes quarterly-focused investors uncomfortable.

4. Financials, Margins and Recent Developments

Here is where the old narrative needs the biggest update. Amazon is not an unprofitable growth story any more.

Revenue is running above an $800 billion annual pace and grew 20% year over year in Q2 2026, which is extraordinary at this scale. Operating income rose 43% to $27.5 billion. Operating margin expanded from 11.4% to 13.7%, because revenue grew 19.6% while operating expenses grew 18.0%. Gross margin came in near 35.9%.

The mix shift is doing the heavy lifting. AWS produced 21.1% of revenue and 60.5% of operating income. A year earlier those figures were 18.4% and 53.0%. Every point of mix moving toward cloud changes the earnings profile of the entire company. AWS segment operating income grew 63% to $16.6 billion, with margin expanding roughly 650 basis points to 39.4%.

Now the number that requires care. Reported net income was $62.6 billion, or $5.75 per share, which more than tripled year over year. That figure includes $53.4 billion of non-operating pre-tax income, driven primarily by an upward revaluation of Amazon’s stake in Anthropic. That is a real economic gain, but it is a mark, not cash flow, and it will not repeat. Trailing twelve-month EPS is $12.60 including it and roughly $6.73 excluding non-recurring items. Reported ROE of about 30.6% is similarly flattered; consensus has ROE settling near 16.6% in three years.

If you take the reported P/E of roughly 20 at face value, you will conclude Amazon is cheap. It is not. That is the single most common analytical error being made on this stock right now.

The other number that deserves attention is capital expenditure. Trailing twelve-month purchases of property and equipment reached $169 billion, up 64%, and management raised the 2026 cash capex guide from about $200 billion to about $220 billion, citing higher memory costs and continued AI and cloud infrastructure investment. The consequence is that trailing free cash flow has gone negative, at roughly minus $7.6 billion. On EBITDA the picture stays comfortable, in the low-to-mid twenties as a percentage of revenue, because depreciation and amortisation are now running at a high single-digit share of revenue and climbing with the build-out. But anyone screening on free cash flow yield will discard this stock immediately, and that is precisely why the opportunity exists.

5. Valuation: A Fair P/E of 30

I want to keep this simple, because complexity in a valuation model usually hides a weak assumption.

I use a fair P/E approach. For Amazon I set the fair multiple at 30. That is a premium to the market and a discount to what the market has historically paid for Amazon, whose ten-year average P/E sits near 60. I think 30 is defensible for a business with mid-teens earnings growth, an expanding mix toward 39%-margin cloud revenue, an advertising business compounding at 26%, and durable competitive positions in three separate large markets. I would not pay 45 for it. I do not think 20 is realistic on the downside for a franchise of this quality.

The inputs:

• Share price today: $254.92

• Normalised 2026 EPS (excluding the Anthropic mark): approximately $8.00

• Consensus normalised EPS: roughly $10.25 for 2027 and $13.22 for 2028

• Current normalised P/E: roughly 31 times

So on today’s price, Amazon trades essentially at my fair multiple. There is no valuation cushion and no valuation excess. That is a much more comfortable place to be than at any point in the last decade.

The return then comes from earnings growth. If I assume roughly 14% annual EPS growth over five years, which is below the 2026-to-2028 consensus trajectory and therefore deliberately conservative, earnings power reaches roughly $15.30 per share by 2031. Applying the fair multiple of 30 gives a fair value near $460.

From $254.92, that is a total return of roughly 80% over five years, or approximately 13% per year.

Thirteen percent annually, from a business of this quality, with the AI infrastructure cycle as a tailwind rather than a headwind, is an outcome I will take. It does not require multiple expansion. It requires only that Amazon keeps executing at a rate below what its own recent results imply. Any multiple above 30, or any upside surprise from advertising or silicon, is a bonus rather than a requirement.

6. Risks and Why the Opportunity Exists

A stock 11% off its high, six weeks after a blowout quarter, is telling you something. Four things explain the price action.

The capex question. Two hundred and twenty billion dollars of cash capital expenditure in a single year is an enormous bet. If AI demand digests rather than compounds, Amazon will own a great deal of depreciating hardware and a badly impaired earnings profile. Negative free cash flow removes Amazon from a large number of institutional screens.

Earnings quality. The $53.4 billion Anthropic revaluation cuts both ways. It made the headline print spectacular and it can reverse. Marks on private AI holdings are not a stable asset. Investors who correctly discount it arrive at a normalised P/E near 31, which is a very different stock from the one the headline suggests.

Regulation, which is now live rather than theoretical. On 31 August the FTC, joined by 22 state attorneys general, sued Amazon in the Western District of Washington, alleging it secretly and systematically inflated prices in its search advertising auctions for more than 1.2 million advertisers going back to 2018 or 2019, potentially generating over $20 billion. Amazon called the complaint misguided, said it fundamentally misunderstands how advertisers operate, and stated that its auction systems saved advertisers $8 billion between 2021 and 2025. This is the third major federal action against the company. The stock fell roughly 2.8% on the filing. The market is right to care, because advertising is the highest-margin line in the business, but the discount being applied looks larger than the plausible economic damage.

Competitive pressure in cloud. Azure at 40% and Google Cloud at 82% growth are not rounding errors. AWS is reaccelerating, which is the correct rebuttal, but the days of assuming permanent AWS dominance are over.

I would add one structural risk that gets discussed too little. Amazon is capital-intensive and cyclical in retail, politically exposed in cloud and advertising, and now dependent on a small number of very large AI counterparties for a meaningful share of its incremental growth. Anthropic and OpenAI have both made multi-gigawatt commitments. Concentration of that magnitude is a feature until it becomes a risk.

7. Latest News and Earnings

The second quarter, reported on 30 July, was the strongest operating print Amazon has delivered in years. Revenue of $200.6 billion crossed $200 billion for the first time and beat consensus near $196.5 billion. AWS at $42.2 billion beat expectations of $40.5 billion decisively, with analysts having modelled 31% growth against the 36.7% delivered. It was AWS’s fastest growth in 18 quarters, back when the division was less than half its current size, and the fifth consecutive quarter of acceleration. Shares jumped over 9% after hours.

Jassy disclosed on the call that the AI business and the chips business each exceeded a $25 billion annualised run rate, both growing at triple-digit rates. Management said 2027 capacity is already largely reserved with some 2028 capacity spoken for, and that they expect to keep adding significant data centre capacity for the next two to three years.

Guidance for Q3 is $197 billion to $202 billion in revenue with operating income of $22.5 billion to $26.5 billion, against $17.4 billion a year ago. The optically soft top-line growth reflects Prime Day shifting into Q2 this year and an 80 basis point FX headwind. Excluding the Prime Day effect, growth would have been nearly 400 basis points higher.

Elsewhere: the satellite programme formerly called Project Kuiper has been rebranded Amazon Leo, is in enterprise beta with partners including Verizon, AT&T, Vodafone, JetBlue and NASA, and has around 361 production satellites in orbit, well behind the FCC’s July 2026 milestone, with an extension requested and 22 additional launches contracted. Alexa has been rebuilt as Alexa+ with conversational and agentic capabilities, with the legacy voice-assistant architecture retired. Zoox continues expanding its robotaxi footprint. AWS is opening a region in Saudi Arabia backed by a $5.3 billion investment. And in the week since the FTC filing, at least one major bank has told clients to buy the resulting selloff.

Sentiment on X and across the investment newsletters has swung hard in both directions over five weeks, from euphoria on the AWS acceleration to something closer to regulatory panic. That gap between operating reality and narrative is usually where the money is made.

8. Conclusion

Amazon is, in my view, still the best business in the world, and for the first time in its history you can defend that view with an earnings model rather than a story.

What makes it interesting today is not the retail business, which is fine, or Prime, which is excellent, or even AWS, which is the crown jewel. It is the combination of three separate compounding engines inside one holding structure, wrapped in a culture that reliably converts capital into new engines. You are buying a diversified, internationally positioned, multi-model business with better governance and better capital allocation than the average index constituent. In a portfolio it functions almost like a very high-quality mini-ETF, and it did so long before the AI cycle arrived.

The specific option I think is most underpriced is the silicon business. Amazon designs Trainium and Graviton in-house through Annapurna Labs and does not sell the chips. It rents the compute. That vertical integration gives AWS a structural cost advantage over any competitor buying merchant GPUs. Anthropic runs Claude on more than a million Trainium2 chips through Project Rainier and has committed to up to five gigawatts of current and future Trainium capacity alongside a commitment of more than $100 billion to AWS technologies over ten years. OpenAI has committed roughly two gigawatts. Trainium3 UltraServers reach rack-scale performance broadly comparable to Nvidia’s flagship configuration at a materially lower total cost of ownership, and Trainium4 is already in development. That is a $25 billion run-rate business growing at triple digits, embedded inside a company valued as a retailer.

The second underpriced option is advertising, precisely because the FTC lawsuit has made investors nervous about a segment that is compounding at 26% with the best incremental margins in the group.

So where does that leave me? At $254.92, Amazon trades at roughly my fair multiple of 30 on normalised earnings, offering something close to 13% annualised over five years on conservative growth assumptions. I am not going to tell you this is a bargain. Amazon has almost never been a bargain, and waiting for one has cost investors more money over the last fifteen years than any other single decision.

There are perhaps five or six companies where I think the greater risk is not owning them rather than owning them, especially when you are measured against a benchmark. Amazon is one of them. The current combination of a regulatory headline, a capex-driven free cash flow trough and a misread headline P/E is exactly the kind of setup that creates entry points in great businesses. I remain a holder, and at these levels I am comfortable adding rather than trimming.

Risk Disclaimer

This article reflects my personal opinion and is intended for information and educational purposes only. It does not constitute investment advice, a recommendation, or an offer or solicitation to buy or sell any security. Equity investments carry substantial risk, including the total loss of capital. Share prices, valuation multiples, forecasts and market data referenced here are as of 2 September 2026 and may change at any time. Forward-looking statements, earnings estimates and the fair value derived in this article are based on assumptions that may prove incorrect. Past performance is not an indicator of future results. Please conduct your own research and consult a qualified advisor before making any investment decision.

Disclosure: I hold a position in Amazon. Amazon is a holding in the portfolio of the cost-efficient Haas Invest4 Innovation investment fund (invest4.net). A family member is employed by Amazon. I may buy or sell the security discussed at any time without notice.

Introduction: the boring half of the AI trade

Most of what I read about artificial intelligence at the moment is about who builds the models. Almost nobody writes about who cleans up the mess they create.

Every new large language model that gets deployed inside a Japanese bank, every agent that is given read access to an internal document store, every web application that a mid-sized manufacturer suddenly exposes to the internet — each one is a new door. And doors need locks, hinges and somebody who checks at night that they are still shut. That business is unglamorous, it is recurring, and it is priced as if nothing interesting were happening.

Cyber Security Cloud, Inc. is one of those companies. It sits on the Tokyo Growth market under the code 4493, it is followed by essentially no sell-side analyst outside Japan, and at ¥1,940 per share (close of 31 August 2026) the entire business is worth about ¥20.2 billion — roughly $126 million. Inside that market capitalisation sits around ¥3.4 billion of net cash. So the operating business is being valued at something like ¥16.8 billion, against trailing net income of about ¥0.96 billion.

Here is the shape of the case as I see it. This is a company whose revenue has gone from ¥3.06 billion in 2023 to ¥3.86 billion in 2024 to ¥5.08 billion in 2025, whose operating margin crossed 20% and stayed there, whose recurring revenue base (ARR) reached ¥5.27 billion at the end of June 2026, and whose share price is almost exactly where it was five years ago. In December 2020 the market valued this company at ¥31.8 billion on a fraction of today’s earnings. Today it values it at ¥20.2 billion on roughly four times the revenue. That is not a broken business. That is a multiple that has spent five years deflating while the business quietly grew underneath it.

I hold a position in the Haas Invest4 Innovation Fund. What follows is why — and, just as importantly, what has to go right.

1. Product, business model, brand and moat

What they actually do

The simplest way to describe Cyber Security Cloud is this: they protect web applications that live in the cloud.

If you run a website, an e-commerce shop, a booking platform or a mobile app backend, that application is reachable from the open internet. It gets probed constantly — SQL injection, cross-site scripting, credential stuffing, DDoS floods, and increasingly automated attacks generated by AI tooling. A Web Application Firewall (WAF) sits in front of the application and decides which requests are legitimate and which get blocked.

The company operates four things that matter:

Shadankun (攻撃遮断くん) is the original product, launched in 2014. It is a cloud-delivered WAF with the number one revenue share in the Japanese cloud WAF market. In June 2026 it was repackaged and relaunched as “Web/DDoS Security Type Neo” with expanded features and a new price structure, aimed at making the entry tier affordable to smaller sites while adding the capabilities larger customers had been asking for.

WafCharm is the product that made the company interesting to me. Amazon, Microsoft and Google all sell you a WAF as part of their cloud. What they do not sell you is somebody to write, tune and continuously update the detection rules. That is genuinely hard, genuinely tedious, and requires a security team most companies do not have. WafCharm automates rule operation on top of AWS WAF using the company’s own attack intelligence, and it has since been extended to Azure Front Door. It holds the number one revenue share in Japan for WAF managed operation services.

CloudFastener, launched in late 2023, is the ambition. It is a managed detection and response plus cloud security posture management service spanning AWS, Azure and Google Cloud, delivered by the company’s own analysts alongside the customer’s team. It is a bigger contract, a stickier relationship and a much larger addressable spend per customer. In 2026 the company added an incident response and digital forensics option — the logic being that blocking attacks in peacetime is only half the job; when something does get through, somebody has to reconstruct who came in and through which door.

Security for AI is the newest leg, opened in mid-2026. AI MONBAN, released in June, sits between a company’s employees, its AI tools and its internal data, monitoring and where necessary masking what flows between them. In July they added a vulnerability diagnostic service specifically for AI and LLM applications. This is early — there is no disclosed ARR yet — but the direction is right.

How the money comes in

Around 85% of revenue is recurring subscription income, billed monthly and reported as ARR. The rest is initial setup fees, vulnerability assessments and project work. Gross margins run in the 65–72% range. Once a WAF is sitting in front of a production website, nobody removes it casually; Shadankun’s churn has historically run around 1% per month at the low end of the customer base and considerably lower among enterprises.

A meaningful and growing slice comes through the AWS Marketplace, billed in US dollars. Overseas recurring revenue passed 10% of the total for the first time in 2025, reaching 10.6%. For a Japanese small cap, that is unusual and it matters — it means the product travels.

Brand

In its niche, the brand is strong. Shadankun has taken the Leader award in the WAF category of ITreview’s Grid rankings for seventeen consecutive assessments and ranked first overall in the cyberattack countermeasure software category of BOXIL’s first-half 2026 rankings. Both are Japanese buyer-review platforms, which means the recognition comes from customers rather than from a marketing budget.

The credential that carries more weight commercially is the AWS relationship. Cyber Security Cloud was the first Japanese software company to obtain the AWS Level 1 MSSP Competency. WafCharm has since added the AWS Small and Medium Business Competency, and CloudFastener is listed on the AWS Marketplace with Amazon Security Lake Subscriber Partner status. Inside the AWS ecosystem, those badges are how a small vendor gets shortlisted by a large customer’s procurement team.

2. The market

The Japanese cybersecurity market was worth roughly $10.3 billion in 2025 and is forecast to reach something close to $18.8 billion by 2031, a compound rate around 10.5%. Cloud-delivered controls are the fastest-growing slice of that.

Three things make this market structurally attractive rather than merely large.

First, regulation is a tailwind rather than an obstacle. Japan’s Active Cyber Defense Law has pushed organisations from reactive incident response toward continuous monitoring. METI has stated an intention to grow the domestic cybersecurity sector from roughly ¥0.9 trillion to ¥3 trillion over a decade. In 2026 the government issued specific AI-threat guidance to the financial sector, and management confirmed on the second-quarter call that financial institutions have already come to them with exactly those requirements. When the regulator writes your sales deck, customer acquisition gets cheaper.

Second, the demand is close to non-cyclical. Security spending is not the first line item cut in a recession, because the downside of cutting it is asymmetric and career-ending. In the second quarter of 2025 the company observed 526 million attacks against the web applications it monitors, up 78% year over year. That number does not fall when GDP does.

Third, political intervention risk is low. This is domestic infrastructure protection sold to domestic customers by a domestic vendor, with an expanding but still modest overseas footprint routed through AWS. There is no tariff exposure, no export licence regime, no single dominant government customer.

The obvious counterweight: Japan’s cybersecurity market growing at 10–13% means a company growing at 18–20% has to take share to do it. That is a real requirement, not a formality.

3. Culture and management

Toshihiro Koike has run this company as President and CEO throughout its listed life. The founding stated mission — creating a cyberspace that people everywhere can use safely — is the kind of thing that reads as boilerplate until you notice that the product roadmap has actually followed it: from a single WAF in 2014, to AWS rule automation in 2018, to a US entity, to a Singapore entity, to a managed cloud security service, to AI governance tooling in 2026.

Three signals tell me more than the mission statement does.

The first is that they hit their targets and then set harder ones. The previous mid-term plan called for ¥5.0 billion of revenue and ¥1.0 billion of operating profit in 2025. They delivered ¥5.08 billion and ¥1.10 billion. Having cleared it, in February 2026 they published a new plan targeting ¥20 billion of revenue by fiscal 2030, built around growing the number of customers spending more than ¥10 million a year from 48 to over 500. That is roughly a 31% annual growth ambition. I do not underwrite it. But I note that this is management setting a bar it can be publicly measured against rather than issuing vague aspirations.

The second is candour. In an investor Q&A published on 27 August 2026, management stated plainly that ARR accumulation had come in below their own expectations, that some products had missed new-customer targets, that churn rose temporarily in the first half, and that the causes were internal — an organisational restructuring done at the start of the year and its knock-on effect on sales activity. They also said CloudFastener’s ARR of ¥427 million, while growing faster than anything else in the portfolio, is not growing fast enough. Japanese small-cap management teams are not famous for this kind of self-criticism. I would rather own a company that tells me its sales reorganisation misfired than one that blames the macro.

The third is alignment. In July 2026 they completed a share buyback of up to ¥450 million, taking the maximum number of shares contemplated, and simultaneously doubled the employee share purchase plan subsidy from 15% to 30% — a rate that puts them in roughly the top hundred listed companies in Japan. Internally, the stated goal is for every employee to become “AI-native” and to triple operating productivity by 2028. In a business where headcount is the main cost, that is not a slogan; it is the margin thesis.

The strategy is long-term and the capital allocation so far has been rational: small bolt-on acquisitions (Generative Technology, DataSign), a modest and rising dividend, and buybacks when the stock is cheap. They have also confirmed that further M&A is under review, with external advisers engaged. That is worth watching in both directions.

ARR reached ¥4.997 billion at the end of 2025, up 22%, and ¥5.27 billion at the end of June 2026.

The first half of 2026

Revenue of ¥2.771 billion, up 14.6%. Operating profit of ¥620 million, up 30.0%. Ordinary profit of ¥640 million, up 51.6%, helped by foreign exchange gains on the dollar-denominated business. Net income of ¥450 million, up 43.7%. Full-year guidance was left unchanged and the dividend was raised to ¥6 from ¥5.

Read the quarters separately and you see something the headline hides. The first quarter produced ¥262 million of net income on ¥1.39 billion of revenue — an 18.8% net margin and a 26% operating margin. The second quarter produced ¥185 million on ¥1.38 billion, a 13.4% net margin. The step down was driven by higher personnel costs from planned hiring, external consulting fees related to M&A evaluation, and rising AI service costs that outweighed the server savings AI was supposed to deliver. Management has said AI-related costs will stay elevated through the full year.

Margin profile

Gross margin sits in the 65–72% band. Trailing operating margin is around 21.7%, net margin around 16.2%, and return on equity around 19–20%. On a trailing basis EBITDA runs a little above operating profit — this is an asset-light business with limited capitalised development, which is why free cash flow tracks net income closely rather than flattering it.

The balance sheet holds roughly ¥3.6 billion of cash against about ¥0.25 billion of debt. Debt to equity is around 0.06. There is no refinancing risk here and no scenario in which this company is forced to raise equity at a bad price.

What the numbers tell me

Trailing twelve-month net income is approximately ¥0.96 billion, or about ¥92 per share. Against a ¥1,940 share price that is roughly 21 times trailing earnings, and about 17 times once you strip out the net cash.

Company guidance of ¥865 million net income for 2026 would represent a decline against the trailing figure — which cannot happen without a materially weaker second half than the first. The first half already delivered 52% of the full-year net income target and 54% of the ordinary profit target. My own estimate is approximately ¥1.0 billion for the year, or around ¥96 per share, putting the stock on roughly 20 times current-year earnings. Guidance here is conservative, and I think deliberately so.

The honest caveat on the other side: hitting the ¥6.0 billion revenue line requires second-half growth of around 21% against a first half that grew 14.6%. Management expects the June repricing of Shadankun and further price revisions from the third quarter to do that work. That is the single most important thing to verify in the November report.

5. Valuation

My approach is the same one I apply across the portfolio: estimate what the business earns three years out, apply a fair multiple, and see what annualised return that implies from today’s price.

Fair P/E: 24.

I get there from four inputs. Return on equity around 20%. A revenue base that is roughly 85% recurring with 65–72% gross margins. Net cash equal to about 17% of the market capitalisation, which lowers the risk of the equity meaningfully. And a market position that is number one in its two core categories in Japan. Twenty-four times is not a heroic multiple for that combination — Japanese software peers with comparable profitability trade above it, and the stock itself has traded above 24 times for most of its listed history.

The 2029 estimate.

I start from my ¥1.0 billion net income estimate for 2026 and assume revenue compounds at roughly 20% annually to about ¥10.4 billion by fiscal 2029. That is materially below the company’s own ¥20 billion by 2030 ambition, and it assumes CloudFastener and the Security for AI products contribute but do not transform. I assume net margin improves from around 16.7% toward 21% as the recurring base scales against a cost structure whose largest line is headcount the company is explicitly trying to make three times more productive.

That gives net income of roughly ¥2.2 billion. Against a share count of about 10.3 million — modestly lower than today after buybacks — that is earnings per share of approximately ¥213.

The result.

¥213 × 24 = a target of roughly ¥5,100 per share. From ¥1,940, over three years, that is an annualised return of about 38%, before the small and rising dividend.

Two things are worth noting about the shape of that return. The great majority of it comes from earnings growth, not from multiple expansion — the stock only needs to move from roughly 20 times forward to 24 times. And the starting point already embeds a conservative company forecast, so a simple in-line year plus a re-rating toward the historical average does much of the work.

Scenarios.

A downside case where revenue compounds at 12%, net margin stays flat around 16.5%, and the market awards 18 times gets me to roughly ¥2,500 by 2029 — around 9% annually. That is my floor case, and it is positive rather than catastrophic, which is what the net cash and the recurring revenue buy you.

An upside case where the mid-term plan trajectory is even half-credible, CloudFastener finally inflects and AI MONBAN turns into a real product line, produces a number I am not going to publish because it would sound unserious. It is comfortably above 50% annually. That is the option I am not paying for.

6. Risks, and why the opportunity exists

If the numbers are this reasonable, why is the stock at ¥1,940 rather than ¥3,000?

The de-rating is the whole story. In late 2020 this was a post-IPO Japanese growth stock trading on a triple-digit multiple. Since then the multiple has compressed every single year while earnings grew. Market capitalisation went from ¥31.8 billion at the end of 2020 to ¥15.9 billion a year later, and has spent the four years since oscillating between roughly ¥16 billion and ¥21 billion. Anyone who bought at the top has spent six years underwater, and that leaves a persistent supply of sellers into every rally. The 52-week range of ¥1,448 to ¥2,143 tells you this is still a stock people trade rather than own.

Growth has genuinely decelerated. Revenue growth of 26% and 32% in 2024 and 2025 became 14.6% in the first half of 2026. Some of the prior-year figure was acquisition-assisted. ARR growth has slowed from the mid-twenties to the mid-teens. If the market is pricing this as a mid-teens grower rather than a 20%-plus grower, the market is not being unreasonable — it is being literal.

CloudFastener is behind plan. This is the product the equity story depends on, and management has openly said it is not scaling at the pace they wanted. The stated reason is that a comprehensive managed cloud security service requires more customer education and a longer decision cycle than a WAF, because the buyer has to work out how it divides responsibility with their existing operations team. That is a credible explanation. It is also exactly what a company says before a product quietly plateaus.

Execution wobble. The organisational restructuring at the start of 2026 disrupted sales activity and coincided with a temporary rise in churn. WafCharm’s ARR edged slightly lower. These are self-inflicted and fixable, but they are the reason the stock gave back its post-earnings gains through late August.

Cost inflation from AI. The company is spending more on AI services than it is currently saving from them. That is a common pattern right now and probably a temporary one. It still compresses margins in the near term.

Liquidity and market structure. This is a Tokyo Growth listing with a ¥20 billion market cap and around 10.4 million shares outstanding. Average daily volume is roughly 124,000 shares. Position sizing has to reflect that. There is no analyst coverage to speak of and no institutional floor under the price.

Hyperscaler encroachment. Discussed above. Low probability in the near term, high impact if it happens.

The opportunity exists precisely because the last twelve months contained a genuine operational stumble, and the market has extrapolated it. My read is that the stumble is organisational rather than structural, and that the repricing actions taken in June, which only start showing up in ARR from the third quarter, address the revenue side of it.

7. Latest news and earnings

The half-year report landed on 14 August 2026 and the management presentation followed on 19 August. Revenue, operating profit and ordinary profit all reached record levels for a first half. Ordinary profit rose 51.6% to ¥640 million against a full-year plan of ¥1.2 billion, a 54.1% progress rate almost exactly in line with the five-year seasonal average. Guidance was maintained. The dividend was raised.

Alongside the numbers, four developments matter.

Shadankun’s relaunch as Web/DDoS Security Type Neo in June brought expanded functionality and a restructured price list that management expects to lift both order conversion and retention. The ARR impact begins in the third quarter, with further product repricing to follow.

The Security for AI line opened with AI MONBAN in June and an AI/LLM vulnerability diagnostic service in July. Management reported inbound interest and active sales conversations but has not yet disclosed ARR for either.

The capital return actions — the completed ¥450 million buyback and the doubling of the employee share plan subsidy — were framed explicitly as corporate value initiatives, which in the current Tokyo Stock Exchange environment is the language that matters.

And on 27 August the company published a detailed written Q&A addressing the ARR shortfall head-on, attributing it to internal organisational factors and setting out the response: clearer targeting by customer segment and use case, cross-selling into the existing base, and deeper joint selling with AWS and channel partners. The stock closed at ¥1,940 on 31 August, down 4.3% on the day and roughly 10% below its post-earnings high.

8. Conclusion

I like this position for a reason that has little to do with cybersecurity being a hot theme.

It is a founder-era management team that has met every financial target it has publicly set, running a business with 70% gross margins, 85% recurring revenue, net cash equal to a sixth of the market value, a 20% return on equity and the number one share in its two core Japanese categories — priced at roughly 20 times my estimate of this year’s earnings and about 17 times excluding the cash. The company’s own guidance is conservative enough that it implies a second-half profit decline the first half gives no reason to expect.

The investment case has three legs, in descending order of confidence.

The base case is simply that the existing WAF business keeps compounding in the mid-to-high teens, the June repricing does what management expects, margins normalise as the AI cost bulge passes, and the market pays a fair multiple for a 20% ROE software business with no debt. That alone underwrites the roughly 38% annualised return in my model.

The second leg is CloudFastener working. If the managed cloud security service reaches the scale management is targeting, the average revenue per customer transforms and the company graduates from a WAF vendor to a security operations platform. It is behind plan today, which is exactly why you are not paying for it.

The third leg is Security for AI. Every enterprise deploying AI agents against internal data has a governance problem it does not yet know how to solve, and a company with a decade of monitoring, blocking and masking expertise is unusually well placed to sell them the answer. This is a free option today. It could be the whole story in five years.

What would make me sell: a second consecutive half of ARR deceleration with no visible response, evidence that CloudFastener’s slowdown is demand-side rather than execution-side, or a large debt-funded acquisition that changes the risk profile of the balance sheet.

Until then, I am content to own a company that locks the doors, and to be paid for the patience.

Risk disclaimer and disclosure of conflicts of interest

This article is my personal opinion and is provided for information and educational purposes only. It does not constitute investment advice, a recommendation, an offer or a solicitation to buy or sell any security, and it does not take account of the individual circumstances, objectives, risk tolerance or financial situation of any reader. I am not acting as an investment adviser to you.

Equity investments carry substantial risk, including the risk of total loss of the capital invested. Small-cap stocks listed on the Tokyo Stock Exchange Growth market are particularly exposed to low trading liquidity, wide bid-ask spreads and high price volatility, and positions may be impossible to exit at the quoted price. For investors outside Japan, returns are additionally exposed to movements in the Japanese yen. Past performance and historical growth rates are not indicative of future results.

All figures, estimates and projections in this article are based on publicly available information believed to be reliable as of 1 September 2026, but no representation is made as to their accuracy or completeness. Forward-looking statements, including the earnings estimates and the fair value multiple applied, are inherently uncertain and rest on assumptions that may prove incorrect. Please conduct your own research and consult a qualified adviser before making any investment decision.

Conflict of interest: Cyber Security Cloud, Inc. (TSE: 4493) is a holding in the cost-efficient Haas Invest4 Innovation Fund (invest4.net), which I manage. I therefore have a direct financial interest in the performance of this security, and it is possible that I or the fund may buy or sell shares at any time, including in a manner inconsistent with the views expressed above.

Philipp Haas — investresearch.net

Why I care about this one more than most

I run a small business. Not a big one —, a fund, a research operation, invoices going out, receipts coming in, a tax advisor on the other end of an email chain. And every year I am reminded that the software layer between a small entrepreneur and the tax office is one of the most boring, most necessary, most sticky products in the entire economy.

That is exactly what Macompta.fr does. Accounting, invoicing, payroll, expense reports, fixed assets, VAT and social declarations — the whole administrative spine of a company with somewhere between one and twenty employees, plus the connection to the accountant who signs off on it all. I am close to this product category because I am the customer profile.

Here is the setup, in one paragraph. Macompta.fr is a French SaaS company listed in Paris under ALCPA (ISIN FR001400NQB6). At today’s price of €5.50, the entire company is worth €16.6 million. It just closed its financial year (30 June 2026) with revenue of €5.21 million, up 28.1% — the third consecutive year above 28%. It is profitable, has no debt, pays a dividend, is run by its founder, and generated a 23% operating margin last year. The stock is down roughly 25% over twelve months. Tomorrow, 1 September 2026, a regulatory change lands in France that pushes every single VAT-registered business in the country toward exactly the kind of platform Macompta operates.

The market is treating this as a small, illiquid, AI-threatened microcap. I think it is a compounder that has been derated for reasons that have very little to do with the business. On my numbers I get to a fair P/E of 24 and an annual return potential of roughly 44%.

Let me walk you through it, and let me also be honest about why this is a small position in my fund and not a large one.

1. The product, the business model, the brand and the moat

What they actually sell

Founded in 2007 in Lagord, just outside La Rochelle, by a chartered accountant named Sylvain Heurtier, Macompta.fr sells an integrated online suite: bookkeeping, tax filings, invoicing, payroll, social declarations, expense management, asset and depreciation tracking. Web plus mobile apps. The tagline they have used for years translates roughly as “management made accessible to everyone,” and it is not marketing fluff — the pricing is deliberately at the very low end of the French market.

The target customer is precise: companies and associations with fewer than 20 employees, independent professionals, artisans, retailers, and the sports and cultural associations that make up a surprisingly large slice of French organisational life. Since inception, more than 100,000 users have run their books on it.

Since 2024 there is a second leg: selling to the accounting firms themselves — chartered accountants and payroll bureaus who use Macompta as the engine underneath their own client work.

How the money comes in

Pure subscription SaaS. Annual and monthly licences, per-file pricing, with modules layered on top. Revenue is recognised across the year but the business has a pronounced seasonality that trips up anyone reading a single half in isolation: the January-to-June half carries around 60% of annual revenue, because French fiscal year-ends and filing deadlines cluster there. Anyone who models this company off H1 alone will get it wrong.

Two revenue streams that did not contribute a single euro to the FY 2025/26 numbers are about to start: e-invoicing platform fees and paid AI agents. I will come back to both.

The brand

For a company this size the brand punches above its weight. Macompta is recommended by major French banking networks, by accounting bodies, by management associations, and by national sports federations such as ASPTT and UNASS on the association side. In May 2026 it was named among the best online service platforms in the Palmarès Capital rankings. In a market where trust is the entire purchase decision — you are handing over your books — that third-party endorsement layer is worth more than an ad budget.

The moat, honestly assessed

Macompta is not the leading provider of small-business software in France. Cegid, Sage and EBP are far larger. Pennylane has raised enormous venture money and is winning accounting firms. Indy, Tiime, Evoliz, Axonaut, Dougs and half a dozen neobank-adjacent offerings all crowd the same space. Qonto bundles invoicing into a bank account.

What Macompta is: the price-performance champion at the very bottom of the market, in a segment the venture-funded players find structurally unattractive because the ticket sizes are too small to support their cost base.

The moat is therefore not technology. It is three things stacked:

1. Switching friction. Once your chart of accounts, your VAT history, your payroll files and your accountant’s access all live in one system, moving is a project nobody wants to run in February.

2. A cost structure nobody else can match. A company doing €5 million of revenue with a 23% operating margin can profitably serve a customer paying €10 a month. A company that has raised €100 million cannot.

3. Regulatory compliance as a barrier. French payroll and tax filing rules are a moving target. Staying compliant is a permanent engineering tax that keeps casual entrants out.

Is that an unassailable moat? No. It is a narrow one. But narrow and real beats wide and imaginary.

2. The market: big, growing, and about to be forced open by law

France has roughly four million VAT-registered businesses, of which the overwhelming majority — around 3.8 million — are micro-enterprises and small companies. That is Macompta’s addressable universe, and the company currently touches a low single-digit percentage of it.

The structural driver is digitalisation of a segment that is still, in 2026, partly running on spreadsheets and shoeboxes. But the acute driver is regulation.

Tomorrow, 1 September 2026, e-invoice reception becomes mandatory for every VAT-registered business in France, without exception. Large companies and mid-caps must also start issuing electronic invoices on that date. On 1 September 2027, the issuance obligation extends to every SME, micro-enterprise and independent — including those below the VAT threshold. To comply, each business must designate an approved platform in the central directory.

Macompta obtained definitive registration as an approved platform (Plateforme Agréée) from the French tax authority in January 2026. It is one of roughly 120 to 150 approved platforms, which sounds crowded — and it is. But almost all of those platforms are built for mid-market and enterprise invoice volumes. Macompta is one of the very few offering an integrated invoicing + accounting + platform bundle designed for a three-person business at a three-person business price.

This is the rare situation where the state hands a company a mandatory purchase decision for four million potential customers. It is also a market that is essentially non-cyclical: businesses file taxes in recessions too.

The obvious counterpoint on political risk: this reform has been postponed repeatedly. It was originally due in July 2024. The current calendar has held through the 2025 pilot phase and looks stable, but a French political system that changes government roughly annually is not a source of certainty. And the same government that mandates the platform also just appealed a tax ruling against this company, which I will get to.

3. Culture and management: a founder who never took the money

This is the part of the story I find most attractive, and it is the part a screener will never show you.

Sylvain Heurtier founded Macompta in 2007 and still controls it. His holding company, Limule Capital, owns the majority. There are double voting rights — around 5.47 million votes against 3.02 million shares — so control is not going anywhere.

Read the company’s own framing of why it listed: it describes itself as one of the very few French management software publishers that is neither a venture-backed startup nor a subsidiary of a multinational. A family business, human scale, choosing gradual and self-financed development. That is not the language of a company looking to flip.

The listing history reflects it. Macompta came to Euronext Access in 2024 not to raise a war chest but for visibility and a regulated framework. Over two years the free float went from 8% to 27% — widened deliberately, in steps, rather than dumped in one placement. On 1 April 2026 the company stepped up to Euronext Growth, added a liquidity contract with CIC CIB, and picked up its first analyst coverage from Allinvest Securities, who initiated with a Buy and an €8 target. One of the stated reasons for the transfer: making employee shareholding easier. Again — long-term thinking.

In April 2026 the company professionalised its governance. Heurtier stepped back to Chairman, keeping strategy and oversight, while Sylvain Badina became CEO and Claire Alin Deputy CEO as of 15 April. His son Thibault Heurtier joined the board in late 2025. Founder-led, generationally minded, but with operational management now separated from the founder’s chair.

And the tell I always look for: the founder’s holding was buying stock in the open market in May 2026, twice, for roughly €48,000 combined. Small numbers in absolute terms — meaningful ones for a company with a €16 million market cap and a founder who already controls it.

The dividend history rounds out the picture: €0.10 per share in 2024, €0.11 in 2025, €325,000 paid out in total. A company that grows 28% a year and still pays a dividend is a company that believes its growth is self-funding.

4. Financials: growth, margins, and one messy half-year

Let me put the numbers on the table.

Equity at the last full year end stood at €2.68 million, cash at €0.97 million, financial debt at zero. Return on equity works out around 27–30%, which for a company with no leverage is the number that tells you the underlying economics are genuinely good.

The FY 2025/26 breakdown by segment matters:

• Businesses and associations: €4,466K, +24.2%

• Chartered accountants and payroll professionals: €739K, +58.0%

The second market was only launched in 2024 and is now 14% of revenue growing at nearly 60%. That is the growth engine hiding inside a company most people would describe as “the cheap accounting software.”

Now the messy part, and I want to be direct about it. The first half of FY 2025/26 (July–December 2025) showed revenue up 29% but EBITDA of only €404K, an 18.7% margin against 28.8% for the prior full year. Margins compressed hard.

Why? Because management doubled R&D spending. Product development went from around 10% to 20.4% of revenuein a single half. Capitalised development rose from €167K to €441K. External charges — subcontractors brought in to accelerate — went from €349K to €666K. Personnel costs went from €1,091K to €1,506K.

That money went into two things: AI agents and the e-invoicing platform. In other words, the margin didn’t break — it was spent, deliberately, on the two products that are supposed to drive the next leg of growth. Management held full-year guidance of 25–30% revenue growth and a 25–30% EBITDA margin, which requires the seasonally strong second half to carry an EBITDA margin somewhere around 30–38%. We find out on 15 October 2026 whether they delivered.

One thing to keep in mind when you model this: these are French statutory accounts, not IFRS, and capitalised development flatters reported operating profit relative to cash generation. Look at EBITDA less capitalised development, not EBITDA alone.

5. Valuation: what I think this is worth

Here is my arithmetic, laid out so you can disagree with any single input.

Starting point. FY 2025/26 revenue is known at €5.21 million. The net margin was 18.1% last year and R&D spending stepped up sharply this year, so I model a net margin of around 18%, giving net profit of roughly €0.94 million and earnings per share of about €0.31 on 3.02 million shares.

At €5.50, that puts the stock on a P/E of approximately 17.7 for the year just completed. Enterprise value, adjusting for the net cash position, is roughly €15.6 million — call it 3.0x sales and around 11–12x EBITDA for a business compounding at 28%.

For a European SaaS company growing revenue near 30% with a 23% operating margin, no debt and a 27%+ return on equity, that is a striking multiple. Comparable listed SaaS assets in France and Germany with those metrics have historically traded at three to five times that revenue multiple.

Fair P/E: 24. This is the crux. I do not think a company of this quality deserves a 40x multiple — it is tiny, it is illiquid, minorities have no voting influence, and the moat is narrow rather than deep. But 24x for a business with recurring revenue, 28% growth, real profitability, a founder with skin in the game and a legislated demand tailwind is, in my view, conservative rather than aggressive. It is roughly what a slow-growing, mediocre-quality mid-cap trades at on the Paris exchange today.

Growth assumption: 30% annual earnings growth over three years. Revenue growth I assume continues at the guided 25–30%, and I add modest operating leverage as the R&D step-up of FY 2025/26 stops repeating, plus the first contribution from e-invoicing platform fees and paid AI modules.

The result:

• FY 2028/29 estimated EPS: €0.31 × 1.30³ = €0.68

• Fair value at 24x: €16.34 per share

• From €5.50 over three years: approximately 44% per annum

• Plus a dividend yield of roughly 2% on top

For reference, Allinvest’s published target of €8 implies a P/E of around 26 on the year just reported — a very different route to a similar conclusion about the direction of travel.

What has to go right for this to work? Growth has to stay above 25%, margins have to normalise back toward the guided 25–30% EBITDA range, and the market has to eventually price a €25–30 million company more sensibly than it prices a €16 million one. That last point matters: a large part of the return here is not earnings growth at all, it is the re-rating from 17.7x to 24x. If the re-rating never comes, you still own a business compounding earnings at 30% — which is a perfectly acceptable downside case.

6. Risks: why this opportunity exists at all

A stock does not trade at 17x earnings while growing 28% for no reason. Here is what the market is pricing.

Illiquidity, first and foremost. Average daily volume runs somewhere between 1,600 and 4,600 shares. At €5.50 that is €9,000 to €25,000 of turnover a day. No institution can build a meaningful position. Even the newly installed liquidity contract cannot manufacture a market that isn’t there. This is the single biggest reason the discount exists, and it is also the reason I hold only a minimal position in the Haas Invest4 Innovation Fund. Position sizing here is a function of the order book, not of conviction.

The AI fear. This is the fashionable bear case: large language models will commoditise bookkeeping software, and small accounting SaaS will be disintermediated. I understand the argument and I do not dismiss it. But note the direction of the evidence here — Macompta launched its first AI agent, Emma, in August 2026, an onboarding assistant that helps new users create and configure their files. Management has signalled a next generation of agents for day-to-day administrative and accounting work, targeted at the fourth quarter of 2026, to be sold as a paid option into both customer segments. Founder Heurtier described the launch as the start of a new innovation cycle, with the ambition of embedding AI not just in administrative tasks but in analysis and business steering. AI is a threat to a software company with no data, no compliance layer and no customer relationship. Macompta has all three.

The tax dispute, which just got worse. This deserves attention because most people missed the follow-up. On 17 March 2026 the Poitiers administrative court ruled in Macompta’s favour on the application of the IP Box regime — the 10% reduced rate on income from licensing its own software — for the years ended June 2020 and 2021. The tax authority refunded €150,000 plus interest and costs in June 2026, and the company announced the matter closed on 23 July.

Then on 19 August 2026 the company issued a correction: on 10 August it had been notified that the tax authority had in fact appealed to the Bordeaux Court of Appeal, seeking to overturn the judgment and reinstate the corporate tax. The favourable outcome is no longer final. Two consequences: the €150,000 could be clawed back, and — more importantly — the ongoing applicability of the 10% IP Box rate to Macompta’s licensing income is now uncertain. If that regime is denied, the effective tax rate rises materially and my EPS assumptions come down. This is the risk that most changes the numbers, and it is the one I watch closest.

Margin execution risk. The half-year EBITDA margin of 18.7% is a long way from the 25–30% guidance. Management is betting that a heavy investment half is followed by a strong harvest half. If the October results miss the guided range, the story of “growth and profitability” — which is the entire reason the stock is interesting — takes a real hit.

Competition in e-invoicing. Being one of roughly 120–150 approved platforms is not a monopoly. Pennylane, Qonto, Cegid and the banks are all going after the same mandatory purchase decision, several with far bigger marketing budgets. Macompta’s edge is bundling and price, not distribution.

Governance. The founder controls the votes. Minorities are along for the ride. That is a feature when the founder is a good allocator and a bug the day he isn’t.

French macro and small-cap sentiment. French small caps have been out of favour for years, and the country’s fiscal and political instability has not helped. This stock fell around 25% over the past twelve months and about 15% year to date, from a high of €7.35 to a low of €4.90, with almost no company-specific bad news to explain it. That is the derating that creates the entry point.

7. Latest news flow

The last six weeks have been unusually eventful for a company this quiet.

• 15 July 2026 — Q4 revenue of €1,595K, +27.0%; full year €5,205K, +28.1%. Growth across all products: accounting, payroll, invoicing. Explicitly noted: no e-invoicing revenue and no AI revenue included.

• 23 July 2026 — Announcement of victory in the IP Box tax case and recovery of €150K.

• 19 August 2026 — Correction: the tax authority has appealed to Bordeaux. The company intends to defend its position.

• 26 August 2026 — Launch of Emma, the first AI agent, with a next generation of intelligent agents in preparation.

• 1 September 2026 — E-invoice reception becomes mandatory for all VAT-registered French businesses.

• 15 October 2026 — Full-year FY 2025/26 results. The key datapoint: did the EBITDA margin recover into the guided 25–30% range?

Retail investor chatter around the stock has picked up noticeably since the Emma announcement, mostly focused on two threads: whether agentic AI turns Macompta into a partial substitute for a chartered accountant, and whether the e-invoicing mandate multiplies the addressable base. Both are the right questions. Neither is answerable yet, and I would caution against paying for either in advance — my numbers above assume essentially nothing from AI monetisation.

8. Conclusion: why I own it, and why only a little

Strip away the noise and this is what you are buying at €5.50.

A €16.6 million company with €5.2 million of recurring revenue growing at 28% for the third year running, an 18% net margin, no debt, a return on equity near 30%, a dividend, a founder who buys his own stock, and two brand-new revenue streams — regulated e-invoicing and paid AI agents — that contributed exactly zero to the numbers you are valuing it on. At an estimated 17.7 times trailing earnings.

There are three ways this works out.

The base case is that Macompta simply keeps doing what it has done: compounding revenue at 25–30% with margins in the mid-twenties, while the market slowly notices that a Euronext Growth listing, analyst coverage and a liquidity contract have made it investable. Re-rate to 24x on FY 2028/29 earnings and you get to roughly €16 per share, or about 44% per year.

The upside case is that the e-invoicing mandate does what mandates do. Four million businesses need an approved platform within twelve months, and a meaningful minority of the smallest ones choose the cheapest integrated option available. Add paid AI agents on top of an existing 100,000-user base. Growth accelerates rather than decays, and the fair multiple I have used looks far too conservative.

The consolidation case is the one nobody talks about. A profitable, growing, approved e-invoicing platform with 100,000 small-business users, priced at three times sales, is an extremely cheap way for Cegid, Sage, a bank or a venture-backed competitor to buy distribution into a segment they cannot serve profitably themselves. The founder’s control means it only happens if he wants it to — but the double voting rights that block a hostile approach also mean any friendly one gets done at a price he sets.

The bear case is real too: the IP Box appeal goes against them, margins fail to recover in October, and the stock stays illiquid and ignored for years. You would still own a compounding business, but you would own it for a long time before the market agreed with you.

That is why this is a minimal position in my fund rather than a core one. Not because the thesis is weak, but because the order book is thin, the tax situation is genuinely unresolved, and I have no interest in being the marginal buyer and the marginal seller of a €16 million stock. Size accordingly. Use limits. Do not chase.

But as a piece of unloved, profitable, founder-run, structurally advantaged French small-cap software, priced as though its growth is about to stop when every operating datapoint says otherwise — this is exactly the kind of situation I built my process to find.

Risk disclaimer

This article reflects my personal opinion and analysis as of 31 August 2026. It is not investment advice, not a recommendation to buy or sell any security, and not a solicitation of any kind. It does not take into account your personal financial situation, objectives or risk tolerance.

Equity investments carry the risk of total loss of capital. This risk is materially elevated for microcaps such as Macompta.fr, where very low trading liquidity can make it impossible to enter or exit a position at a reasonable price, where bid-ask spreads are wide, and where individual news items can move the share price by double-digit percentages in a single session. Euronext Growth is a multilateral trading facility with lighter disclosure requirements than a regulated market.

All figures, estimates, growth assumptions, fair value multiples and return projections in this article are my own and may prove to be wrong. Forward-looking statements are inherently uncertain. Financial data is drawn from company publications; FY 2025/26 profit figures are estimates, as audited annual results are not published until 15 October 2026. The IP Box tax dispute is unresolved and pending before the Bordeaux Court of Appeal; an unfavourable outcome would negatively affect the company’s tax position and my earnings assumptions.

Conflict of interest disclosure: Macompta.fr is a holding in the portfolio of the cost-efficient Haas Invest4 Innovation Fund (invest4.net). I therefore have a financial interest in the performance of this security and am not impartial. I may buy or sell shares at any time without prior notice or subsequent update to this article.

Please do your own research and, where appropriate, consult a licensed financial adviser before making any investment decision.

The label that keeps this stock cheap

Every time Newborn Town comes up in conversation, someone calls it “the Grindr of China.” I understand why. The company owns Blued, which really is the dominant LGBTQ dating and social app in the Chinese market, acquired through the takeover of BlueCity in 2023. It is the single most quotable fact about the business.

It is also, at this point, roughly a tenth of the story, and the shorthand is actively costing shareholders money. Because if you file Newborn Town under “Chinese dating app,” you inherit every reflex the market has about Chinese consumer internet: regulatory landmines, opaque governance, a rating that never expands. And you miss what the company actually is.

What it actually is: a Hong Kong-listed operator of a portfolio of social entertainment apps whose revenue comes overwhelmingly from outside China. MICO for live streaming. YoHo for audio rooms. SUGO for companionship-based social. TopTop for game-flavoured social. HeeSay as the international LGBTQ community. The strongholds are the Middle East and North Africa and Southeast Asia, with Latin America now being pushed hard. If you want a listed comparison, Yalla Group is the closest — the same voice-and-live-room economics in the same Gulf markets — except Newborn Town is growing considerably faster and running a much broader product portfolio.

Here is the setup. At HK$8.84, where the shares closed today, roughly 1.40 billion shares outstanding put the market capitalisation near HK$12.4 billion, about US$1.6 billion. The company just reported first-half 2026 revenue of US$607 million, up 37.0%, with profit attributable to shareholders of US$99 million, up 45.8%. It holds substantial net cash. It is buying back its own stock every month. And it trades on roughly 8.8 times my estimate of this year’s earnings.

The stock made an all-time high of HK$14.09 on 15 January 2026 and has since given back about 37%. It is down 21% year to date and 28% over twelve months, in a period during which the underlying business accelerated.

I do not think this is a high-quality business. I want to be blunt about that up front, because it shapes everything that follows. But I think a fair P/E of 17 is defensible, and from a starting multiple below nine, that gap plus the earnings growth gets me to a return potential above 50% per year. I have opened a first, deliberately small position.

1. Product, business model, brand and moat

What the apps actually do

Strip away the app names and there is one product: a room where strangers meet, talk, perform and pay.

MICO is live streaming — hosts broadcast, audiences watch, viewers buy virtual gifts. YoHo is audio-first, the group voice-chat format that has proven enormously durable in Gulf markets where text and video both carry social friction. SUGO is companionship-oriented one-to-one social. TopTop wraps light games around the social layer and has become genuinely popular in Saudi Arabia. HeeSay is the international LGBTQ content community, launched in January 2024, now the leading platform of its kind in Southeast Asia. Blued serves the Chinese domestic LGBTQ market.

Alongside this sits a smaller but fast-growing “innovative business” segment: Playlet, the short-drama app, plus games including Alice’s Dream, and a handful of adjacent products. Playlet is the one to watch. It ranked number one among free iOS entertainment apps in Japan in early July, and the company is targeting exactly the high-spending markets — the United States, Japan, South Korea — where short-drama monetisation works.

How the money comes in

Almost entirely through virtual gifting and in-app purchases. Users buy coins, coins buy gifts, gifts go to hosts, the platform keeps a share. This is a high-gross-margin model — group gross margin runs around 56% — but it is not software margin, because a large slice of revenue is shared with the hosts and content creators who generate the engagement.

Social networking delivered US$539 million in the first half, up 36.5%, roughly 89% of group revenue. The innovative segment grew faster in percentage terms but remains around a tenth of the total.

The brand question

There is no single consumer brand here, and I would argue that is deliberate. Newborn Town runs a portfolio, and the corporate name means nothing to the end user in Riyadh or Jakarta. What the company has instead is a publishing and localisation machine. In December 2025 it ranked fourth on the industry table of Chinese non-gaming publishers by overseas revenue, up one place. That is the relevant scoreboard: not brand equity, but a demonstrated ability to take a format that works in one market and stand it up in another.

Management calls this “product replication plus market replication,” and for once the corporate phrasing is accurate rather than decorative. Take a proven room format, restaff it with local operators, re-tune the payment rails and the content moderation for the target culture, launch. Repeat.

The moat, and I will not oversell it

Let me be straightforward: this is not a wide moat, and it may not be a durable one.

There is no data asset a competitor cannot rebuild. Apple and Google sit between the company and every payment, taking their cut and holding the power to change the rules.

What does exist is narrower and more operational:

Localisation depth. Running a live social product in Saudi Arabia means understanding what content gets a licence pulled, which payment methods people trust, how Ramadan reshapes usage, and how to recruit and retain local hosts. That is accumulated operational knowledge, not technology, and it takes years to build. It is why Western social apps have consistently underperformed in these markets.

Host and creator supply. The top hosts on these platforms earn real money and have real followings. Rebuilding that supply base is the genuinely hard part of entering the category.

Cost position through AI. More on this below, because it matters and because the popular version of the story is not quite right.

Portfolio effect. Any single app can decay. A company running six can afford to have two roll over. That is resilience rather than a moat, but it is worth something.

Could a well-funded competitor replicate this? Yes, and several are trying. That is precisely why I am not paying 25 times earnings for it.

2. The market: large, growing, and politically exposed

The addressable market is global social entertainment, and the specific niches Newborn Town targets are among the fastest-growing pockets of it.

MENA is the anchor. Young populations, extremely high smartphone penetration, high disposable income in the Gulf, and cultural conditions that make voice-based and semi-anonymous social interaction unusually attractive relative to face-to-face or video formats. Monetisation per user in Saudi Arabia and the UAE rivals developed markets.

Southeast Asia provides the volume — large populations, rising incomes, lower per-user monetisation but enormous headroom.

Latin America is the current expansion frontier, and the company reports its flagship products strengthening there through the first half.

Short drama is a genuine new market rather than a repackaged old one. Industry research points to global short-drama revenue reaching around US$14 billion by the end of 2026, from a base that barely existed three years ago. Newborn Town is using AI to compress production cost and cycle time, reporting an improvement of more than 60% in per-title launch efficiency.

Cyclicality: moderate. Virtual gifting is discretionary spending, and a Gulf consumer recession would hurt. But the spend is small-ticket and habitual rather than large and deferrable, which makes it more resilient than most discretionary categories.

Political intervention: this is the real exposure, and it cuts both ways. Live social content is regulated everywhere, and the rules are neither stable nor transparent. Gulf regulators have removed apps from stores over content judgments. Indonesia and Malaysia have done the same. In China, Blued operates in a domain where the regulatory posture toward LGBTQ content has tightened rather than loosened. Add the standard Hong Kong-listing overhang — a Chinese-founded company that Western institutional money structurally underweights regardless of where its revenue comes from.

The company’s structural answer is diversification: no single market, no single app, no single regulator can take out more than a slice. That is a real mitigant. It is not a solution.

3. Culture and management

Founded in 2009 by Liu Chunhe and Li Ping, listed on the Hong Kong main board in December 2019 at HK$1.68 a share. Li Ping serves as CEO. Around 1,785 employees.

The track record on capital allocation is better than the sector average, and the BlueCity acquisition in August 2023 is the proof point. They bought a distressed, US-delisted asset, folded it into the group, and turned it into a contributor. Buying well in a category where most acquirers overpay for growth is not nothing.

What I like more is the shareholder-return behaviour, which is rare among Hong Kong-listed Chinese founders. In March 2026 the board announced a repurchase programme of approximately HK$300 million, funded from internal resources and distributable profits. This is not a shelf authorisation that sits unused — they are executing. Between November and December 2025 they bought over HK$80 million of stock in the open market. In June 2026 alone they repurchased 5.03 million shares for HK$39.8 million. Repurchased shares are being cancelled, not warehoused: the issued share count came down from 1,413,208,391 to 1,408,034,391 in June, with further cancellations since. Buying and cancelling stock at eight times earnings is the single most accretive thing management can do with cash right now, and they are doing it.

Two further signals worth noting. In July 2026 the company switched its reporting currency to US dollars, an honest acknowledgment that this is not a renminbi business and a move that makes the accounts far easier for international investors to read. And the board considered an interim dividend alongside the August results — a company thinking about returning cash rather than hoarding it.

Where I am less comfortable: a management RSU scheme holds around 11% of shares outstanding, which is a large incentive pool and a real dilution offset against those buybacks. And the strategy, while coherent, is fundamentally opportunistic — find a format, find a market, replicate. That is entrepreneurial. It is not the same thing as a twenty-year strategic vision, and investors should not pretend otherwise.

4. Financials and margins

The growth record is the reason to look at all.

First half 2026 (first period reported in US dollars):

• Revenue: US$607 million, +37.0%

• Social networking: US$539 million, +36.5%

• Profit attributable to shareholders: US$99 million, +45.8%

• Adjusted EBITDA: US$111 million, +23.6%

Three things in those numbers deserve attention, and only two of them are good.

First, the monetisation shift. Monthly active users grew just 7.5% in the half. Monthly revenue per social user rose about a quarter, from US$1.97 to between US$2.47 and US$2.54. Almost all of the top-line growth came from getting more money out of roughly the same people, not from adding people.

This is where the AI story actually lives, and it is worth being precise, because the popular version — “AI lets them acquire users cheaply” — is not what the data shows. AI is being applied to content recommendation, user matching and paid-conversion efficiency across SUGO and TopTop, and to content moderation and translation, which strips human labour cost out of operating in dozens of languages. That is monetisation and cost control, not cheap acquisition.

Is that good or bad? Both. Higher revenue per user with flat costs is exactly the operating leverage you want. But growth built on extracting more from an existing base has a ceiling that growth built on adding users does not. Watch the MAU line. If it goes flat while ARPU keeps climbing, the model is on borrowed time. If MAU reaccelerates as Latin America scales, the story is intact.

Second, the margin flag. Revenue grew 37%. Adjusted EBITDA grew 23.6%. That is meaningful margin compression at the operating level, driven by the cost of pushing into new markets and building out short drama. Meanwhile attributable profit grew 45.8% — faster than EBITDA. Profit growing faster than EBITDA while EBITDA grows slower than revenue means below-the-line items did work: interest income on the cash pile, currency effects, or other non-operating contributions. I would not extrapolate that. The operating margin trend is the one that matters, and it went the wrong way this half.

Group operating margin runs around 11%, net margin around 12.6%. Return on equity is strong at roughly 37%, and return on capital is similarly high, because this is an asset-light model with almost no capital expenditure — around RMB 23 million against RMB 2.25 billion of operating cash flow on a trailing basis.

Third, the balance sheet, which is the best part. Cash and equivalents of roughly RMB 2.85 billion against total debt of around RMB 96 million. Free cash flow of roughly RMB 2.2 billion on a trailing basis. Net cash amounts to something on the order of a fifth to a quarter of the entire market capitalisation, which means the operating business is being valued at a materially lower multiple than the headline P/E suggests.

5. Valuation

Here is my arithmetic, laid out so you can substitute your own inputs.

Starting point. First-half attributable profit was US$99 million. Applying a normal second-half contribution with growth moderating from the first-half pace, I model full-year 2026 attributable profit of roughly US$180 million, which on approximately 1.40 billion shares is earnings per share of about HK$1.00.

At today’s HK$8.84, that is a P/E of approximately 8.8. Adjusting for net cash of roughly US$390 million, the enterprise value multiple on the operating business drops to somewhere near 6.7 times earnings.

Let me state plainly what that multiple means. The market is pricing Newborn Town as though earnings will stop growing, or as though a meaningful portion of them will be taken away by a regulator. At under nine times, with net cash and a buyback running, you are being paid to accept those risks rather than being charged for the growth.

Fair P/E: 17. This is the judgment call and I want to defend it properly.

Seventeen is not a quality multiple. It is roughly what you would pay for an average business with average prospects. I am deliberately not applying a software or platform multiple, because the moat does not justify one — no lock-in, no switching costs, app-store dependency, real regulatory exposure. Equally, 17 is well above where the stock sits, and I think the current single-digit rating reflects sentiment about the listing venue and the founding country far more than it reflects the economics of a business earning 37% on equity while compounding revenue in the mid-thirties.

Seventeen splits the difference between what the business deserves on quality and what it deserves on growth. If you think the regulatory risk is existential, use 12 and you still get an attractive number. If you think the market is simply wrong about Chinese-founded companies with predominantly non-Chinese revenue, 20 is arguable.

Growth assumption: 23% annual earnings growth over three years. That is a material deceleration from the 45.8% just delivered and from the 94.6% of last year. I am assuming revenue growth fades from the high thirties toward the low twenties as the base builds and MAU growth stays modest, with buybacks contributing a point or two to per-share earnings.

The result:

• 2029 estimated EPS: HK$1.00 × 1.23³ = HK$1.86

• Fair value at 17x: HK$31.60 per share

• From HK$8.84 over three years: approximately 53% per annum

Roughly half that return comes from earnings growth and roughly half from the multiple moving from 8.8 to 17. That second half is the part I cannot control and cannot forecast. Which is exactly why the position is small.

For context, the analyst consensus target sits around HK$14, with CLSA maintaining Outperform at HK$16 earlier in the year. My three-year number is far above both, because they are working on twelve-month horizons and I am not.

6. Risks, and why the opportunity exists

A business growing 37% does not trade at 8.8 times because the market is stupid. Here is what is being priced.

The category is structurally fragile. Live social entertainment has a history of apps that grow explosively and then decay just as fast. Users are not locked in. Trends move. The company’s answer is portfolio breadth and constant new launches, which is a treadmill rather than a fortress. Anyone underwriting this needs to accept that individual products will roll over and the question is whether new ones arrive fast enough.

Regulatory risk in every direction. Gulf and Southeast Asian regulators can and do remove social apps over content. Chinese authorities set the terms for Blued domestically. Apple and Google control distribution and payments and can change their take rate or their content policies unilaterally. Any one of these could remove a chunk of revenue with little warning.

Margin compression is already visible. The 23.6% adjusted EBITDA growth against 37% revenue growth is not a rounding error. It says that buying growth in Latin America and building short drama costs real money. If that gap widens, the earnings growth I have modelled does not happen.

Earnings quality. Attributable profit growing faster than EBITDA means non-operating items are contributing. That is fine in any single period and unreliable as a trend. I would want to see the full interim report before treating the 45.8% as clean.

The China discount is real, whether or not it is fair. Hong Kong-listed, Chinese-founded companies trade at persistent discounts to Western peers with identical financials. That discount has narrowed and widened repeatedly over the past decade and shows no sign of disappearing. If it never closes, the multiple expansion half of my return thesis simply does not arrive.

Dilution offsets the buyback. An 11% management RSU pool is substantial. Cancelling shares with one hand while issuing them with the other produces a smaller net effect than the buyback headlines suggest.

Volatility. The stock went from HK$1.05 in October 2022 to HK$14.09 in January 2026 and is now at HK$8.84. That is a 13-fold move followed by a 37% drawdown in seven months. Position sizing is not optional here.

So why does the opportunity exist? Three reasons, stacked. The stock ran too far into January and the correction is partly just air coming out. The market cannot decide whether this is a Chinese company or a global one and prices it as the former. And the AI narrative that lifted every social platform in 2025 has rotated elsewhere, leaving a company that is actually deploying AI productively priced as though it is not.

7. Conclusion

Strip away the framing and here is what HK$8.84 buys.

A business compounding revenue at 37% with attributable profit up 45.8%, roughly 37% return on equity, minimal capital expenditure, net cash worth a meaningful fraction of the market value, an active buyback cancelling shares monthly, and a management team that just switched its reporting currency to dollars because that is what the business actually earns. At 8.8 times this year’s earnings, and closer to seven on the operating business net of cash.

Three ways this could work.

The re-rating case. Nothing changes operationally, but the market gradually stops filing this as a Chinese dating app and starts pricing it as a globally diversified social entertainment operator. Earnings compound at 23%, the multiple travels from 8.8 toward 17, and you get roughly 53% per year over three years.

The second-leg case. Playlet works. Short drama is a genuinely new market heading toward US$14 billion, Newborn Town has AI-driven production cost advantages and existing localisation infrastructure in exactly the right high-spending markets, and the innovative segment grows from a tenth of revenue to a third. Growth reaccelerates instead of decaying and the fair multiple I have used looks far too cautious.

The capital-returns case. The quietest and possibly the most reliable. If management simply keeps buying and cancelling stock at these prices, per-share earnings rise mechanically regardless of what the multiple does. Add an initiated dividend and the stock becomes ownable by a class of investor who will not touch a non-payer.

The bear case. A core app decays, a regulator acts, margins keep compressing, the discount never closes, and you own a mediocre business at a fair price rather than a decent one at a cheap price.

That is why this is a first, minimal position in the Haas Invest4 Innovation Fund and not a core holding. The valuation gap is wide enough to be interesting even after you haircut it heavily for everything that could go wrong. But the quality is not there — no durable moat, real regulatory exposure, a category with a history of fast decay — and I do not size positions on upside alone.

At 25 times earnings I would not look at this. At under nine, with net cash and a buyback, the risk-reward is asymmetric enough that I want to own some and learn more. Sometimes the interesting question is not whether a business is excellent, but whether the price already assumes it is terrible.

Risk disclaimer

This article reflects my personal opinion and analysis as of 31 August 2026. It is not investment advice, not a recommendation to buy or sell any security, and not a solicitation of any kind. It does not consider your personal financial situation, objectives or risk tolerance.

Equity investments carry the risk of total loss of capital. Newborn Town carries risks well above those of a typical listed equity: a share price that has moved from HK$1.05 to HK$14.09 and back to HK$8.84 within four years; concentration in social entertainment products with historically short life cycles; dependence on the Apple and Google app stores for distribution and payment; and material regulatory exposure across China, the Middle East and North Africa, and Southeast Asia, where content rules can change abruptly and app removals occur. Hong Kong-listed, Chinese-founded companies are additionally subject to geopolitical, listing and sentiment risks that can affect valuation independently of business performance. Currency risk applies to non-HKD investors.

All figures, estimates, growth assumptions, fair value multiples and return projections here are my own and may prove wrong. Forward-looking statements are inherently uncertain. Financial data is drawn from company publications and market data providers; full-year 2026 earnings figures are my estimates, not reported results, and the detailed interim report should be read before relying on any half-year figure quoted above. Reported net cash and balance-sheet figures are approximate and drawn from third-party data.

Conflict of interest disclosure: Newborn Town is a holding in the portfolio of the cost-efficient Haas Invest4 Innovation Fund (invest4.net). I therefore have a financial interest in the performance of this security and am not impartial. I may buy or sell shares at any time without prior notice or subsequent update to this article.

Please do your own research and, where appropriate, consult a licensed financial adviser before making any investment decision.

Philipp Haas — investresearch.net |

The stock I first bought as a customer

There is a small category of companies I own where I did not come to the investment case through a screener or a broker note. I came to it through my own credit card statement.

Westwing is one of them. Years ago, when we moved and had to furnish a house, I ran into the same wall a lot of people in their thirties and forties run into. On one side sits IKEA — functional, cheap, and everywhere. On the other side sits the designer furniture segment, where a single armchair costs what a decent used car costs. In between there was, for a long time, almost nothing that felt like a curated point of view rather than a warehouse catalogue. Westwing filled exactly that gap for us. We bought chairs from an Italian manufacturer I would never have found on my own. My wife has bought more of the small things than either of us would care to add up.

I was sceptical about European e-commerce for most of the last decade, and I said so publicly. The reasoning was simple. If your business is to sell a commodity product online, you eventually have to explain what you do better than Amazon, and most of the answers were unconvincing. On top of that, the valuations these players carried in 2020 and 2021 were absurd. Westwing itself is the cautionary tale: at the 2021 Capital Markets Day, management put a €1 billion revenue target and a €100 million adjusted EBITDA target on the table for 2024/25. Revenue in 2025 came in at €449 million. That is not a rounding error, that is a strategy that got run over by reality.

What interests me today is that the company that emerged from that crash is a fundamentally better business than the one that made the promise. Westwing now generates around €44 million of adjusted EBITDA on that same revenue base — more than it ever produced in the boom — with a private label at 63% of gross merchandise volume, a negative working capital cycle, €68 million of net cash, no debt, and 26 European countries served instead of eleven.

And the stock, at €12.95 on Xetra as I write this on 28 August, sits about 28% below its February high of €18.05, back near a six-month low, on a market capitalisation of roughly €230 million.

The short version of the case: Westwing is a profitable, structurally growing, brand-led specialty retailer trading at roughly 12x this year’s expected earnings and something closer to 8x once you strip out the net cash. Growth has re-accelerated to 13–15%, driven by a European expansion that is working. Margins took a visible knock in Q2 from a one-off systems migration and macro-driven freight costs, and the market treated a temporary dent as a structural one. On my Fair P/E of 22 applied to what I think the business earns in 2029, I get an implied annualised return of roughly 38% per year. That is my base case, not a price target.

1. Product, business model, brand and moat

What they actually sell

Westwing is a home and living retailer. Furniture, lighting, textiles, rugs, tableware, decoration. What matters for the economics is the split: the large, heavy, complicated furniture is the minority of the business. The bulk sits in accessories, textiles and smaller pieces, with an average basket size of €260 in the last quarter.

This is not a cosmetic detail. Large furniture e-commerce is a brutal business. Returns are catastrophic on the cost line — a customer who orders a wardrobe, half-assembles it, and sends it back destroys the unit economics of three good orders. Made.com found this out and went bankrupt. Home24 found it out and ended up inside the XXXLutz group. Westwing deliberately built a mix where the heavy stuff is a service and a halo, not the profit engine.

How the money is made

Two engines, and understanding the difference between them is most of the analysis.

The first is the third-party assortment: curated design brands sold on the platform. More than 500 partners, including names like Flos, KitchenAid, Kartell, Georg Jensen, Le Creuset and Louis Poulsen. Lower margin, but it brings credibility and search traffic, and it makes Westwing a destination rather than a private-label shop.

The second, and the important one, is the Westwing Collection — own-brand product, largely contract-manufactured, sold at retail margins the company keeps for itself. In 2022 this was 41% of GMV. In 2023, 47%. In 2025, 63%. That mix shift is the single biggest reason gross margin sits near 52% and why a business with flat revenue between 2022 and 2024 could still swing from losses to €44 million of adjusted EBITDA.

Layered on top: a B2B offering, a paid interior design service, a delivery and assembly service, and a small but growing physical store network — six standalone stores and four store-in-stores, with new openings in Frankfurt and a permanent Munich location this year.

Push, not pull — the part most analysts underweight

The structural insight I keep coming back to is that Westwing is a push business, not a pull business. Most e-commerce is pull: the customer knows they need a thing, they search, someone pays Google for the click. That model has no defence, because the cost of the click rises until the marginal retailer earns nothing.

Westwing inverted it. Customers arrive through the newsletter, through Daily Specials, through Instagram, through curated themed worlds where a look is assembled for you. Nobody wakes up needing a velvet pouf. They see one in a styled room, and they want it. That is a demand-creation model, and it has three consequences that show up directly in the P&L: customer acquisition cost is structurally lower, the addressable wallet is larger than stated need, and repeat purchase behaviour is habitual rather than event-driven. 1.32 million active customers, up 13% year over year, and two consecutive quarters of active customer growth after a long period of decline.

The brand and the moat

Delia Lachance, the founder, is still with the company as Chief Creative Advisor. She came out of ELLE Decoration, she has a substantial personal following, and she is treated in the German-speaking market as a genuine taste authority. That is not a marketing budget — it is the thing marketing budgets try and usually fail to buy.

Is Westwing easy to replace? Not trivially. You would need a curated multi-country supply base, a private label with real design capability, an audience that opens your emails, and the logistics to serve 26 markets. Amazon can undercut on price all day and does not compete on taste. The two most credible European challengers of the last cycle are gone or absorbed. Westwing came out of the shakeout as the surviving premium player, which is a much better place to stand than it was in 2021.

The moat is not a fortress. It is a brand-plus-habit moat, of the kind that erodes slowly if neglected and compounds quietly if fed. I would rate it moderate and improving.

2. The market

The European home and living market across Westwing’s existing footprint is roughly €150 billion, with the UK adding around €20 billion on top. Westwing’s €507 million of 2025 GMV is a rounding error inside that. Share gain, not market growth, is the driver here.

Online penetration in home and living remains structurally below fashion or electronics, for the obvious reason that people historically wanted to sit on a sofa before buying it. That gap is closing generationally, and the demographic Westwing serves — urban, design-conscious, predominantly female, comfortable buying a €200 rug from a phone — is precisely the cohort where the behaviour has already flipped.

Two honest qualifications.

This is a cyclical market. Home and living spend is discretionary and correlated with housing transactions, moving activity, consumer confidence and real income. The 2022–2024 European consumer downturn hit this category harder than most. Anyone who tells you Westwing is a defensive compounder has not looked at 2022.

Political interference risk is low, input cost risk is not. There is no regulator deciding what Westwing may charge. But the business imports, ships and delivers physical goods across borders, so it is exposed to freight rates, fuel and energy prices, and tariff regimes. Management explicitly flagged elevated transportation costs from higher fuel prices as a Q2 margin headwind and built the current Middle East conflict into the 2026 guidance. That is the real macro sensitivity, and it is unhedged.

3. Culture and management

Dr Andreas Hoerning became CEO in the summer of 2022, in the middle of the crash, when the previous CEO resigned on two weeks’ notice. Hoerning was previously Chief Commercial Officer and — this matters — he is the person who founded the Westwing Collection. The highest-margin part of the business is his creation, and he now runs the company. His term was extended in July 2025.

Sebastian Westrich has been CFO since early 2023. He bought 1,200 shares on the open market in May 2026 at €14.45, well above today’s price. A supervisory board member bought stock on 13 August, after the Q2 sell-off. Insiders here buy, and they buy after bad news rather than after good news, which is the pattern I want to see.

The strategy is stated as a three-step value creation plan, and the company has actually executed it in sequence rather than pursuing everything at once: fix profitability and cash generation first, then expand the footprint, then scale with operating leverage. They are now in phase three. Ten new countries launched in 2025, the UK in February 2026, the three Baltic markets at the end of July 2026. The UK is already the largest of all markets launched since 2024. Their own track record on country economics is that Portugal, the first expansion market, was loss-making initially and profitable within a year.

Capital allocation deserves specific credit, because German small caps are usually terrible at it. Westwing ran a public tender offer in November 2024 — genuinely unusual here. It cancelled 1.25 million treasury shares. It completed an €8 million buyback at the end of July 2026, repurchasing 512,118 shares. It carries no debt. There is no dividend, which for a business still compounding at this rate I think is the correct choice.

One governance feature to know about: Rocket Internet holds 29.99% of the voting rights. That is not an accident — the position sits deliberately a hair under the 30% mandatory offer threshold, and BaFin has granted a conditional exemption in case buyback-driven share cancellations push it over. Read it how you like. I read it as a large, informed, long-standing holder who has been increasing rather than decreasing exposure.

4. Financials, margins and recent developments

Look at what happened between 2022 and 2025. Revenue went essentially nowhere for four years. Adjusted EBITDA went from negative to €44 million. That is the entire private-label and cost story, and it means the operating leverage on the revenue growth now arriving is real rather than hoped for.

Margins. Gross margin around 52%. Adjusted EBITDA margin 9.8% in 2025, guided to 7.7–9.7% for 2026 as expansion costs ramp. Management maintains a clear path to 10%+ over time. Reported 2025 net margin was 6.5% and reported EPS €1.55, but I would caution against anchoring on that: the Q4 2025 profit of €28.6 million on €143 million of revenue is obviously not a run-rate, and reflects the recognition of deferred tax assets from prior losses rather than operating performance. The good news buried inside that accounting event is that the loss carry-forwards mean cash taxes will stay well below the P&L rate for some years. Return on equity on reported 2025 numbers is roughly 34% against €86 million of equity, and even normalised sits in the mid-twenties — respectable for a retailer, and helped considerably by negative working capital of −€5.5 million, which means suppliers finance the growth.

The recent developments that matter. H1 2026 revenue was €233 million, up 13%. Q2 was the strongest growth print in years: GMV up 15% to €127 million, revenue up 14% to €113 million, orders up 15%, active customers up 13%. International GMV grew 22% against DACH at 9%, and international is now roughly half the group. Third-party GMV grew 23% against 11% for the Westwing Collection, which nudged the private label share down two points to 63% and cost 0.7pp of gross margin. That is onboarding success in the partner brand portfolio, not own-brand weakness, but it does dilute mix.

Adjusted EBITDA was €5.4 million, a 4.8% margin, down €0.8 million year over year. Free cash flow was −€9.4 million. Both figures look worse than the business is: the cash outflow included €9.5 million to settle stock options, two-thirds of it legacy programmes struck at €3.10 from before 2020, plus €3.5 million of buybacks. Roughly 70% of those legacy options should be cleared by mid-2027, which removes an overhang rather than creating one. Net cash still stood at €68 million, up €18 million year over year.

5. Valuation: the Fair P/E view

My framework here is the same one I apply to everything in the fund. I do not ask what the stock is worth today. I ask what the business plausibly earns three years out, what multiple a business of that quality deserves, and what annualised return the combination implies from today’s price.

Starting point. €12.95 per share. Around 17.8 million shares outstanding net of treasury stock, so a market capitalisation of roughly €230 million. Net cash of €68 million takes enterprise value to about €162 million. Against this year’s expected adjusted EBITDA of roughly €42 million, that is under 4x EV/EBITDA on my numbers. NuWays arrives at 6.5x on a lease-inclusive basis and rates the stock Buy with a €23.50 target. Either way, this is not a growth multiple.

Earnings today. Consensus for 2026 sits around €19 million of net income, roughly €1.07 per share. That is a forward P/E of about 12, or roughly 8.5x once you net out the cash. For a company growing revenue 13% with a 63% private label and no debt.

Earnings in 2029. My assumptions, deliberately unheroic:

• Revenue compounding at about 8% from the €490 million I expect this year, reaching roughly €620 million

• Adjusted EBITDA margin recovering to 10.5% as expansion markets mature and the systems migration converts into the promised warehouse efficiency, giving about €65 million

• Depreciation and amortisation around €23 million, share-based compensation normalising lower as legacy programmes clear, leaving EBIT near €40 million

• A ~30% effective tax rate (with cash tax materially lower), giving net income of about €28 million

• Share count drifting down to roughly 18 million through continued buybacks

That produces 2029 EPS of approximately €1.55.

The multiple. I set the Fair P/E for Westwing at 22. The argument for higher: 60%+ private label, structurally low customer acquisition cost, negative working capital, net cash, a founder-brand asset, a 26-country platform with expansion optionality, and a demonstrated 10%+ margin path. The argument for lower: this is a discretionary consumer business with real cyclicality, a €230 million German small cap with limited liquidity, and a management team whose predecessors set targets they missed by 50%. Twenty-two is where those two arguments meet. It is a demanding multiple for a retailer and a modest one for a brand platform, which is roughly what Westwing is.

The result. €1.55 × 22 = €34 per share, against €12.95 today. Over three years that is an implied annualised return of approximately 38% per year, and I have given no credit for the net cash pile, which by then should be considerably larger.

Two sanity checks on the downside. If margins never recover past 8% and revenue compounds at only 5%, 2029 EPS lands nearer €1.05, and at a P/E of 16 the stock is worth around €17 — still 10% annualised from here. If the private label share pushes back toward 70% and the margin reaches 12%, EPS approaches €1.90 and the upside compounds well beyond my base case. The asymmetry is what makes this position worth holding, not the point estimate.

6. Risks, and why the opportunity exists

The stock traded at €18.05 in February on the back of a strong 2025, the UK launch and the buyback. It now trades at €12.95, having touched a six-month low last week. Nothing structural broke in between. What happened is that a market conditioned by 2022 saw a 150 basis point margin decline in Q2 and immediately priced the possibility that the turnaround is reversing.

I think that is the mistake, but let me be precise about what could make it correct.

Margin recovery may not arrive on schedule. The Q2 compression came from three sources: macro-driven contribution margin pressure, higher transportation and fuel costs, and one-off costs from replacing the order and warehouse management systems with SaaS solutions. The first is outside management’s control. The third should reverse and turn positive from Q4. If it does not, the whole operating leverage argument weakens.

Consumer weakness could deepen. Guidance explicitly assumes temporary headwinds from the Middle East conflict and elevated energy prices, and explicitly does not assume a prolonged conflict or a severe energy shortage. That is management being honest about the limits of the forecast, and investors should read it that way.

H2 comparisons are harder. H1 grew 13%, but the implied H2 growth inside full-year guidance is only 4–9% against a stronger prior-year base. Management is pointing at the upper half of the revenue range while being cautious on the second half. A soft Q3 print on 5 November is entirely possible and would probably be treated as confirmation of the bear case.

Mix dilution. Third-party growing faster than own brand is good for assortment and bad for gross margin. Watch the Westwing Collection share.

Liquidity and ownership. This is a €230 million company with a 30% holder. Position sizing matters. There is no exit at scale on a bad day.

Execution across 26 countries. Running a curated, service-heavy proposition in 26 markets with roughly 1,200 employees is operationally demanding, and a stumble in a big expansion market would be expensive.

The opportunity exists because the market is applying a 2022 template to a 2026 business. In 2022 the fear was correct: growth was collapsing and the company was unprofitable. Today growth is accelerating, the company is profitable, cash-generative and debt-free, and the margin issue is largely identifiable and largely temporary. That gap between narrative and numbers is where the return lives.

7. The latest news

Beyond the Q2 numbers themselves, the last few months delivered several things I would file as confirmatory:

The UK launch in February has been the most successful market entry in the company’s history, already the largest of every country opened since 2024, with the full Westwing Collection, curated local and international partner brands, and the complete service suite from day one. Next absorbed Made.com’s brand after its insolvency, but the premium curated position in the UK has been open for years, and Westwing is walking into it.

Estonia, Latvia and Lithuania went live at the end of July, taking the total to 26 countries.

The systems migration is complete — legacy order and warehouse management replaced by SaaS, which is what caused the Q2 one-off costs and what should deliver faster shipping, flexible delivery options and better warehouse efficiency from Q4.

The €8 million buyback finished at the end of July with 512,118 shares retired.

Insiders have been buying into the weakness.

Next scheduled catalyst is the Q3 report on 5 November 2026.

8. Conclusion

I want to be clear about what Westwing is and is not.

It is not a compounder you buy and forget. It is a discretionary consumer business in a cyclical category, in a small-cap German listing!

What it is: a profitable, growing, cash-generative, debt-free brand platform with a private label at 63% of volume, 1.3 million active customers acquired far more cheaply than its peers manage, negative working capital, €68 million of net cash, and a 26-country footprint that costs a competitor years and a great deal of money to replicate. It is priced at roughly 12 times this year’s earnings and under 4 times enterprise value to EBITDA.

Three ways this works from here:

The base case is operational. Margins normalise toward 10%+ as the systems investment pays off and expansion markets mature, revenue compounds high single digits, and the market eventually pays a normal multiple for a normal business. €34 by 2029, roughly 38% annualised.

The re-rating case is faster. Two clean quarters of margin recovery, and a stock that already carries a Buy with a €23.50 target from its covering analyst can close most of that gap in months rather than years. Small caps do not re-rate gradually.

The strategic case is the option nobody is paying for. A debt-free, €230 million European brand platform with 26 markets, a private label engine and a 1.3 million customer base is an obvious asset to a large furniture group wanting genuine online capability, or to a private equity buyer looking at €68 million of net cash and a 10% margin path. XXXLutz bought Home24. Next bought Made.com’s remains. Westwing is the last independent premium player standing, with a shareholder at 29.999995% who knows the asset intimately. I do not underwrite takeovers, and I never buy for that reason. But when it sits underneath a case that works without it, I will happily take it for free.

I bought this company’s products before I bought its shares. Both times, the reasoning was the same: there is a specific gap in the European home market between cheap and unaffordable, and Westwing is the only one filling it at scale, with taste, and at a profit.

Risk disclaimer and conflict of interest

Westwing Group SE (ISIN DE000A2N4H07) is a position in the portfolio of the cost-efficient Haas Invest4 Innovation Fund (invest4.net). I therefore have a direct financial interest in the price development of this security. Conflicts of interest may arise from this. I may buy or sell shares in this company at any time without prior notice.

This article is not investment advice, not a recommendation to buy or sell, and not a solicitation to enter into any securities transaction. It reflects my personal opinion and analysis at the time of writing and nothing more. Every reader must make their own investment decisions, based on their own research, their own financial situation and their own risk tolerance, and where appropriate with the help of a qualified adviser.

Equity investments involve substantial risk, including the total loss of capital invested. Small-cap stocks such as Westwing carry additional risks: limited liquidity, higher price volatility, concentrated shareholder structures, and greater sensitivity to single operational setbacks. Past performance is not indicative of future results.

All financial data, estimates and projections in this article are based on publicly available sources including company reports, earnings calls and market data as of 27 August 2026. Forward-looking statements — including my Fair P/E model, EPS estimates and implied return calculations — are assumptions, not forecasts, and are subject to significant uncertainty. Actual results will differ, possibly materially. Errors cannot be excluded.

Philipp Haas is the manager of the Haas Invest4 Innovation Fund and publishes equity research at investresearch.net.

Some of the best businesses I have owned were boring on the surface and beautiful underneath. Screws. Gloves. Bearings. Lubricants. Safety boots. Cutting discs. The stuff nobody writes a thread about, which every factory on earth cannot run for a single shift without.

In the United States, the market has understood this for decades. Grainger is a roughly sixty-billion-dollar company changing hands near thirty-three times trailing earnings, and it grows sales at high single digits. Fastenal is valued at more than forty times earnings and grows a little faster than that. Investors pay these multiples not because the growth is spectacular but because the cash flows are relentless, the customer relationships are sticky, and the industry consolidates towards whoever can deliver a thousand different low-value items reliably, tomorrow morning, at an auditable price.

Now consider the same business in a country with roughly four trillion renminbi of annual industrial MRO procurement — maintenance, repair and operations — where less than ten percent of that spend has moved online. Where the buyer is often a state-owned enterprise under explicit political instruction to make procurement transparent and traceable. Where the incumbent supply chain is a fog of local distributors, verbal pricing and paper invoices.

That business exists. It is called JINGDONG Industrials, it is the industrial arm of JD.com, it listed in Hong Kong on 11 December 2025 at HK$14.10 per share — and as I write this it trades at HK$13.67. Eight months after the IPO, the stock is below its issue price, despite revenue growth having accelerated from 17% to 27% in the meantime.

That gap between operating reality and share price is what this article is about.

The thesis in one paragraph. JINGDONG Industrials is the number one industrial supply chain technology and service provider in China by gross merchandise value, growing 27% with expanding gross margins, no debt whatsoever, roughly 14.5 billion renminbi of cash resources against a market capitalisation of about 34 billion renminbi, and negative working capital. On my 2026 numbers the headline forward P/E is around 21 times — but strip out the cash the business does not need and you are paying roughly twelve times next year’s earnings for a structural digitisation winner in a market with sub-10% online penetration. I apply a fair P/E of 22, which points to about HK$25 per share on 2028 earnings and an expected return of roughly 28% per annum. It is a very small position in the Haas Invest4 Innovation Fund.

1. Product, business model, brand and moat

What the company actually does

Strip away the language of “digital-intelligent supply chains” and the proposition is simple. A large Chinese manufacturer — an automotive plant, a power utility, a chemical works — needs perhaps forty thousand distinct consumable items to keep running. Historically it bought them from several hundred local suppliers, each with its own catalogue, its own pricing, its own delivery promise and its own invoice. Nobody in head office could tell you what the group paid for a particular fastener across twelve sites.

JINGDONG Industrials replaces that with one entry point. Around 205,000 connected manufacturers, distributors and agents. Roughly 97.7 million SKUs. Delivery tiers running from 48-hour and 72-hour down to same-hour in dense areas, riding on JD Logistics infrastructure. And, critically, a single auditable record of what was ordered, by whom, at what price, delivered when.

Two product families sit inside this. MRO is the classic maintenance and repair basket — the recurring, non-cyclical consumption of a running factory. BOM is production material that goes into the finished good: fasteners, seals, labels, packaging, structural parts, components. BOM is the newer push, it is a lower-margin and slower-turning category, and it is roughly a fifth of the size of the MRO opportunity — but it is the natural extension once you already control a customer’s indirect spend.

Wrapped around the goods is a software layer the company calls Taipu, its end-to-end supply chain solution, sitting on a standardised product master database named Mercator that gives each significant item a digital identity traceable from factory to end user. In 2025 the company standardised more product data using large language models than in the previous five years combined. On top of that sits JoyIndustrial, China’s first supply-chain-specific industrial large model, and a growing fleet of agents — more than seventy deployed in the first half of 2026 alone, covering sourcing, fulfilment and operations.

How the money is made

Here is the part most investors get wrong, so it is worth being precise.

In 2025, of 23.95 billion renminbi of revenue, 22.49 billion came from selling goods — first-party e-commerce, booked gross, at a product gross margin of roughly 11%. Only 1.46 billion came from services — marketplace commissions, advertising, technology — but that revenue carries gross margins above 90%.

Blended, that produced a group gross margin of 17.4% in 2025, up from 16.2% the year before, driven by better purchasing efficiency and a cleaner supplier network. The mix shift towards services and towards own-brand goods (Falichuang for professional-grade items, Huixiang for value-priced generics) is the single most important margin lever in the model, and it is working.

Below the gross line the cost structure is remarkably light. Research and development ran at 307 million renminbi — 1.3% of revenue. General and administrative at 1.1%. Capital expenditure for the entire year was 8.7 million renminbi. Depreciation and amortisation totalled 17.6 million. This is not a company that builds warehouses; it rents capability from its parent and spends its money on code and category management.

The working capital position is where the model becomes genuinely attractive. At the end of 2025 the company carried 1.69 billion of inventory and 277 million of trade receivables against 6.48 billion of trade payables. Customers and suppliers finance the business. It generated 2.09 billion of operating cash flow on 1.13 billion of adjusted profit. There are no bank borrowings of any kind — a detail worth pausing on, because the finance cost line in the accounts is not interest at all, it is the fee paid to JD Technology for factoring receivables from large enterprise customers.

Brand and moat

Two questions matter here: how hard is this to replicate, and how hard is it to leave?

On replication, the honest answer is that anyone can build a website with a million SKUs. What is hard is the combination of a supplier network of two hundred thousand counterparties, a standardised product taxonomy that lets you tell a customer that item A and item B are functionally identical, a fulfilment network capable of same-day delivery on low-value items across a continent, and — the piece almost nobody outside China appreciates — a compliance record good enough to survive an audit of a state-owned enterprise’s procurement department. That last element is not technology. It is institutional trust, and JD’s brand carries it.

On stickiness, the numbers speak. Key enterprise customer transaction value retention was 116.6% for 2025 and reached 119% for the twelve months to 30 June 2026. Existing customers spent nearly a fifth more than the year before. The customer count moved from roughly 10,600 in 2024 to 13,300 in 2025. Around 60% of China’s Fortune 500 industrial companies are on the platform.

Once a customer has integrated its procurement system with Taipu, standardised its material master data on Mercator, and started running approval workflows through the platform, switching means redoing an internal transformation project. That is not a price decision; it is a change management decision. It is exactly the dynamic that makes Grainger’s national accounts so durable.

Two caveats I want on the record. Concentration is low but so is dominance: no single customer exceeds a tenth of revenue and the top five are under 30%, which is healthy — but with 33.5 billion of GMV against a four-trillion market, “market leader” here means leading a landscape so fragmented that the leader has barely one percent of it. And the JD relationship cuts both ways: revenue originating on JD’s platform fell from 47.1% of the total in 2022 to 39.7% in 2024, which is the right direction, but the umbilical cord is real.

2. The market

The addressable number is close to four trillion renminbi of annual industrial MRO procurement in China. For scale, total enterprise materials procurement across the economy ran at 196.3 trillion renminbi in 2025, growing 4.2%.

The important variable is not the size but the penetration. Under 10% of industrial goods procurement in China happens online today. The digitisation of this category began around 2015 with the “sunshine procurement” push in the state sector, and it has been slow because industrial buying is genuinely hard: non-standard specifications, long tails, technical selection, delivery-time sensitivity.

Structural growth therefore comes from three stacked drivers, not one. The category itself grows with industrial output. The online share of the category grows from a very low base. And the leading platform takes share within the online segment. When those compound, a 20%-plus grower in a low-single-digit end market is arithmetically unremarkable rather than heroic.

Cyclicality. MRO spend tracks factory utilisation, not factory capex — a running plant consumes gloves and lubricant regardless of whether it is expanding. That makes the core basket meaningfully less cyclical than capital goods. BOM is a different animal: it moves with production volume, and it is the faster-growing piece, so cyclicality in the mix is rising rather than falling. China’s persistent industrial price deflation is also a genuine headwind to nominal revenue growth, and it has not gone away.

Political interference — and the other side of that coin. Your notes were right that Beijing is behind this. The State Council’s opinion on the “AI+” initiative and a joint action plan from the Ministry of Commerce and seven other ministries on accelerating digital-intelligent supply chains both explicitly target exactly what this company sells. State-owned enterprises are being pushed towards traceable, compliant, cost-audited procurement, and JINGDONG Industrials is the most obvious vehicle. In the first half of 2026 the company signed or deepened relationships across the state sector, including a strategic partnership with Zhejiang Communications Investment Group and a showcase role at a State-owned Assets Supervision Commission-guided procurement benchmarking event.

But I would not describe this as protection. The same state that promotes the category regulates it. The group holds its value-added telecom licence through a contractual VIE arrangement rather than direct ownership, because foreign capital is restricted in that activity. The registered shareholders of the onshore entity were changed as recently as March 2026. This is a structure that works until a regulator decides it does not, and every China investor should size positions with that in mind.

3. Culture and management

This section is the one that changed most between when your original notes were written and today, and it deserves candour.

The chairman. Liu Qiangdong sits as non-executive chairman and controls roughly 76% of the shares through the JD structure. He is not running the company day to day. What he did do, in May 2026, was accept a restricted share unit grant equivalent to about 2% of the issued capital of each of JD’s listed subsidiaries, including this one, vesting over four years. I read that the way I read most founder equity grants: as a signal about where he intends to spend attention. It is also a dilution the market has to absorb, and I would rather have both facts on the table.

The CEO change. Song Chunzheng built this business. He joined JD in 2013, incubated the enterprise procurement operation, took the industrial unit independent in 2017, became CEO in 2020 and carried it to the Hong Kong listing. On 23 June 2026 the company announced he was stepping down for personal health reasons, effective 1 July — barely six months after the IPO. The stock fell hard on the news, and the company was buying back shares in the low elevens the following day.

His successor is Wang Peinuan, 53. Two decades at Lenovo and Digital China running large-account and commercial businesses, then key accounts at 360, joining JD in 2018 where he ran JD Cloud’s major accounts, then group strategic partnerships, then the group enterprise business, and from November 2025 the BOM business inside JINGDONG Industrials itself. He has signed a three-year service contract.

How do I read it? Not as a disaster, and not as a non-event. The bear case is straightforward: the founder-operator who understood this business from the inside is gone, six months after listing, and “personal health reasons” so soon after an IPO always invites suspicion. The bull case is that the successor is not a parachuted-in administrator — he was already running the newest and strategically most important growth vector inside the company, and his background in enterprise and government-facing sales is precisely the skill set the next phase requires. The board confirmed no disagreement.

The results give the successor the benefit of the doubt so far. The first set of numbers reported under the new arrangement showed the fastest growth in several years. But this is a genuine reduction in the quality of the management leg of my framework, and it is why my fair P/E here is 22 rather than the higher multiple the business quality alone might justify.

The rest of the bench. CFO Wang Xuedong came from the CFO seat at JD Technology after eleven years at PwC in Beijing and London. The board includes Xu Bingdong from Granite Asia, and independent directors with serious pedigree: Gu Baofang, previously head of Asia-Pacific private equity at the Abu Dhabi Investment Authority and before that CEO of GE Capital China; and Sun Hanhui, former CFO and later president of Qunar, a Chinese CPA who sits on several US and Hong Kong listed boards. For a recently listed subsidiary of a Chinese conglomerate, that is above-average governance.

Long-term orientation. The clearest evidence is the 2025 profit and loss statement itself. Adjusted profit grew only 5.3% that year, because management pushed fulfilment expenses up 56.7% — from 5.5% to 7.4% of revenue — to build out infrastructure for overseas expansion and new categories. They accepted a flat profit year to fund the next growth leg. The first half of 2026, with 27.4% revenue growth and 43.4% adjusted profit growth, is what that spending bought.

4. Financials, margins and recent developments

The revenue line

Revenue has compounded from 14.13 billion renminbi in 2022 to 17.34 billion in 2023, 20.40 billion in 2024 and 23.95 billion in 2025. That is a three-year compound rate above 19%, but the trend within it was decelerating — 22.6%, then 17.7%, then 17.4%.

Then it turned. First quarter 2026 revenue came in at 5.66 billion, up 25.3%. First half revenue reached 13.06 billion, up 27.4%, which implies second-quarter growth of roughly 29%. Growth accelerated for two consecutive quarters, and at this run rate the company clears 30 billion renminbi of revenue for the full year.

Underneath, GMV grew 16.5% to 33.5 billion in 2025 — but the two customer cohorts diverged sharply. Key enterprise GMV rose 26.5% to 16.5 billion; the SME and micro cohort grew only 8.3% to 17.0 billion. The large-account business is doing the work, which is both a strength (higher retention, deeper integration) and a concentration to watch.

Profitability

You have to be careful with the headline numbers here, because 2025 contains a large one-off. Reported profit for 2025 was 2.31 billion renminbi, up 203.8% — but 1.40 billion of that was a non-cash fair value gain on convertible preferred shares that converted to ordinary shares at listing. The honest figure is the adjusted one: 1,130.7 million, up 5.3%, on the deliberate investment year described above.

The trajectory on the adjusted line: 765.8 million in 2022, 901.1 million in 2023, 1,073.4 million in 2024, 1,130.7 million in 2025, and then 750 million in the first half of 2026 alone — up 43.4% year on year. Operating profit rose 38.1% to 530 million and reported net profit 37.2% to 619 million in the same period.

Margins. Gross margin at 17.4% and rising. Adjusted net margin of 4.7% in 2025 moving to 5.7% in the first half of 2026. Adjusted operating margin around 3.4% in 2025 and roughly 4.1% in the recent half. Because depreciation is immaterial in this model, EBITDA and EBIT are within a rounding error of each other — the adjusted EBITDA margin is around 4.3%, and anyone quoting a large EBITDA-to-EBIT gap here has the wrong company.

Let me be direct about the comparison to the US names, because it is the most common error I see in the bull case. Grainger runs a gross margin near 39% and an operating margin around 15%. JINGDONG Industrials will never reach those levels on a 94%-first-party-goods revenue mix. The two models book revenue differently and carry different functions. What matters is not whether the Chinese company converges on American margins — it will not — but whether its own margin direction is upward on a growing base. It is, on both counts.

Returns. Reported return on equity looks pedestrian at first glance, and the reason is the cash. Post-IPO equity of 11.2 billion sits against cash resources of 14.5 billion. Strip the excess cash out and the operating business runs on negativeinvested capital — it is financed by supplier payables. Return on operating capital employed is, for practical purposes, not a meaningful number because the denominator is negative. That is the correct way to think about a business like this, and it is exactly why the American comparables earn returns on capital in the thirties and forties.

What is new

The first half of 2026 brought the first disclosure of AI contribution as a hard number rather than a slide. More than seventy agents deployed across sourcing, fulfilment and operations. AI-driven incremental GMV equal to 6.5% of key enterprise customer GMV. Productivity in core roles up 16.6%. The company’s IndLens data platform — twenty-seven autonomous agents compressing data processing from months to hours — was featured in the World Economic Forum’s AI white paper at Summer Davos this year.

I take these disclosures seriously precisely because they are specific and quantified. It is easy to say “we use AI.” Attaching a percentage of GMV to it invites the market to check the claim next period.

Elsewhere: the “Empowering Thousands of Industries, Trillions in Cost Savings” initiative has connected more than 5,000 core enterprises. Overseas services — accompanying Chinese manufacturers as they build plants abroad — served roughly a hundred companies in 2025, the first full year of the effort, growing fast off a tiny base. And M&G crossed the 5% disclosure threshold in July 2026, buying nearly a million shares in a single day.

5. Valuation

The starting point. At HK$13.67 with 2.713 billion shares outstanding, the market capitalisation is approximately HK$37.1 billion, or about US$4.75 billion, or roughly 34.0 billion renminbi.

The cash. At the end of 2025 the company held 14.5 billion renminbi of total cash resources — cash, restricted cash, term deposits and short-dated wealth management products — with zero borrowings. That is about 43% of the market capitalisation sitting in the bank. Given negative working capital, most of it is genuinely surplus. I will be conservative and treat roughly 12 billion as excess, which is about HK$4.80 per share.

My estimates. For 2026 I model revenue around 30 billion renminbi, up roughly 25%, with adjusted net profit of about 1.63 billion — the first half is done at 750 million and the second half is seasonally stronger. That is earnings per share of roughly RMB 0.60, or HK$0.66.

For 2027 I assume growth moderates to 22% on revenue of about 36.5 billion, with the adjusted net margin reaching 6.0% for roughly 2.2 billion of profit. For 2028, 20% growth to about 44 billion of revenue at a 6.5% adjusted margin gives roughly 2.87 billion renminbi of adjusted net profit, or approximately RMB 1.04 per share on a modestly diluted count — call it HK$1.14.

Where the stock trades today. On my 2026 estimate the headline forward P/E is about 21 times. Deduct the surplus cash and the enterprise value per share is around HK$8.90, which puts the operating business on roughly twelve to thirteen times next year’s earnings. For a business growing revenue at 27% with expanding gross margins and no debt.

Fair P/E. I apply 22. That is a deliberate discount to what quality alone would suggest — the American comparables carry 33 to 43 times, and the ten-year median for Fastenal is around 30 — and the discount reflects four things I do not wave away: the VIE structure, the 76% parent shareholding and small free float, the recent CEO transition, and structurally thinner margins that leave less room for operational error.

The maths. HK$1.14 of 2028 earnings at 22 times gives a target of approximately HK$25 per share. From HK$13.67, over the roughly two and a half years to when those results are reported, that is a total return near 85% — or approximately 28% per annum.

Sanity checks on that number. Consensus among the seven analysts covering the stock sits near HK$20 with a unanimous positive rating, and UBS initiated coverage with a Buy and a HK$23.50 target — so my number is above the street but not in a different universe. And the ex-cash multiple gives a second line of defence: even if the fair multiple proves to be 18 rather than 22, the 2028 target is still above HK$20 and the annualised return still comfortable.

6. Risks — and why the opportunity exists

A stock does not trade below its IPO price for eight months while accelerating unless somebody is unhappy. Here is my read on who, and why.

The mechanical explanation. The six-month lock-up from the December listing expired in June 2026. Almost to the day, the share price broke from around HK$13.50 in mid-June to under HK$11 by 24 June. On top of that, on the evening of 23 June, the CEO resignation landed. Two separate shocks inside a fortnight, in a stock with a free float under 30% and modest daily turnover. The company responded by buying stock — six repurchase tranches in the first half of 2026, some executed in the low elevens.

That is a supply-and-sentiment story, not a business story. The recovery to HK$13.67 since then, following the August results, is the market slowly re-underwriting the operating case.

The real risks, ranked.

Parent dependence. This is the one I watch most closely. The company paid JD Group roughly 937 million for logistics and warehousing, 495 million for technology and traffic support, 432 million for shared services and 59 million for payment processing in 2025, plus 204 million to JD Technology in factoring fees. It received 751 million back for marketing services. These are disclosed continuing connected transactions on arm’s-length terms with independent director sign-off and auditor confirmation — but a subsidiary whose logistics, traffic, payments and receivables financing all come from its 72% shareholder does not have fully independent economics, and the market is entitled to discount that.

The VIE. Discussed above. Structural, unquantifiable, permanent.

Margin thinness. At a 4% operating margin, a 100 basis point mistake on purchasing or fulfilment is a quarter of the profit. The 2025 fulfilment cost overshoot demonstrates exactly how quickly the P&L can absorb an investment decision.

Chinese industrial deflation. Falling producer prices compress nominal revenue growth and squeeze the pass-through spread. This has been a live drag for years.

The customer mix. Growth is concentrated in large enterprises, many of them state-owned, whose procurement digitisation is partly policy-driven. Policy priorities change.

Competition. ZKH Group, listed in New York, is the pure-play comparable — it reached operating profitability for the first time in the second quarter of 2026 with GMV up 18.9%, on trailing revenue around 1.2 billion dollars and a market capitalisation near half a billion. It is growing, but it is a fraction of the size and has only just crossed into profit. Alibaba’s 1688 sits in the SME end of the market, and thousands of offline distributors still hold the vast majority of the volume. Competition is real but nobody is currently out-executing the leader.

Governance and control. One shareholder holds three quarters of the equity. Minority shareholders are along for the ride.

Why I think the opportunity is real. Because none of the above explains a business accelerating from 17% to 27% growth, with expanding gross margins and improving profit conversion, trading below its issue price at twelve times ex-cash earnings. Lock-up supply is a temporary condition. A management transition is resolved by results, and the first set of results under the new CEO were the best in years. The mismatch between the operating trajectory and the tape is the entire reason a position exists.

7. Latest news and earnings

The interim results published on 13 August 2026 were the most important data point of the year.

Revenue of 13.06 billion renminbi for the six months, up 27.4%. Adjusted net profit of 750 million, up 43.4% — profit growing at roughly one and a half times the rate of revenue, which is the operating leverage the model was always supposed to deliver. Operating profit of 530 million, up 38.1%. Reported net profit of 619 million, up 37.2%.

Key enterprise customers served rose from around 11,000 to around 13,000. Transaction value retention over the trailing twelve months hit 119%, up from 116.6% at the end of 2025 — customers are not just staying, they are handing over a larger share of wallet.

The AI disclosures were new and, to me, the most interesting part of the release. Seventy-plus agents in production. AI-attributable incremental GMV at 6.5% of key enterprise GMV. Core-role productivity up 16.6%. Management framed the Taipu solution as having become standardised and replicable — no longer dependent on the individual expertise of a consultant on site. If that claim holds, it is the difference between a services business that scales linearly with headcount and one that does not, and it is the single most important thing to verify in future reporting periods.

Commercially, the half saw a broad strategic agreement with Zhejiang Communications Investment covering digital enablement, logistics fulfilment, supply chain integration, green energy and financial services, and a prominent role at a State-owned Assets Supervision Commission-guided procurement benchmarking conference hosted by China Coal, where the company presented joint results in industrial large-model deployment, price control and green supply chain work.

Prior to that, first quarter revenue of 5.66 billion was up 25.3%, and the full-year 2025 results in March delivered 23.95 billion of revenue and 1.13 billion of adjusted profit.

8. Conclusion

The investment case rests on a single observation. Two companies do essentially the same thing on opposite sides of the Pacific, and the market prices them as though they inhabit different centuries.

Grainger sells maintenance and repair products in a mature North American market with high online penetration, grows sales at high single digits with 2026 guidance of $19.2 to $19.6 billion, earns a 46% return on equity, and trades at roughly 33 times trailing and 27 times forward earnings. Fastenal, growing at a similar pace, trades above 40 times trailing earnings — well above its own ten-year median near 30.

JINGDONG Industrials sells maintenance and repair products — plus production materials — in a market of comparable absolute size where online penetration is under 10%, grows revenue at 27% with profit growing at 43%, carries no debt and 43% of its market capitalisation in cash, and trades at 21 times headline forward earnings or roughly twelve times excluding that cash.

I am not arguing that the Chinese company deserves an American multiple. It does not: the VIE structure is a real risk, the parent relationship compromises economic independence, the margin structure is thinner by construction, the free float is small, and the founder-CEO left six months after listing. Those are the reasons my fair P/E is 22 rather than 30.

But 22 against a market price implying roughly 21 today — on a business whose earnings I expect to roughly two-and-a-half times by 2028 — produces a target near HK$25 and an expected return of about 28% per annum. The valuation gap does not need to close for that to work. The earnings alone deliver most of it. Any narrowing of the discount to the Western comparables, any re-rating as the CEO transition fades from memory, any dividend from the 10.1 billion renminbi of distributable reserves, is upside I have not underwritten.

There is one more thing I keep coming back to. The last two decades of Chinese equity investing have mostly been a story about the consumer. The next decade, on the evidence of both policy direction and where the capital is flowing, looks more like a story about industrial productivity. If that is right, then owning the digital rail on which Chinese factories buy the things they need is not a niche position. It is one of the more direct expressions of the theme available in public markets — and it currently costs less than it did on its first day of trading.

A note on the numbers in this article

All financial figures are drawn from the company’s 2025 annual report, its 2025 full-year results announcement of 5 March 2026, its first quarter 2026 disclosure of 12 May 2026, and its interim results announcement of 13 August 2026. Share price of HK$13.67 and peer valuations are as of 26 August 2026. Financial statements are reported in renminbi; the shares trade in Hong Kong dollars, which introduces a translation effect between reported earnings and quoted price that is not hedged.

Risk disclaimer

This article reflects my personal opinion and is provided for information and educational purposes only. It is not investment advice, not a recommendation to buy or sell any security, and not an offer or solicitation of any kind. It does not take into account your financial situation, investment objectives or risk tolerance.

Equity investments carry the risk of substantial loss, up to and including the total loss of capital invested. This is particularly true of the security discussed here, which combines several elevated risk factors: a Chinese variable interest entity structure whose enforceability depends on the continued tolerance of Chinese regulators; a controlling shareholder holding approximately 76% of the equity; a limited free float and correspondingly limited liquidity; a recent listing with a short public track record; sensitivity to Chinese industrial activity, producer price deflation and policy direction; and currency risk between the renminbi reporting currency and the Hong Kong dollar quotation. Foreign investors in Chinese issuers may also face legal, custody, tax and information risks that do not apply in domestic markets.

All forecasts, estimates, target prices and expected returns in this article are my own assumptions about an uncertain future. They may prove materially wrong. Past performance and past growth rates are not indicative of future results. Figures were compiled with care from public company disclosures but no warranty is given as to their accuracy or completeness, and they may become outdated without notice.

Please form your own judgement, do your own research, and consult a qualified, independent adviser before making any investment decision.

Conflict of interest disclosure: JINGDONG Industrials (7618.HK) is a holding in the Haas Invest4 Innovation Fund(invest4.net), the cost-efficient investment fund I manage. I therefore have a direct financial interest in the performance of this security and may buy or sell shares at any time without prior notice or subsequent disclosure. Please read the article with that conflict of interest in mind.

Philipp Haas — investresearch.net

Why I own a small position in Sweden’s most unloved life science holding company — and what has to happen for it to work

Introduction: The one corner of the market where I hire help

I want to start this one with a confession, because I think it makes the rest of the article more useful to you.

There is exactly one sector where I can say, with a clear conscience and no false modesty, that I do not have an edge: biotech. And within biotech, the part that is hardest of all — early-stage companies with no revenue, no profit, and a valuation that rests entirely on the probability-weighted outcome of a clinical trial I am not qualified to read. My Fair PE model, the framework I run almost everything through, is useless here. There is no E. There is no normalised multiple. There is a scientist, a protocol, and a regulator.

For most of my career the honest answer was simply: skip it. But biotech is one of the great long-duration growth sectors, and I do not love the idea of having a permanent zero in a part of the market where the winners can compound at extraordinary rates. So I did what I do in every domain where I lack expertise — I looked for someone who has it, and then I looked for a way to buy their work at a discount.

That search led me to Flerie AB (Nasdaq Stockholm: FLERIE).

The abstract of the case, in one paragraph: Flerie is a Swedish life science investment company founded and controlled by Thomas Eldered — the man who co-founded Recipharm and built it from a single Swedish tablet factory into one of the world’s five largest contract development and manufacturing organisations before selling it to EQT in 2021. Flerie holds roughly two dozen mostly private biotech and pharma companies, marked to the last financing round or the last listed price. Reported net asset value at the end of July 2026 was SEK 39.26 per share. The stock trades around SEK 21. That is a discount of roughly 46% — and since around 16% of NAV is cash and short-term assets related to portfolio companies, on which no rational person would apply a discount at all, the operating portfolio is being marked down by something closer to 55%. You are buying a venture portfolio you could otherwise never access, run by an industry insider with his own fortune in it, for less than half of what its most recent transaction prices say it is worth. There are two ways to make money: the NAV goes up, and the discount closes. There are also two ways to lose: the NAV goes down, and the discount does not close. The rest of this article is about which of those is more likely.

1. Product, business model, brand and moat

Flerie does not sell a product. It owns companies. That distinction matters, because it changes what you are actually underwriting.

What the company does. Flerie is an active, long-term life science investor with a portfolio spread across Europe, Israel and the United States, split into two segments. Product Development — early- and mid-stage biotech and pharma companies working toward clinical proof-of-concept and approval — accounted for 64.5% of NAV at the end of July. Commercial Growth — businesses that already sell products or services — accounted for 19.7%. The remainder is cash, receivables and other assets tied to the portfolio companies. A third segment, Limited Partnerships (fund investments), was divested in late 2025 for SEK 109 million; management concluded that direct ownership was where their edge actually lived, and I agree with that call.

The largest individual holdings at the end of July give you a fair picture of the shape of the thing: Lipum (100% owned, 10.2% of NAV) with its anti-BSSL antibody SOL-116 heading into Phase II in rheumatoid arthritis; Prokarium (50%, 8.4%) with its engineered Salmonella immunotherapy platform in bladder cancer; KAHR Medical (34%, 6.0%) in Israel; Empros Pharma (79%, 6.0%) in obesity; Atrogi (44%, 5.8%); and the listed Xspray Pharma (17%, 4.0%). On the commercial side sit Symcel (6.2%), NorthX Biologics (61%, 5.9%), Chromafora (4.0%) and Nanologica (2.1%).

How they earn money. They don’t, in the conventional sense — and this is the single most common misunderstanding I see. Flerie’s income statement is close to irrelevant. Reported revenue is a rounding error; the entire economic result flows through the fair-value line on the portfolio. Returns come from exits, from up-rounds that lift carrying values, and from the occasional listing. The reference point is Cormorant Pharmaceuticals, an early Flerie holding sold to Bristol Myers Squibb in 2016 for up to USD 520 million. That is the shape of the win.

Brand. Small, but genuinely valuable where it counts. Flerie is not a brand for consumers; it is a brand for founders and co-investors deciding whom to let onto their cap table. Here the Recipharm lineage does real work — Eldered’s name opens doors that a generalist fund’s would not. The most persuasive evidence is the syndication ratio management discloses each quarter: in Q2 2026, co-investors put in roughly 3.3 times what Flerie put in. Other people with capital and scientific judgement keep choosing to invest alongside them. That is a brand doing its job.

Moat. Thin as a corporate moat, real as an access moat. Anyone can start a life science holding company. Almost nobody can assemble Flerie’s specific portfolio at Flerie’s specific cost basis, or replicate three decades of relationships across European drug manufacturing. And there is one asset that genuinely differentiates: NorthX Biologics, a biologics manufacturing site in Matfors, Sweden. Flerie is not purely a financial owner — it owns industrial capacity in the exact discipline the founder mastered. For a portfolio company needing GMP material, having your largest shareholder own a plant is not nothing.

2. Market

The addressable market here is the private biotech funding market, and it is enormous, structurally growing, and currently miserable — which is precisely why the entry price exists.

Long term, the drivers are as sturdy as any I follow. Ageing populations across the developed world. Big pharma facing a patent cliff of historic scale toward the end of this decade, with balance sheets full of cash and pipelines that need replenishing — which means acquirers with both motive and means. Continued outsourcing of manufacturing. And a scientific frontier — cell and gene therapy, engineered bacteria, next-generation antibodies — that keeps producing genuinely new modalities rather than incremental reformulations.

Against that: the cyclicality is brutal, and we are living through the down leg. Small-cap and private biotech has been in a funding winter since rates normalised, and the pain has been concentrated at exactly Flerie’s end of the market. Management put it plainly on the last call — the current climate demands disciplined capital allocation, and their focus has been on funding existing holdings to reach milestones rather than chasing new deals. When capital is scarce, private marks drift down, down rounds happen, and valuations compress even where the science is progressing.

Political exposure is real but two-sided. Drug pricing pressure in the US, FDA staffing and process uncertainty, and tariff noise around pharmaceutical supply chains all weigh. Yet the same political pressure that squeezes pricing also accelerates reshoring of European manufacturing capacity — a tailwind for an asset like NorthX. And a portfolio spread across Sweden, the UK, Israel and the US is at least not a single-jurisdiction bet.

The honest framing: this is a deeply cyclical market near a cyclical low, with secular growth underneath. That is usually where I want to be buying. It is also where you can be early and stay early for a long time.

3. Culture and management

Thomas Eldered is not a financier who discovered healthcare. In 1995 he and Lars Backsell bought a Pharmacia tablet plant in Årsta out of a management buyout. He ran Recipharm as CEO from 2008 to 2021 and grew it into a global top-five CDMO with a workforce in the thousands, before the EQT transaction gave him the capital he has been redeploying ever since. He founded Flerie in 2011 while still running Recipharm, meaning he has been at this for fifteen years. He chairs or sits on the boards of a long list of portfolio companies — Prokarium, Amarna, NorthX, Chromafora, Nanologica, KAHR. This is not a man reading investment memos; this is an operator in the boardrooms.

He holds roughly 58.9 million shares, about two-thirds of the company, held through his own vehicles. His interests and mine are, structurally, the same interests.

The clearest test of that alignment came this May, and it is the single fact that moved me from “interesting” to “small position.” Flerie conducted a bonus issue of 11,888,785 shares. Eldered waived his entire allocation without compensation and returned the shares he was entitled to for cancellation. The net effect was to increase every other shareholder’s ownership of the company by approximately 15.5% — a material, voluntary transfer of value from the controlling owner to the minority. I have covered a lot of founder-controlled companies. Transfers almost always run the other way.

CEO Ted Fjällman — a scientist by training who came up through the portfolio before joining as venture partner in 2018 — runs investment operations with a team of fewer than ten people. The cost ratio was 1.2% of NAV in the last quarter. For access to a curated private portfolio, that is materially cheaper than the two-and-twenty you would pay a venture fund, and unlike a fund, there is no lock-up and no capital call.

4. Financials, margins and recent developments

Let me be direct about the conventional metrics: they do not apply here, and any screen that ranks Flerie on them will produce nonsense.

Reported revenue in 2025 was SEK 2.5 million — literally noise. The reported net loss for 2025 was SEK 541 million, and the trailing loss per share sits around SEK 9. Return on equity is negative. EBITDA margin is meaningless. If you run this company through a standard quality screen, it fails every test, which is a significant part of why it is priced where it is.

The metric that matters is NAV per share, and its trajectory has been poor:

That is a decline of roughly 27% in nineteen months. Q4 2025 alone took 13.8% out of NAV per share through write-downs. There is no way to dress that up: the value creation has not happened yet, and the marks have gone the wrong way.

What I do take from the more recent data is the rate of change. The Q1 2026 quarter was -3.9%. Q2 2026 was -2.9%, with net loss of SEK 92 million and EPS of SEK -1.02. July was +0.3% — the first positive month in a while. The bleeding has slowed markedly from the Q4 2025 shock. Whether that is a bottom or a pause, I genuinely do not know, but a stabilising NAV is the precondition for everything else in this case.

Liquidity is adequate but no longer abundant. Cash stood at SEK 585 million after Q1 (helped by a SEK 76 million directed issue done at market price with no discount, plus SEK 132 million from selling Lipum shares), and SEK 306 million at the end of Q2 — about 9.0% of NAV — after funding the redemption programme and portfolio commitments. There is no debt of consequence, and no dividend. Management’s stated position is that this is sufficient to fund milestones in the existing portfolio. It is enough; it is not a war chest, and a serious new commitment would likely require a realisation first.

5. Valuation: the discount and the two engines

Here is where I have to abandon my usual toolkit and build a different one — the same approach I used for VNV Global. No Fair PE. A sum-of-the-parts, and then an honest scenario tree.

Start with what you are buying. At SEK 21 against NAV of SEK 39.26, the headline discount is 46.5%.

But decompose the NAV. Roughly 9.1% sits in other assets and liabilities — predominantly cash — and 6.7% in assets related to portfolio companies. Call it SEK 6.16 per share of NAV that is cash or near-cash. No sane buyer applies a 46% haircut to Swedish kronor sitting in a bank account. If you value that at par, you are paying SEK 14.84 per share for an operating portfolio marked at SEK 33.10 — an effective discount on the actual assets of around 55%.

Put differently: the market is telling you that Flerie’s private marks are overstated by more than half. That is a strong claim. It might be right — private marks in a funding winter are notoriously sticky — but “more than half” is a lot, particularly for holdings like NorthX (a real plant with real customers) and Xspray (marked at a live, observable market price).

The two engines. In a holding company, your return has two sources, and they multiply. NAV growth compounds. Discount narrowing is a one-time re-rating. Getting both at once is where the outsized outcomes live.

Bear case. The funding winter persists, a couple of the larger private holdings take down rounds, NAV compounds at -10% annually to SEK 28.6 in three years, and the discount stays at 50%. Share price SEK 14.3. Roughly -12% per year.This is a real scenario, not a strawman.

Base case. NAV grinds forward at 5% annually — a couple of milestones hit, a couple disappoint — reaching SEK 45.5. Sentiment normalises enough that the discount narrows to 35%, still wide by investment-company standards. Share price SEK 29.5. Roughly +12% per year.

Bull case. The biotech cycle turns, one or two holdings deliver clean clinical or regulatory wins, NAV compounds at 12% to SEK 55.2, and the discount closes to 20% as bargain hunters and generalist investment-company money arrive. Share price SEK 44.1. Roughly +28% per year.

Blue sky. A single portfolio company gets acquired at a Cormorant-style multiple. On a market cap of only about SEK 1.8 billion, one exit at a few hundred million dollars is transformational rather than incremental — this is the leverage a small holding company gives you that a Kinnevik or a VNV cannot. NAV compounds at 20%, discount closes to 10%: north of 40% per year.

I weight the base and bull cases at maybe 60% combined, the bear at 25%, blue sky at 15%. That is not the 25%+ expected return I demand from a Fair PE candidate. But this is a portfolio, not a single company — the diversification across roughly two dozen shots on goal is doing work that no individual biotech position could — and the asymmetry is genuine. Sized appropriately, I think that combination earns a place.

6. Risks: why this opportunity exists at all

I want to be very clear that the discount is not a free lunch. Someone is on the other side of it, and they have arguments.

The marks may simply be wrong. Private holdings are valued at the last financing round or listed price. In a funding winter, the last round can be eighteen months stale and struck in a different world. If Flerie has to raise the next rounds at lower valuations, NAV falls to meet the price rather than the price rising to meet NAV. Q4 2025’s 13.8% write-down was exactly this dynamic in action, and it is the single most important risk in the case.

The mechanical selling from the redemption programme. The April 2026 oversubscription created a wave of frustrated sellers and knocked the stock roughly 15% in a session, permanently resetting the discount from the mid-teens to the twenties and then wider. That is a technical wound, not a fundamental one — but technical wounds in illiquid small caps take a long time to heal.

Liquidity, in both senses. Daily trading volume is thin, the free float is around one-third of shares, and the market cap is about SEK 1.8 billion. Getting in and out is a real cost, and the stock is effectively uninvestable for institutions of any size. That is precisely why the discount can persist for years.

No visible catalyst to close the discount. The redemption programme is gone. There is no dividend. There is no buyback. Without a mechanism, a wide discount can simply be the price for a very long time. This is the objection I find hardest to answer, and I do not have a clean rebuttal beyond “eventually, cheap assets find owners.”

Key person risk. Take Eldered out of the picture and much of the sourcing advantage, board influence and credibility goes with him. He was born in 1960. There is no obvious successor with the same network.

Sector beta. If small-cap biotech stays out of favour for another two years, none of the above matters. You will be right on the arithmetic and wrong on the outcome.

And a governance caveat that cuts both ways: two-thirds ownership means the founder’s alignment protects you — right up until the moment his interests diverge from yours, at which point you have no recourse whatsoever. The May bonus issue was strong evidence of good faith. It is evidence, not a guarantee.

7. Latest news and earnings

The Q2 2026 report, published 9 July, is the most recent full picture, and I found the CEO’s commentary unusually candid. NAV per share fell 2.9% to SEK 39.14; portfolio fair value came in at SEK 2,849 million; the net result was SEK -92 million against +174 million a year earlier. Cash of SEK 306 million represented 9.0% of NAV. Fjällman explicitly noted that despite the bonus issue transferring substantial value from the majority owner to everyone else, the gap between share price and NAV had widened to a full 48% by quarter end. When management is publicly naming the discount, they are aware of the problem.

On the portfolio, the newsflow is genuinely mixed, which is what an honest early-stage portfolio should look like:

• Prokarium reported positive interim data from its clinical study in bladder cancer — one of the more encouraging datapoints in the portfolio this year.

• Lipum was selected for a Horizon Europe grant of EUR 8 million to advance SOL-116 in rheumatoid arthritis. With the Lipum merger now complete and Flerie owning 100%, that non-dilutive funding flows directly to Flerie’s benefit.

• Xintela and AnaCardio both reported positive clinical progress.

• Strike Pharma received a EUR 1 million Eurostars grant for an in-vivo CAR-T project with South Korea’s AbClon.

• On the negative side, share price declines at Xspray Pharma and Xintela were direct drags on the quarter’s NAV.

• A merger with Biosergen is in progress, with Biosergen reporting its half-year results in August ahead of completion.

The most immediate catalyst is tomorrow. Xspray Pharma’s PDUFA date for Dasynoc — its improved amorphous dasatinib formulation for chronic myeloid leukaemia — is 25 August 2026. This has been a long road: a Complete Response Letter in October 2025 driven by GMP observations at a third-party manufacturer, resubmission in February, acceptance in March. A further CRL arrived for the sister candidate Nilopki (nilotinib) in June. Xspray represents 4.0% of Flerie’s NAV, so an approval is not transformational on its own — but it would be the first genuine commercial validation to emerge from this portfolio in some time, and sentiment on a name trading at a 46% discount is arguably worth more than the arithmetic.

Beyond that, the next NAV update lands in early September and Q3 results on 15 October 2026.

8. Conclusion: why I own a little of this

Let me tie the thread back to where I started.

I do not have an edge in early-stage biotech. I am not going to develop one. So the question is not “can I pick the winning molecule” — it is “can I buy a competently assembled portfolio of shots on goal, from someone who genuinely can pick, at a price that gives me a margin of safety even if I’m wrong about several of them.”

At SEK 21 against a NAV of SEK 39.26 — and an effective 55% discount once you strip out the cash — I think the answer is yes, in small size. What I am underwriting is not any single clinical outcome. It is: a founder with thirty years of operating credibility in exactly this industry, who owns two-thirds of the company and just voluntarily handed 15.5% more of it to the rest of us; a portfolio of roughly two dozen holdings across early-stage and commercial-stage, with co-investors consistently putting in three times what Flerie does; a cost structure of 1.2% of NAV with no lock-up; and an entry price where either NAV recovery or discount normalisation alone produces a decent return, and both together produce a very good one.

The investment cases, as I see them:

1. The discount-closing case. Nothing dramatic happens operationally, but biotech sentiment normalises and a 46% discount on a founder-aligned holding company becomes indefensible. Re-rating alone gets you a solid double-digit return.

2. The cycle case. The biotech funding winter ends, private marks recover, NAV compounds again, and you collect both engines at once.

3. The single-exit case. One holding gets acquired at a meaningful multiple. On a SEK 1.8 billion market cap, this is the option you get for free, and it is worth more here than in any large holding company.

4. The access case. Even ignoring all of the above, this is one of the very few liquid ways for a public-market investor to own private European and Israeli biotech at all — and the only one I know of currently available at half price.

What would make me sell: another Q4-2025-style write-down cycle showing the marks were fiction; any sign that the alignment I have described has reversed; or a NAV decline that outpaces the discount narrowing for two more years.

This is a small, high-risk, illiquid position — a satellite holding, sized so that being wrong costs me a little and being right matters. That is exactly how I think a generalist should approach the one sector where he knows he is not the expert.

If you know Flerie or hold it yourself, I would genuinely like to hear your view — particularly from anyone closer to the science than I am.

Risk disclaimer

This article is my personal opinion and is provided for information and educational purposes only. It is not investment advice, not a recommendation to buy or sell any security, and not a solicitation of any kind. I am not a licensed financial adviser, and nothing here takes account of your personal circumstances, risk tolerance, tax situation or investment objectives.

Investing in equities involves the risk of total loss of capital. This applies with particular force to Flerie AB: a micro-cap, illiquid, founder-controlled holding company whose net asset value depends on the valuation of unlisted early-stage biotechnology companies. Such valuations are inherently subjective, are typically based on the last financing round, and can be revised sharply downward. Early-stage biotech companies frequently fail outright. Currency risk (SEK) applies to non-Swedish investors. Past performance is not indicative of future results.

All figures cited are drawn from company reports and public sources and were correct to the best of my knowledge at the time of writing. Data can change quickly — please verify everything independently before acting.

Conflict of interest: I hold a position in Flerie AB. The stock is a constituent of the portfolio of the cost-efficient Haas Invest4 Innovation investment fund (invest4.net), which I manage. I may buy or sell shares at any time without prior notice. Please always do your own research and, if in doubt, consult a licensed adviser.

— Philipp Haas, investresearch.net

I first looked at ZhongAn Online back at its IPO in 2017, and I did what most disciplined investors did back then: I passed. The stock came public at HK$59.70, ran to almost HK$98 within days, and was valued as though it had already conquered global insurance. It hadn’t. It was losing money on every policy it wrote.

What followed was one of the more brutal de-ratings I have watched in Asia. Today the share sits around HK$9.50 — roughly 90% below the all-time high, near the lowest level it has ever traded, with a market capitalisation of about HK$16bn (~US$2bn). Meanwhile net assets at the end of 2025 stood at RMB 25.4bn, or roughly HK$16.60 per share. Read that twice: the market is currently paying around 0.57 times book for a company that has now delivered five consecutive years of underwriting profit and just tripled its adjusted earnings.

That is the setup that makes me interested. Not a story stock. A cheap, misunderstood, structurally advantaged financial that the market has decided to file under “China risk, forget it.”

My thesis in one paragraph: ZhongAn has quietly completed the transition every insurtech promises and almost none delivers — from growth-at-any-cost premium accumulation to disciplined underwriting. The combined ratio has gone from a disastrous 133% in the loss-making years to 95.8% in 2025. The health ecosystem is compounding above 20% with a 92% combined ratio. ZA Bank, the digital bank nobody believed would ever make money, turned its first full-year profit. And the drag on reported earnings — the consumer-finance book — is being deliberately shrunk by management, which depresses 2026 optics while improving the quality of everything underneath. On my Fair PE framework I arrive at a fair P/E of 21 and a potential return of roughly 32% per annum over three years. ZhongAn is a position in the Haas Invest4 Innovation Fund.

Let me walk you through it properly.

1. Product, business model, brand and moat

What the company actually does. ZhongAn was founded in October 2013 in Shanghai as China’s first fully online property & casualty insurer. No branches. No agents. No paper. Every policy is sold, underwritten, serviced and paid out through an app, a mini-programme or an API embedded in someone else’s platform. It became the first fintech listed on the Hong Kong Stock Exchange in September 2017.

The business is organised into four insurance ecosystems plus two adjacent businesses:

• Health — the crown jewel. FY2025 gross written premiums of RMB 12.68bn, up 22.7%, at a combined ratio of 92.1%, a 3.6 point improvement. The flagship “Personal Clinic Policy” (尊享e生) million-yuan medical product has now been through more than 25 iterations in a decade. Alongside it sits “Zhong Min Bao,” a deliberately inclusive line covering pre-existing conditions, the elderly and chronic patients — a segment traditional Chinese insurers have historically refused to touch.

• Digital lifestyle — the original business. Return-shipping insurance sold through e-commerce, plus pet, flight delay, travel and cancellation cover. Still the largest premium pool at RMB 15.97bn, but it shrank 1.4% in 2025 at a combined ratio of 99.9%. In plain language: this segment now contributes scale and data, not profit.

• Auto — the fastest-improving line. Insurance service revenue of RMB 2.37bn, up 28.4%, combined ratio 93.1%. Family cars make up 87.9% of auto premiums, and NEV insurance is growing rapidly off a small base. ZhongAn secured independent operation of compulsory traffic insurance in Shanghai and Zhejiang in December 2024 — a genuine regulatory unlock.

• Consumer finance — credit guarantee insurance for online lenders. This is being wound down on purpose. FY2025 premiums of RMB 4.3bn, down year-on-year; the outstanding loan balance has been falling from RMB 27.7bn. Management guides to a 50–60% decline in 2026 and further contraction into 2027/2028.

• Technology — selling ZhongAn’s core insurance systems, AI and blockchain stack to other insurers domestically and across Asia. Still loss-making, but losses have narrowed sharply.

• Banking — ZA Bank in Hong Kong, held via ZA Global. More on this below, because I think it is the single most underappreciated asset on the balance sheet.

How they make money. Two engines, exactly as Buffett taught us to look at insurers. First, underwriting profit: premiums earned minus claims minus expenses. In 2025 that was RMB 1.41bn, up 42.5%, from a combined ratio of 95.8% — 57.1% loss ratio, 38.7% expense ratio. Second, investment income on the float: total investment income rose 59.1% in 2025 as Chinese capital markets recovered. Together those produced adjusted net profit attributable to shareholders of RMB 1.80bn, up 198.3%.

That combined ratio number deserves emphasis because it is the whole investment case in a single metric. An insurer running above 100% is paying for the privilege of holding your money. An insurer running at 95.8% is being paid to hold it, and then earns a return on it as well. ZhongAn crossed that line and has stayed on the right side of it for five straight years.

Brand. ZhongAn is now the 8th largest P&C insurer in China by gross written premiums, and it is gaining share. That is remarkable for a company with zero physical distribution competing against giants like PICC and Ping An. The brand strength is concentrated where it matters: young, digitally native customers who will never walk into a branch, and who ZhongAn acquires at a fraction of the cost of an agent-led insurer. Proprietary direct-to-consumer channels now generate over a fifth of total premiums, and that share is climbing — which matters enormously, because D2C premiums carry far better economics than premiums acquired through a platform partner taking a cut.

The brand is not spotless. There are more than 20,000 complaints on China’s Heimao consumer platform, clustered around “first month RMB 0” and “first month RMB 1” promotional pricing with automatic renewal — customers feel they were enrolled into recurring charges they did not fully understand. A pet insurance advertisement in May 2026 drew public criticism for tasteless imagery. Neither is existential, but I track them, because in Chinese financial services regulatory patience with consumer-conduct issues is finite.

Moat. This is where I push back on the lazy bear case that “insurance is a commodity, anyone can copy it.” Three things are genuinely hard to replicate:

Cost structure. ZhongAn’s expense ratio reflects a business with no branch network, no agent commissions and no legacy IT. Every point of expense advantage flows straight into either margin or price competitiveness. A legacy insurer cannot simply decide to have this — it would have to dismantle its distribution.

Embedded distribution. ZhongAn sits inside the checkout flow of e-commerce platforms, travel apps, telecom carriers and lending platforms. The insurance is invisible; it is a checkbox. Winning that integration requires the API, the underwriting speed and the regulatory licence simultaneously.

Data and claims automation. This is not a slide-deck claim. In auto, 88.2% of claims are now self-reported online, over half are handled through instant video assessment using NFC “tap-to-report,” AI damage assessment runs as fast as 116 seconds, and claims under RMB 10,000 settle in about 13.3 minutes. The in-house AI platform “Lingxi” runs a growing fleet of agents across the entire insurance value chain. In a business where claims handling is the largest cost line and the primary driver of churn, this is a durable operating advantage.

Where the moat is thinnest is digital lifestyle. Return-shipping insurance is close to a commodity, priced razor-thin, and its 99.9% combined ratio proves it. I do not underwrite this position on that segment.

2. The market

Chinese property & casualty insurance is enormous and still structurally underpenetrated relative to developed markets. Insurance density and penetration in China remain well below the US, Japan or Germany, and the gap closes as household wealth and consumption grow. Within that, the two sub-markets ZhongAn is levered to are the attractive ones.

Health insurance is the standout. China’s public healthcare system covers a basic floor; anything beyond it — imported drugs, private wards, cancer therapies, medical devices — is out of pocket. Commercial health insurance fills that gap and grows structurally faster than GDP as the population ages and as healthcare payment reform pushes more cost onto private cover. ZhongAn’s 22.7% growth here is not a share-grab; it is riding a genuine expansion.

Pet insurance is a smaller but explosive adjacency. Chinese dog and cat household penetration sits in the high teens and is climbing, and it remains roughly half the level of Japan or Thailand — countries with comparable urban living conditions. Pet insurance in China is where pet insurance in Japan was fifteen years ago.

NEV auto insurance is the third leg. China dominates global electric vehicle production, NEVs are taking an ever-larger share of the domestic fleet, and NEV policies carry higher premiums than combustion equivalents. ZhongAn has an early, growing position.

Political intervention. I will not sugar-coat this. Insurance in China is a licensed, closely supervised industry, and Beijing intervenes. The “报行合一” reform — forcing insurers to charge what they file with the regulator — compressed distribution expenses across the entire industry. That happened to help ZhongAn (its auto expense ratio fell 2.8 points), but it demonstrates that pricing is not fully in management’s hands. Consumer-finance and credit-guarantee insurance have been under sustained regulatory scrutiny for years. And the Hong Kong listing means the shares are hostage to the broader geopolitical mood in a way the underlying business is not.

Cyclicality. The old argument that “insurance is defensive because people insure regardless of the economy” is only half right for ZhongAn, and I want to be honest about that. Health and auto premiums are genuinely resilient. But return-shipping insurance is tied directly to e-commerce volumes, credit-guarantee insurance is tied to consumer credit quality, and — critically — a large share of earnings comes from investment income on the float, which is tied to Chinese equity and bond markets. The 2025 profit explosion was partly an underwriting story and substantially a market story. Investors who model ZhongAn as a bond-like defensive will be surprised in both directions.

3. Culture and management

ZhongAn’s founding story is legendary and, frankly, over-told. It was launched in 2013 with backing from Jack Ma’s Alibaba/Ant, Pony Ma’s Tencent and Ma Mingzhe’s Ping An — “the three Mas.” It was the first insurance licence in China granted to a purely online entity, and that pedigree is genuinely what made the venture possible.

But the ownership picture today is very different, and anyone still buying this stock for the “three Mas” narrative is buying a memory. Over the past several years the original strategic holders have materially reduced their positions. In 2026, founding shareholder Ou Yaping consolidated family holdings and returned as the single largest shareholder with roughly 11.94%, ahead of Ping An at around 8.90%. The centre of gravity has shifted from a consortium of Chinese tech champions to a founder-anchored ownership structure.

I regard that as a net positive, with a caveat. The positive: founder-led companies with concentrated skin in the game make better long-term capital allocation decisions than committee-governed joint ventures, and this is exactly the profile I look for. The caveat: the strategic distribution advantages that came bundled with Alibaba and Tencent ownership are no longer guaranteed by shareholder alignment — ZhongAn must now win that shelf space commercially, every year.

On the entrepreneurial dimension, management has earned my respect in a specific and unglamorous way: they have shown willingness to shrink. Voluntarily cutting a profitable-looking premium line by 50–60% because the risk-adjusted returns are not good enough is the single hardest thing to do in insurance, where every incentive pushes toward writing more. The consumer-finance runoff will make 2026 headline growth look mediocre. They are doing it anyway. That is long-term thinking.

The same is true of ZA Bank. Building Hong Kong’s first virtual bank meant absorbing years of losses to reach the scale where a digital bank’s unit economics finally work. In 2025 it worked: net revenue up 62.7%, cost-to-income ratio down 32 percentage points, and the first full-year net profit in the bank’s history. That is what patient capital deployment looks like.

Balance sheet discipline is intact: comprehensive solvency ratio of 242% at end-2025 and 241% at the end of Q1 2026. Moody’s rates the insurance financial strength Baa1 with a positive outlook, and AM Best affirmed A- (Excellent) and revised its outlook to positive in November 2025. Two independent rating agencies moving in the same direction, while the equity market moves the other way, is precisely the kind of divergence I hunt for.

4. Financials and recent developments

Reported net profit came in around RMB 1.10bn — the gap to the adjusted figure is largely a one-off, non-cash impairment of roughly RMB 750m at ZA Global, the international arm. I use the adjusted number for the earnings power discussion and the reported number for the P/E screens, and investors should be aware that many data providers show only one of the two, which is part of why this stock screens inconsistently.

Margins. Let me be precise, because insurance margins are routinely misread. Net income margin on insurance service revenue was roughly 3.3% reported / 5.4% adjusted in 2025. That sounds thin next to a software company, and it is — but it is normal for P&C, where revenue is gross premium flow, not value-added. Return on equity on the 2025 adjusted result was approximately 7–8%, up sharply from near-zero in 2024, and still well below what this business should earn. EBITDA is not a metric I use for insurers or banks and I would encourage readers to ignore it here entirely; combined ratio, ROE and solvency are the numbers that matter.

The problem: Q1 2026. This is why the stock is where it is. First-quarter insurance business income was RMB 7.93bn, but net profit collapsed to RMB 171m, down 69.9% year-on-year from RMB 569m. Net assets were essentially flat at RMB 25.29bn. The market read that as the 2025 recovery unravelling.

I read it differently, and this is the crux of my thesis. Q1 2025 benefited from an exceptionally strong investment result; Q1 2026 did not. Simultaneously, the consumer-finance runoff was hitting premium and fee income while the fixed cost base had not yet adjusted. The underwriting engine — which is what I am actually buying — did not break. But quarterly earnings for an insurer with a large investment portfolio are noisy by construction, and the market punished the noise as though it were signal.

At the April 2026 briefing, management guided 2026 as follows: excluding consumer finance, total premium growth above 10%; auto around +30%; health above 10%; digital lifestyle low double-digit; consumer finance down 50–60%. Sell-side reacted in both directions — Huatai cut FY2026/27 EPS estimates to RMB 0.75 and RMB 0.84 and trimmed its target to HK$23, while Guosen and BofA raised their forecasts. Consensus across roughly fourteen analysts remains a Buy with an average target near HK$20.

One more recent development worth noting: ZhongAn has taken a stake in an offline insurance agency, which reads to me as an acceleration of the auto push. Pure-online distribution has limits in motor insurance, where local presence still matters for acquisition and claims. If that is the reasoning, it is pragmatic rather than dogmatic — and I prefer pragmatic.

5. Valuation: the Fair PE

Here is how I think about what this company is worth.

I do not value ZhongAn on trailing reported earnings, because 2025 contained an unusual investment tailwind and a one-off impairment, and 2026 contains a deliberate, self-inflicted revenue reduction. Both distort the picture in opposite directions. I value it on a normalised earnings base three years out, applying my Fair PE multiple.

The Fair PE for ZhongAn is 21.

Why 21 and not higher or lower? On the positive side: structural growth in the underlying markets, a genuine and widening cost moat, five years of underwriting profitability, a founder-anchored owner, and an underappreciated banking asset. Against that: Hong Kong listing risk, regulatory exposure, an earnings stream partially dependent on Chinese capital markets, a dilution track record, and a low-margin segment that still accounts for the largest share of premiums. A US or European insurtech with these operating characteristics would carry 25–30x. A generic Chinese P&C insurer would carry 8–10x. Twenty-one reflects the quality without pretending the geography and the volatility are free.

The earnings bridge. Working from an FY2026 base of roughly RMB 0.75 per share, I assume: health compounding in the high teens to low twenties; auto growing around 30% off a small base with a combined ratio in the low nineties; digital lifestyle flat-to-modest and margin-neutral; the consumer-finance drag fully absorbed by 2027 and gone by 2028; the technology segment reaching breakeven; and ZA Bank contributing a growing, no-longer-negative earnings line. I deliberately assume no repeat of the 2025 investment income windfall — a normalised, unremarkable investment return.

That produces a normalised FY2028 EPS of approximately RMB 0.95, or about HK$1.05.

Fair PE 21 × HK$1.05 = fair value of approximately HK$22 per share.

Against the current price of about HK$9.50, that is roughly 130% total upside, or about 32% per annum over three years.

Sensitivity, because a single point estimate is dishonest:

• Bear case — consumer-finance runoff overshoots, health growth halves under competition, investment returns stay poor: normalised EPS around RMB 0.60 (HK$0.66), fair value ~HK$14, still roughly 13% per annum from here. The downside is protected by book value, not by hope.

• Base case — as above: ~32% per annum.

• Bull case — health sustains 20%+, auto scales to a genuine third profit engine, ZA Bank re-rates independently, investment returns normalise upward: EPS around RMB 1.30 (HK$1.43), fair value ~HK$30, or roughly 47% per annum.

The sum-of-the-parts cross-check. This is the part that made me size the position rather than just watch it. ZA Bank — Hong Kong’s first and largest digital bank, now profitable — has been valued by at least one sell-side house at around US$2bn. ZhongAn’s economic interest of roughly 43% is therefore worth in the order of US$0.9bn, or close to HK$7bn. Against a total market capitalisation near HK$16bn, that single stake accounts for more than 40% of the entire company’s value.

Strip it out and the core insurance business — the 8th largest P&C insurer in China, growing health premiums above 20%, writing at a 95.8% combined ratio — is being valued at roughly 0.3–0.4 times book. Listed Chinese P&C peers trade around 1.2 times book. I have run a lot of sum-of-the-parts analyses over the years and rarely found a gap this wide on a business that is actually profitable.

6. Risks: why is it this cheap?

A stock does not fall 90% from its high and 55% in a year by accident. I owe you a clear account of what the market is worried about, because if I cannot articulate the bear case better than the bears, I have no business owning the shares.

The stablecoin round trip. In mid-2025, Hong Kong passed its Stablecoin Ordinance. ZA Bank was the first digital bank in Hong Kong to offer reserve banking services to stablecoin issuers, and ZA Global led a US$40m financing round in stablecoin infrastructure firm RD Technologies. The stock ripped — up more than 40% in a month, then more than 70% at the peak. Management, sensibly, issued equity into that strength. Then ZhongAn did not receive a stablecoin issuer licence, the narrative evaporated, and the shares gave back everything and more. Investors who bought the story and the placement are deeply underwater and, understandably, angry. That overhang of disappointed shareholders is real and it takes time to clear.

The Q1 2026 earnings shock. A 70% year-on-year profit decline is the kind of headline that triggers systematic selling regardless of the underlying explanation. Many holders never got past the number.

Deliberate shrinkage looks like decline. When a company guides one of its four segments down 50–60%, screens flag it as a business in trouble. Distinguishing “shrinking because it must” from “shrinking because it chose to” requires reading the transcripts. Most of the market does not.

The largest segment earns nothing. Digital lifestyle is 45% of premiums at a 99.9% combined ratio. Every bear note leads with this, and the point is fair: if return-shipping insurance keeps contracting industry-wide, ZhongAn loses scale and data without losing much profit — but the optics are ugly and the growth rate suffers.

Dilution. A 13% dilution executed near a speculative high, followed by a halving of the share price, is not something shareholders forget quickly. If capital needs recur, the market will assume more of the same.

China and Hong Kong risk. Currency, geopolitics, regulatory intervention, the persistent valuation discount applied to Hong Kong-listed Chinese financials. Ownership changes at the top of the shareholder register add uncertainty about strategic direction. And the auditor transition, while probably routine, is one more thing for a nervous market to fixate on.

Investment income dependence. If Chinese equity markets are flat or down for two years, a meaningful chunk of earnings simply is not there, regardless of how well the underwriting performs.

Where does the opportunity come from, then? From the gap between what is actually deteriorating (one deliberately-shrunk segment, one quarter of weak investment returns, one failed narrative) and what the price implies (that the whole enterprise is worth less than its accounting equity). Markets are very good at pricing direction and very bad at pricing the difference between temporary and permanent. That is the entire game.

7. Latest news and the upcoming catalyst

The near-term calendar is unusually well-defined, which is rare and useful.

ZhongAn’s board meets on 25 August 2026 to approve interim results for the six months to 30 June 2026 — and, notably, to consider the recommendation of an interim dividend. That second item is the one I would flag hardest. ZhongAn has never paid a dividend. It has been building capital, absorbing losses at ZA Bank and the technology arm, and reinvesting. The mere fact that a dividend is on the agenda signals a company that believes it has passed peak capital intensity. Whether or not they declare one, the discussion itself is a change of posture.

What I will be watching in those numbers, in order of importance:

1. Combined ratio. Does it hold below 96%? Comparison base is 95.6% in H1 2025. Anything at or below that confirms the underwriting engine is intact and Q1’s weakness was investment-driven.

2. Health ecosystem growth and its combined ratio. This is the quality engine. I want to see continued double-digit growth at or below 92%.

3. Auto. Guidance was +30%. Delivery would establish auto as a credible third profit centre.

4. The consumer-finance runoff pace. Faster is better, even though it hurts the headline.

5. ZA Bank’s contribution. It made HK$49m in H1 2025 before finishing the full year with a much slimmer profit. I want to see H1 2026 profitability sustained, not lumpy.

6. Technology segment losses. Narrowing losses here are pure incremental earnings — the R&D is already spent.

Other developments in the current news flow: the shareholder restructuring that returned the founder to the top of the register; the auditor change to Deloitte; the offline agency stake supporting the auto expansion; and continued build-out of the AI infrastructure across claims and underwriting. Sentiment on Chinese fintech Twitter and in the Hong Kong retail community remains poor — the recurring phrase in shareholder forums is that the company “lost the wife and the army” on the stablecoin adventure. When retail sentiment is that sour on a profitable company trading at 0.57x book, I pay attention rather than agree.

8. Conclusion

Let me return to where I started. In 2017 ZhongAn was a fantastic story attached to a terrible business at an absurd price. In 2026 it is a mediocre story attached to a genuinely improved business at a very cheap price. I would take the second combination every single time, because stories re-rate and businesses compound, but you only get paid for both when the entry price is wrong.

The three investment cases I see, in ascending order of ambition:

The value case. You are buying an insurer at roughly 0.57 times book with a 242% solvency ratio, positive rating agency momentum from both Moody’s and AM Best, five consecutive years of underwriting profit, and a bank stake worth over 40% of the market capitalisation. Even if ZhongAn never grows again, a re-rating to book value alone is roughly 75% upside. This is the floor, and it is a reasonably hard one.

The normalisation case — my base case. The consumer-finance drag rolls off by 2028. Health and auto do what management has guided. Technology reaches breakeven. Investment returns are merely average rather than exceptional. Normalised EPS reaches around RMB 0.95, the market applies my Fair PE of 21, and the share reaches roughly HK$22. That is approximately 32% per annum over three years, and it requires no heroics — only that management executes a plan they have already articulated and largely begun.

The franchise case. China’s commercial health insurance market expands for a decade, pet insurance follows the Japanese trajectory, NEV insurance scales with the Chinese auto fleet, ZhongAn’s AI-driven cost advantage widens rather than erodes, and ZA Bank becomes a separately valued digital financial platform in Hong Kong rather than a line item. In that world the company earns well above my normalised estimate and deserves more than 21 times. I am not underwriting this case, but I am not paying for it either — which is exactly how a free option should be acquired.

What makes this a Haas Invest4 Innovation position rather than a watchlist name is the asymmetry. The downside is anchored by tangible book value and a solvency ratio near 240%. The base case pays roughly 32% per annum. And the catalyst is dated: 25 August 2026, with a dividend discussion attached.

That divergence is the opportunity. I own it, and I intend to keep owning it.

Risk disclaimer

This article represents my personal opinion and is intended for information and educational purposes only. It does not constitute investment advice, a recommendation, or an offer or solicitation to buy or sell any security. I am not your investment adviser and I do not know your financial situation, objectives or risk tolerance.

Equity investments carry the risk of total loss of capital. ZhongAn Online is a Hong Kong-listed Chinese financial company and carries above-average risk, including but not limited to: significant share price volatility; currency risk (HKD/RMB versus EUR); geopolitical and regulatory risk relating to China and Hong Kong; the risk of further capital increases and dilution; earnings dependence on capital market returns; regulatory intervention in insurance pricing and distribution; and liquidity and bid-ask spread risk for European investors trading a Hong Kong-listed security. All forecasts, estimates and valuation figures presented here are my own assumptions and may prove materially wrong. Past performance is not indicative of future results.

Financial data referenced is drawn from company reports, earnings calls and publicly available market data and is believed to be accurate as at the date of publication, but no warranty is given. Please always conduct your own research and consult a licensed adviser before making any investment decision.

Conflict of interest disclosure: ZhongAn Online (6060.HK) is held in the portfolio of the cost-efficient Haas Invest4 Innovation Fund (invest4.net), which I manage, and may also be held in my private accounts and in portfolios I manage elsewhere. I may buy or sell the security at any time without notice. This constitutes a conflict of interest, and you should weigh my views accordingly.

Philipp Haas — investresearch.net

Introduction: The Asset Class Private Investors Are Not Supposed to Reach

There is one corner of the market that ordinary private investors are structurally locked out of: venture capital.

If you want to put money into pre-IPO technology companies in Europe, you normally need a notary appointment, a six-figure minimum ticket, an LP commitment locked up for a decade, and enough of a network to see the deals that matter in the first place. I spent part of my early career on that side of the table — at a venture capital firm and later at a family office — and the mindset never left me. What I do today is take that venture lens and apply it to listed companies. Occasionally, though, the two worlds collide, and you find a business where you can buy the venture portfolio itself through a normal brokerage account.

VNV Global is exactly that. It is a Stockholm-listed investment company that has been backing network-effect businesses since it was still called Vostok Nafta, back in 1996. Its historic wins are the kind of thing VC firms print on the front page of a fundraising deck: Avito, held from 2007 to 2019, returned 34 times invested capital at a 37% IRR. Tinkoff Bank delivered 8x at a 42% IRR. Hemnet, the Swedish property portal, another 8x at a 50% IRR. This is not a firm that has never returned capital — a bar that a surprising number of German venture funds still fail to clear.

And yet, as I write this in August 2026, the shares change hands at roughly SEK 16.6 against a reported net asset value of SEK 34.97 per share. That is a discount of about 53%, and it is not far from the widest level in the company’s listed history. The entire market capitalisation is around SEK 2.1 billion, or roughly USD 220 million, for a portfolio marked at USD 484 million.

So here is the thesis in one paragraph. VNV is a concentrated, cleaned-up portfolio of six European growth businesses — carpooling, micro-mobility, rental marketplaces, men’s health, grocery, beauty bookings — that collectively grew their revenue 32% last year and crossed into aggregate profitability. Their combined pro-rata value is roughly SEK 27 per VNV share, and you pay SEK 16.6 for the whole company. The market is pricing in either that the marks are fiction, or that the discount is permanent, or both. My view is that the marks are conservative rather than aggressive, that the discount is explainable but not permanent, and that the two events most likely to break it — a BlaBlaCar listing and a Voi exit — are both closer than they were twelve months ago.

This is a high-risk, high-variance situation, not a compounder you buy and forget. Let me walk through why I think the risk is being mispriced anyway.

1. Product, Business Model, Brand and Moat

What the company actually does. VNV buys minority stakes in private, high-growth companies and holds them for a long time — typically five to twelve years. It is stage-agnostic, from seed through late growth rounds. It does not operate the businesses; it allocates capital, sits on boards, and waits for a liquidity event. The self-description is “patient capital for network-effect businesses,” and unlike most marketing lines, this one is actually visible in the portfolio.

How it earns money. Purely through the change in value of its holdings — there is no operating revenue in any meaningful sense. Reported “earnings” are almost entirely unrealised revaluations, which is why the income statement is close to useless as an analytical tool. In the first half of 2026 the group reported a net result of minus USD 85.1 million, essentially all of it from writing down BlaBlaCar and Voi on lower peer multiples. Operating expenses ran at USD 4.5 million for the half-year against a NAV of USD 461 million — call it 1.9% of NAV annualised, with roughly seven employees. That is lean in absolute terms but not cheap relative to a shrunken asset base, and it is a genuine drag I do not want to gloss over.

A second income stream is now being built. Management is establishing a regulated fund structure — internally referred to as VNV 2.0 — to manage outside money using VNV’s deal flow, generating fee income without deploying much of its own balance sheet. The stated intention is that VNV’s own liquidity keeps going into buybacks. Full details are due at the Capital Markets Day in Stockholm on 16 September 2026. Structurally this matters more than it sounds: holding companies get discounts, asset managers get multiples.

Brand. Among Nordic investors VNV is well known and heavily retail-owned — more than 20,000 private shareholders hold roughly 55% of the register. Outside Sweden, essentially nobody covers it; two analysts follow the stock. Within the European venture ecosystem the brand is real and functional: the firm gets into rounds it wants to be in, and portfolio founders take its calls. But as a consumer or institutional brand, it carries scar tissue, and I will come back to that.

Moat. Let me be blunt: VNV’s own moat is thin. Deal flow, reputation and permanent capital are advantages, not fortifications, and any well-funded competitor can bid against it. The moats that matter sit one level down, inside the holdings:

• BlaBlaCar has the strongest one — a genuine two-sided liquidity network. In long-distance carpooling, the platform with the most drivers gets the most passengers, which attracts more drivers. Nobody has broken that loop in twenty years of trying, and BlaBlaCar filled close to five empty seats per second in 2025.

• Voi’s moat is weaker and different in kind: it is built on city tenders and licences. Winning a municipal concession creates a local quasi-monopoly for a period of years, and Voi has been notably good at winning them. But the moat is granted by regulators, not earned from customers, and what regulators give they can withdraw.

• HousingAnywhere (medium-term rentals) and Bokadirekt (Swedish beauty and wellness bookings) are classic marketplace network effects with embedded payments, the stickiest of the smaller holdings.

• Numan (men’s and women’s digital health) and Breadfast (Egyptian online grocery) are execution and brand stories rather than moat stories. Numan in particular sells into a market where the product itself is increasingly commoditised.

2. The Markets

The portfolio breaks down roughly 51% mobility, 27% marketplaces and 11% digital health, and — this is the biggest single change from VNV’s past — it is now overwhelmingly a developed-market portfolio, concentrated in Europe. The emerging-market swashbuckling that defined the Vostok era is largely gone. There is no Russian exposure left; those positions were written to zero after February 2022 and are not coming back. Breadfast in Egypt is the main frontier-market position remaining.

Shared mobility is the dominant exposure and it is a genuinely large, structurally growing market. Long-distance travel and short urban trips are both enormous spending categories, and both have a durable tailwind: young Europeans in cities are buying fewer cars, fuel and rail prices push travellers toward cheaper alternatives, and municipalities want fewer private vehicles. Cyclicality is moderate — travel demand wobbles with consumer confidence, and micro-mobility has a pronounced summer/winter seasonality — but neither business collapses in a downturn. Carpooling arguably benefits from one.

The problem is political exposure, and it is not theoretical. Micro-mobility operates entirely at the pleasure of city councils. Paris removed shared e-scooters outright after a 2023 referendum, and every European city runs its own permit regime with its own rules on parking, speed and fleet caps. This is the single most policy-dependent business in the portfolio, and any investor in VNV is implicitly underwriting a tender-by-tender regulatory grind across dozens of jurisdictions.

Marketplaces — rentals, second-hand goods, service bookings — remain the most attractive market structure I know: capital-light, high-margin at scale, low political interference beyond the ordinary consumer-protection layer.

Digital health is the most volatile of the three. Numan sits in the UK weight-loss and men’s health market, which has been whipsawed over the past year by GLP-1 pricing changes as manufacturers moved to direct-to-consumer models. The market is vast and the demand is not going away; the economics of who captures the value are genuinely unsettled.

One market-level risk applies across the whole book: capital allocation in technology has become extraordinarily concentrated in artificial intelligence. Money that once funded consumer marketplace and mobility rounds is now funding compute. That both suppresses the valuations of everything else and, at the margin, makes exits harder to arrange.

3. Culture and Management

Per Brilioth has run this business for nearly two decades. Björn von Sivers is CFO. The team is tiny — around seven people managing a half-billion-dollar portfolio — and the long-term incentive plan requires participants to buy shares with their own money and hold them through the measurement period. Lars-Åke Norling was elected chairman at the May 2026 AGM and promptly bought roughly SEK 3 million of stock in the open market. I like all of that. It is the structure you want: small, aligned, personally exposed.

The entrepreneurial instinct is intact. When BlaBlaCar secondary shares became available in 2023 at a price management considered well below fair value, they funded the purchase with a rights issue at SEK 20 per share, raising SEK 328 million and lifting the stake from 10.5% to 14.1%. Issuing your own equity at a record discount to buy someone else’s is an aggressive, high-conviction move — and given where BlaBlaCar sits today, it looks like the right one.

Now the honest part, because this is where the market’s scepticism comes from and it deserves a fair hearing.

VNV had a genuinely bad 2021 to 2023. Babylon Health — at one point the single largest position and pitched as an AI-driven, value-based transformation of healthcare — went to essentially zero in 2023 after a SPAC listing. Swvl, the Middle Eastern bus-hailing business, followed a similar SPAC-and-collapse arc. The Russian book went to zero. Management did not sell enough into the 2021 window when public comparables were at all-time highs, and shareholders have been diluted through multiple equity raises since the 2020 Stockholm listing. NAV per share has fallen from a 2021 peak around SEK 120–130 to SEK 34.97 today. The long-run NAV IRR since 2012, which the company itself put at roughly 11.8% in its 2025 materials, is now a fraction of the 25%-plus figure that used to headline the presentations.

I think the fair reading is this: VNV is very good at finding network-effect businesses early and demonstrably poor at selling them at the top. That is a common venture pathology, and it is exactly why buying the vehicle at a 53% discount — rather than paying NAV — is the only version of this trade that makes sense to me.

Capital allocation today is, to their credit, orthodox. Debt has been cut hard: borrowings fell from USD 46 million at the end of Q1 to USD 27.1 million at the end of Q2 2026, after a partial bond buyback of SEK 166.9 million at 104% of nominal settled in May. Buybacks continue — 2.97 million shares over the last twelve months, about 2.3% of the company, with most cancelled after the AGM. The CFO stated plainly that repurchases will continue while the discount is this wide. At a 53% discount every share bought back is accretive to NAV per share by roughly twice the cash spent. That is the correct use of marginal liquidity and they are doing it.

4. Financials, Margins and Recent Operating Developments

You cannot analyse VNV with a normal P&L, so the right approach is to look through to the underlying companies. Here the picture is considerably better than the share price implies.

On a pro-rata basis — VNV’s share of the numbers, not the gross figures — the top six holdings grew revenue from USD 74.9 million in 2023 to USD 130.4 million in 2025, a 32% increase year-on-year. Over the same period their aggregate adjusted EBITDA went from minus USD 5.9 million to plus USD 3.6 million. The portfolio crossed from cash-burning to cash-generating while still growing above 30%. As of Q2 2026, 71% of portfolio value sat in adjusted-EBITDA-positive companies, down from 81% a year earlier — but that decline is almost entirely the mechanical effect of selling Gett, which was profitable.

Holding by holding:

BlaBlaCar (25% of NAV, USD 121 million, 13.5% owned) carried 150 million passengers in 2025 across more than twenty countries, generating around EUR 2 billion in gross merchandise value with positive EBITDA. The important development is a deliberate margin trade: management is winding down the low-margin operated bus business, which should lift blended gross margin from roughly 50% to above 90%. Revenue will optically shrink; gross profit and cash conversion will not. The company is running ahead of budget year-to-date and pushing into new markets across four continents.

Voi (22% of NAV, USD 106 million, 20.8% owned) is the operational standout. Net revenue on a last-twelve-month basis through Q1 2026 reached EUR 188.3 million with adjusted EBITDA of EUR 29.7 million — a 15.8% margin — and adjusted EBIT positive at around EUR 1.4 million. Growth was 38% year-on-year in Q1, against Lime at 32% and Dott at minus 6%. Step back and look at the five-year arc: from EUR 27 million of net revenue in 2020 with a minus 83% EBITDA margin, to EUR 188 million with a positive 16% margin, and vehicle-level profit margin up from 31% to 57%. Cumulative rides passed 400 million, and the most recent 100 million took a single year versus more than three and a half years for the first 100 million.

HousingAnywhere (8%, USD 39 million, 26.2% owned) raised primary capital in Q1 2026 with VNV participating, grows revenue, and holds positive adjusted EBITDA. Its AI booking assistant now handles half of platform traffic — a real cost-to-serve story, not an AI press release.

Numan (7%, USD 36 million, 13.5% owned) launched its unified “Numan 2.0” platform on 30 June, combining men’s health, women’s health and diagnostics, and is preparing an oral Wegovy launch with an 18,000-person waiting list. The mark came down about 2% on GLP-1 market volatility.

Breadfast (6%, USD 30 million, 6.8% owned) runs 59 fulfilment points in Egypt serving close to half a million monthly users, with annualised gross transaction value of USD 270 million in December 2025 against USD 130 million a year earlier.

Bokadirekt (5%, USD 26 million, 15.8% owned) acquired Zoezi, a Swedish gym and fitness management system, adding roughly 10% to the top line, with payment revenue growing faster than bookings. Its mark rose 5% in the quarter on higher peer multiples.

The single most important thing to understand about the reported numbers: NAV moved down 16% in dollar terms in the first half of 2026 while the underlying businesses grew and improved. The write-downs came from falling public-market comparables feeding into model-based valuations, not from operating deterioration. That distinction is the entire investment case.

5. Valuation: There Is No Fair P/E Here — Only NAV

I normally value companies with a fair P/E applied to forward earnings. That framework is useless for a venture holding company, because reported earnings are just revaluations of unlisted assets. The only sensible anchor is net asset value, and the only sensible questions are: how good are the marks, how fast can the portfolio grow, and will the discount close?

Are the marks credible? This is where I have done most of my work, and the evidence points toward conservatism rather than optimism.

• BlaBlaCar is marked at USD 121 million for 13.5%, implying an equity value around USD 0.9 billion. The company’s last private round in 2024 valued it near USD 2 billion. VNV carries it at roughly half of the last transaction price — on a business that is now larger and structurally more profitable than it was then.

• Voi’s mark implies roughly USD 0.5 billion for a business with EUR 188 million of net revenue, 38% growth and a 15.8% EBITDA margin. Lime — slower-growing, loss-making, and carrying about USD 1 billion of current liabilities with a going-concern warning in its filing — listed on Nasdaq on 1 July 2026 at a valuation of roughly USD 1.66 billion, around two times its 2025 revenue. Lime’s revenue is gross and Voi’s is net, so the multiples are not directly comparable, but the direction is clear: the profitable, faster-growing private asset is not being carried at a premium to the unprofitable listed one.

• In the long tail, three data points from the last few quarters are worth more than any model: Flo raised at over USD 1 billion; the Veristable stake converted into Oura equity at a USD 10.9 billion valuation; and Tise was sold to eBay at a 65% premium to VNV’s carrying value. When third parties transact repeatedly above your marks, your marks are not the problem.

The base arithmetic. NAV per share is SEK 34.97. If the portfolio compounds at 15% annually for three years — which is below the growth rate the top six are currently delivering, and therefore assumes continued multiple compression eating into operating progress — NAV per share reaches roughly SEK 53. Add the accretion from repurchasing 2–2.5% of shares per year at a 50%-plus discount, worth a bit over 1% per year to NAV per share, and you get to roughly SEK 55.

Against a share price of SEK 16.6, the scenarios look like this:

The brief here was to assume the discount vanishes entirely in three years with a portfolio compounding at 15% — that is the bull line, and it produces roughly 49% annually, or a little over three times your money. I want to be honest that a full closure to zero discount is rare for a venture holding company; some discount is warranted permanently, because the assets are illiquid, the holding costs are real at around 1.9% of NAV, and the marks carry genuine uncertainty. My working case is the base line: something in the region of 30% a year if the portfolio performs and the discount normalises toward its historical range.

Note also the asymmetry in the bear case. For an investor to lose money over three years from here, the portfolio has to shrink and the discount has to stay at a record level. Flat NAV with an unchanged discount is roughly break-even. That is what a margin of safety looks like when it is expressed as a discount rather than a multiple.

6. Risks: Why the Stock Is Where It Is

A 53% discount is not a gift; it is a price the market has set for reasons. Here they are, as fairly as I can state them.

The credibility problem. Babylon went to zero. Swvl went to zero. Russia went to zero. Investors who trusted the marks in 2021 were badly hurt, and the market’s response has been to apply a blanket haircut to every number VNV publishes. That scepticism is earned, even if I believe it is now over-applied.

Marks follow public comps with a lag. The valuations are model-based, keyed to listed peer multiples. When technology multiples compress, NAV falls even if the businesses are thriving — precisely what happened in the first half of 2026. This cuts both ways: the same mechanism will lift NAV sharply if multiples recover, but it means reported NAV is not a stable anchor quarter to quarter.

Concentration. The top six are 78% of NAV; BlaBlaCar and Voi alone are 47%. A serious problem at either one is a portfolio event, not a position event.

No near-term catalyst from the core. Management has said explicitly it does not expect major exits from the top six during 2026. Smaller transactions are being negotiated, some at or above NAV marks, but the events that would definitively validate the big positions are not scheduled.

Liquidity and size. This is a roughly SEK 2.1 billion company with daily trading volume that can be measured in the low tens of thousands of shares. Getting in is easy; getting out in size, in a stressed market, is not. For a fund like mine this dictates position size, and for a private investor it dictates that this is a small, patient allocation and nothing more.

Currency. NAV is reported in dollars, the shares trade in kronor, and the underlying assets earn in euros, pounds and Egyptian pounds. Exchange rates alone move reported NAV by several percent per quarter.

The discount can simply persist. Nordic investment companies have traded at wide discounts for years at a stretch. There is no mechanism that forces convergence — only buybacks, exits and time.

So why does the opportunity exist? Because the market is pricing a portfolio of profitable, growing European platforms as though the marks are inflated, when the observable third-party transactions say the opposite; because the two things that would prove it — a BlaBlaCar or Voi liquidity event — have been perpetually “next year”; and because in a market obsessed with artificial intelligence, an illiquid Swedish holding company owning carpooling and scooters is about as far from the narrative as it is possible to be. That combination is where mispricing usually lives.

7. Latest News and Earnings

The second-quarter report on 14 July 2026 was, by recent standards, a stabilisation. NAV came in at USD 461 million, or USD 3.60 per share (SEK 34.97) — flat in dollars over the quarter and up 2% in kronor, after a first half that took NAV down 16% in dollars. The investment portfolio stood at USD 484 million, comprising USD 468 million of holdings and USD 16.3 million of cash, plus USD 9 million in liquidity-management instruments. Borrowings were cut to USD 27.1 million.

BlaBlaCar’s mark rose 1% to USD 121 million; Voi was flat to slightly down at USD 106 million; HousingAnywhere was held flat on a Q1 transaction; Numan slipped about 2%; Breadfast was unchanged; Bokadirekt rose 5%. Buybacks continued at around 600,000 shares year-to-date, with 2.97 million over the trailing twelve months. Shares outstanding now stand at 128.0 million.

The shares fell 2.7% on the day, which tells you the market is no longer paying for operational progress — it wants cash.

Earlier in the year the company completed a partial bond buyback, repurchasing SEK 166.9 million of its 2024/2027 notes at 104% of nominal, settled 6 May. The AGM on 12 May adopted a new long-term incentive plan and elected Lars-Åke Norling as chairman, who subsequently bought roughly SEK 3 million of shares personally.

Two dates matter from here. The Capital Markets Day in Stockholm on 16 September 2026 will lay out the regulated fund structure — this is the first serious attempt to change what kind of company VNV is, and it deserves close attention. The third-quarter report is due 27 October 2026.

8. Conclusion: What Actually Breaks the Discount

I hold VNV as a small position, and I want to be precise about what I own. This is not a quality compounder. It is a discounted claim on a concentrated set of private European platforms, run by a small, aligned team with a mixed but genuinely impressive long-run record, bought at roughly half of a value that independent third parties keep validating from above.

The path to a re-rating runs through four doors, and only one of them needs to open.

A BlaBlaCar listing. Management there has been openly IPO-curious for years, with Euronext Paris the obvious venue, and the company now has the profile a listing requires: profitable since 2022, 150 million passengers, EUR 2 billion of GMV, and a gross margin about to step up dramatically as the operated bus business winds down. A public price on VNV’s largest asset — carried at roughly half its last private round — would resolve the single biggest argument about the marks in one afternoon.

A Voi exit. Lime’s July listing did something more useful than any analyst note: it put a live, public price on a micro-mobility business that grows more slowly than Voi, loses money where Voi makes it, and carries far more balance-sheet risk. Whether Voi’s eventual outcome is a listing or a trade sale, the comparable now exists and it does not flatter Voi’s carrying value.

Buybacks. Every share retired at a 50%-plus discount is worth roughly two kronor of NAV for one krona of cash. At 2–2.5% of the company per year this is not transformative, but it is relentless, it is entirely within management’s control, and it compounds.

The fund structure. If VNV genuinely builds a managed vehicle with outside capital and fee income, the market will eventually have to decide whether it is valuing a holding company or an asset manager. Those two things trade very differently. September’s Capital Markets Day is where that story starts.

None of this is guaranteed, and I would not size this like a core holding. But this is one of the few ways a private investor in Europe can own a venture portfolio, at a discount, alongside a management team that has to buy the shares with their own money. In a market where almost everything interesting is expensive, that combination is rare enough to be worth the risk I am taking.

Risk Disclaimer

This article reflects my personal opinion and is intended for information and educational purposes only. It is not investment advice, not a recommendation to buy or sell, and not a substitute for individual advice from a qualified professional. All figures are based on publicly available information at the time of writing and may contain errors or become outdated; net asset values for unlisted holdings are estimates produced by the company and may differ materially from realisable value.

Investments in equities, and particularly in small, illiquid investment companies holding unlisted assets, carry substantial risk, including the total loss of capital. Past performance is not indicative of future results.

Conflict of interest: VNV Global is part of the portfolio of the cost-efficient Haas Invest4 Innovation Fund (invest4.net), which I manage, and I may hold the security personally or through other vehicles. I may buy or sell at any time without notice. Conflicts of interest may therefore arise. Please do your own research and make your own decisions.

Philipp Haas | investresearch.net

There is a moment in every screen where a name stops you. Mine came a few weeks ago, running my Fair-PE model across Japanese small caps. Sitting at the very top of the upside ranking — above everything else on my Japanese watchlist — was a company called Ai Robotics. Ticker 247A, Tokyo Stock Exchange Growth Market. Market capitalisation about ¥54 billion. Forward PE under ten. Revenue growth of more than 100%.

Ai Robotics does not make robots. It does not sell robots. It has, as far as I can tell, never shipped a single robot. What it actually does is sell vitamin C whitening serum, hair mist and beauty devices to Japanese women — and it does so with roughly thirty employees, a proprietary AI marketing system, and a founder who has publicly committed to a ¥1 trillion market capitalisation by March 2029 and, somewhere beyond that, space.

The robotics in the name is an ambition, not a segment. Founder and CEO Makoto Tatsukawa renamed the company in 2020 from “HowTwo! Inc.” because he does not think of himself as running a cosmetics business. He thinks of himself as running a software company that happens to have chosen beauty as the first category to point its algorithm at. Whether you find that visionary or slightly unhinged depends largely on whether the numbers back it up.

They do, mostly. Revenue has gone ¥7.1bn → ¥14.2bn → ¥29.4bn over three fiscal years. Gross margin sits above 73%. Return on equity is comfortably above 40%. Almost a billion yen of revenue per employee. This is not a promotional shell — it is one of the most capital-efficient consumer businesses I have looked at in Japan.

And yet the stock is down more than 60% from its November 2025 high of ¥2,146, trading around ¥830 as I write this on 10 August 2026, having touched ¥628 in early June. The market has decided something is badly wrong.

My view is that the market has correctly identified a real problem — a debt-funded acquisition and a cash flow statement that turned sharply negative — and then extrapolated it into a solvency narrative that the underlying economics do not support. Applying a fair PE of 26 to my three-year earnings path gets me to an annualised return potential of just under 85%. That is the widest gap I currently have in Japan.

Here is the full case, including the parts that keep me awake.

1. Product, business model, brand and moat

What they actually do

Ai Robotics is a vertically integrated D2C brand builder powered by an in-house AI system called “SELL.”

SELL was built starting in 2018, back when the company still ran performance advertising for third-party clients. It ingests purchase data, creative performance, traffic source, landing-page conversion, per-customer retention curves and CRM response, and it optimises two variables simultaneously: customer acquisition cost and lifetime value. Creative production, ad buying, demand forecasting, inventory allocation and churn intervention are largely automated.

In 2023 Tatsukawa made the decision that defines the company: he stopped selling this capability to clients and turned it inward. If the system could reliably manufacture hit products for other people, why capture an agency fee instead of the product margin?

The portfolio today:

• Yunth — the flagship skincare brand. The hero SKU is a fresh vitamin C whitening serum sold primarily on subscription. This is the profit engine.

• Brighte — mid-priced beauty devices, launched February 2024, over five million units shipped cumulatively.

• Straine — haircare, launched in FY3/2026, and the fastest new-brand ramp the company has produced.

• BJC / CHARIS&Co. — acquired April 2026. Professional-channel beauty products (salons, aesthetic clinics), category leader in eyelash serums and foundation, main brand SPICARE.

How the money is made

Three channels, and the mix is shifting fast. In FY3/2026 own-EC was 48.4% of revenue (down from 61.4%), EC marketplaces 17.6% (down from 27.3%), and retail wholesale 34.0%, up from 11.3% — wholesale revenue grew roughly 6.2x year on year. Distribution now reaches more than 10,000 drugstores nationwide. Post-BJC, management is targeting wholesale at 70–80% of revenue from the second quarter of this fiscal year.

The subscription economics on Yunth are the part I care most about: acquisition spend is recovered by roughly month three, and everything from month four onward is contribution profit. Subscribers reached 174,920 at the March year-end, up 37,601, ahead of the 172,000 plan.

That said — and this matters — from FY3/2027 the company is retiring subscriber count as its headline KPI in favour of brand-level revenue and gross profit. Management frames this as appropriate for a multi-brand, wholesale-weighted business.

Brand

Yunth’s serum won overall honours in Rakuten’s cosmetics awards. Ambassador spend has been aggressive and, in one case, genuinely global: after actresses Emi Takei (Yunth) and Nozomi Sasaki (Brighte), the company signed BTS’s V as Yunth’s brand ambassador in October 2025, with national TV advertising from November. Add Disney/Pixar Toy Storyand Alice in Wonderland collaboration packaging, and you have a brand that has moved from performance-marketing artefact to something with genuine shelf presence.

Recent launches — a penetration beauty device called Deep Boost, a whitening cushion foundation, a PDRN-and-astaxanthin serum line, a nano-mist hair milk — arrive at a cadence most Japanese cosmetics houses could not match with ten times the headcount.

Moat

I want to be careful here, because this is where the bull case is thinnest and where I have seen Japanese D2C skincare go badly wrong before. Anyone who watched Kitano Tatsujin de-rate knows that a beauty D2C model built on paid acquisition can unravel the moment the ad platforms reprice.

What Ai Robotics has that a generic brand does not:

1. A compounding data asset. SELL improves as volume grows, and the company has been running it on live spend since 2018. That accumulated feedback loop is not purchasable.

2. Structural cost advantage. Thirty people generating ¥29bn of revenue is not a marginal efficiency edge; it is a different cost structure entirely. Manufacturing is outsourced to OEMs, so fixed costs stay negligible.

3. Repeatability across brands. Yunth, then Brighte, then Straine — three consecutive successful launches in different sub-categories is evidence of a system rather than a lucky formulation.

4. Channel breadth. Own EC, marketplaces, drugstores, and now the professional salon channel via BJC. Very few Japanese challenger brands hold all four.

What it does not have: formulation IP, manufacturing control, or a defensible position against a competitor willing to buy the same shelf space. This is a process moat, not an asset moat. It compounds while it works and erodes quickly if it stops. That is precisely why I am not willing to pay a software multiple for it — and why a fair PE of 26 rather than 35 is the right anchor.

2. The market

Japan is the world’s third-largest beauty market at roughly ¥2.4 trillion, with skincare alone around ¥1.5 trillion. Ai Robotics’ entire FY3/2027 revenue guidance represents about 2.5% of that. Share gain, not category growth, is the thesis.

The category characteristics are close to ideal for a compounder:

Non-cyclical. Japanese women do not stop buying serum in a recession; they trade down within the category. Ai Robotics sits in the mid-price tier, which is where trade-down lands. In a downturn this business is defensive.

Demographically supported. Japan’s ageing population is a headwind for colour cosmetics and a tailwind for functional anti-ageing skincare — whitening, wrinkle care, pigmentation — which is exactly where Yunth plays. Quasi-drug designation lets these products make efficacy claims, and medicated variants already account for roughly 40% of domestic cosmetics shipments.

Politically benign. Cosmetics is one of the least politicised categories I cover. There are no tariffs of consequence, no subsidy dependence, no strategic-sector scrutiny. The regulatory surface is quasi-drug approval under the domestic pharmaceutical framework — a compliance cost, not an existential variable. If anything the policy wind is favourable: METI treats cosmetics as a strategic export sector, and Japanese beauty exports hit a record of roughly ¥600bn in FY2023.

An underused export option. Overseas revenue was 2.7% in FY3/2025. Management targets 20% by FY3/2029 via cross-border e-commerce into China, then Southeast Asia and Taiwan. J-Beauty carries global brand equity that this company has barely monetised. I assign near-zero value to this in my base case, which means it is free optionality.

The genuine market risk is not the market. It is that the FY3/2029 target of ¥220bn revenue would represent roughly 9% of the entire Japanese beauty market. Domestically, that is not achievable. Getting there requires overseas expansion and further M&A to work — both of which are assumptions, not facts.

3. Culture and management

Tatsukawa is the reason this is investable, and one of the reasons it is risky.

The biography is worth knowing. He started a business as a student, then founded Rocket Venture in 2013 — a women’s curation media property that reached 100 million monthly page views within about six months and was sold for ¥600 million eight months after launch. He founded Ai Robotics in April 2016, took it public on TSE Growth in September 2024 at ¥1,760, and watched it open at ¥2,514.

Three things about how he runs the company stand out to me.

He is explicit about what the company is. At the IPO press conference he stated plainly that this is not a cosmetics company but a software company developing AI systems. He has never softened that framing. It explains the headcount, the OEM structure, and — honestly — the name.

The targets are absurd and he does not walk them back. FY3/2029: ¥220bn revenue, ¥40bn operating profit, ¥1 trillion market cap, and an average employee salary of ¥100 million. He has called the trillion-yen figure a promise to shareholders. He has talked publicly about eventually reaching into space development. After the FY3/2026 miss, he reaffirmed all of it without adjustment.

I have a complicated relationship with this. Founders who set impossible targets and hold them tend to be either the ones who compound at 50% for a decade or the ones who destroy capital spectacularly trying. There is not much middle ground. What I look for is whether the operating decisions match the rhetoric — whether the ambition is a strategy or a marketing device.

So far, the decisions match. The strategy is stated simply: launch one new brand per year, grow several past ¥10bn individually, and use M&A to acquire brands already showing traction. That is exactly what has happened. Straine launched and worked. BJC was bought and is a category leader with a 28% operating margin. The pattern is consistent.

Where I take issue: the fourth-quarter investment decision. Management deliberately front-loaded spending to integrate SELL with physical-store POS data, connecting regional web marketing to in-store purchase behaviour. Tatsukawa’s own comparison is to 2018, when building SELL required several hundred million yen of deliberately inefficient data acquisition before it produced anything.

Strategically I think he is right. This is the connective tissue that makes the wholesale pivot work, and it is not replicable by a competitor without the same data history. But it was communicated poorly to a shareholder base that had been given a specific profit number, and the market punished it accordingly. That is a governance and IR failure, not a strategy failure — but it is a failure, and it is the second time this founder has chosen the long-term option and let the quarter take the damage.

Employee alignment is unusual: a headcount around thirty, no dividend, everything reinvested. There is real key-man risk here. Remove Tatsukawa and I am not confident this company functions.

4. Financials, margins and recent developments

The growth record

Revenue has more than doubled three years running. Critically — and this corrects a widespread misreading — the +106.7% in FY3/2026 was entirely organic. BJC did not close until 1 April 2026. The inorganic contribution begins this year.

The margin picture

Gross margin was 73.4% on gross profit of ¥21.5bn (+93.3%). That is the number that tells you what this business is capable of. Operating margin came in at 12.9%, down from 17.5%, and net margin at 9.0%. Return on equity was roughly 44% on year-end equity and higher on average equity — outstanding by any standard, and a direct consequence of the asset-light OEM structure.

Because there is essentially no owned production, EBITDA margin sits only slightly above the operating margin. There is no depreciation shield making the profitability look better than it is. For FY3/2027 management guides to adjusted EBITDA of ¥9.5–12.0bn on ¥56–60bn of revenue — a 17–20% margin — with the gap to operating profit being goodwill and intangible amortisation from BJC, provisionally assumed at about ¥1.5bn per year.

The miss

Guidance for FY3/2026 was ¥28.0bn revenue, ¥4.80bn operating profit, ¥3.33bn net income. Delivered: revenue beat at ¥29.36bn; operating profit missed by 21% at ¥3.80bn; net income missed by 20%.

The first warning came in November 2025, when first-half results showed operating profit of just ¥701 million — 14.6% progress against the full-year plan. The stock went limit-down the following session, falling ¥400 to ¥1,489. It has not recovered since.

The cash flow problem

This is the part that actually matters and the part most write-ups skip. Operating cash flow for FY3/2026 was negative ¥5.88 billion.

The composition, per management: approximately ¥7bn of uncollected receivables at year-end, plus a large inventory build to support a doubling of volume. Both are mechanical consequences of the shift from own-EC (cash collected at the point of sale) to wholesale (60–90 day retailer payment terms). Total assets were ¥18.4bn, net assets ¥6.05bn, equity ratio 32.8%.

Then, on 27 March 2026, the company announced the acquisition of BJC for ¥25.55 billion, funded entirely by borrowing from Mizuho Bank, completing on 1 April.

Look at those numbers side by side. A company with ¥6bn of equity and negative ¥5.9bn of operating cash flow took on ¥25.6bn of bank debt. That is roughly four times equity. Whatever else you conclude, you cannot call this conservative.

The deal itself is well-priced: BJC generated ¥10.9bn of revenue and ¥3.07bn of operating profit in its October 2025 fiscal year — a 28% operating margin, acquired at roughly 8.3x EV/EBIT. Buying a profitable category leader at that multiple is good capital allocation. Financing it entirely with debt off a stressed cash flow base is the aggressive part.

FY3/2027 by brand

Management has disclosed the plan brand by brand, which I appreciate: Yunth ¥20.0bn revenue at 70.0% gross margin; Brighte ¥15.0bn at 64.7%; Straine ¥6.0bn at 61.7%; BJC ¥15.0bn at 57.3%.

Two observations. First, the descending gross margin ladder is the wholesale transition made visible — as the mix shifts to retail and professional channels, blended gross margin compresses. Second, Yunth remains roughly a third of the plan. Diversification is improving but concentration is real.

5. Valuation

My framework asks one question: what multiple would this business deserve if it were fairly valued for its growth, quality and risk — and what return do I earn if it gets there?

Starting point. At ¥830 with 65.0 million shares outstanding, market capitalisation is roughly ¥54bn. Against FY3/2027 adjusted net income guidance of ¥5.9–7.4bn, that is a PE of 7.3x to 9.1x, roughly 8.1x at the midpoint. On reported net income after goodwill amortisation, closer to 10–11x. Price-to-book is 8.9x, which tells you the earnings power sits in the process, not the balance sheet.

Either way: single-digit to low-double-digit earnings multiple on a business guiding to roughly double its revenue.

Fair PE: 26. How I get there. A business growing revenue above 90% with 73% gross margins and 40%-plus returns on equity would ordinarily command well above 30x. I take that down materially for four reasons: leverage from the BJC deal, negative operating cash flow, a process-based rather than asset-based moat, and a founder whose ambition materially exceeds his current execution consistency. Against that, I add back for genuine capital efficiency, three consecutive successful brand launches, and unmonetised international optionality. Twenty-six is where I settle — the same anchor I use for high-quality compounders whose growth is real but whose risk is above average.

The earnings path. FY3/2027 adjusted net income at the ¥6.65bn midpoint gives roughly ¥102 per share. From there I deliberately depart from management. Rather than the ¥220bn revenue and ¥40bn operating profit implied by the FY3/2029 plan, I assume growth decelerates hard to roughly 40% and then 35% — a fraction of the company’s own ambition. That produces around ¥143 for FY3/2028 and approximately ¥193 for FY3/2029.

The result. ¥193 × 26 = roughly ¥5,000 per share on a three-year view, against ¥830 today. That is a compound annual return of just under 85%.

I want to be clear about what is doing the work. This is not primarily a growth story — it is a re-rating story with growth attached. Roughly a third of the return comes from earnings compounding; the remainder comes from the multiple travelling from 8x to 26x. Multiple expansion of that magnitude requires the market to change its mind, and markets change their minds about leveraged small caps slowly, and only after seeing cash.

Which brings me to the sensitivity that matters. If the fair PE is 18 rather than 26, the three-year target is roughly ¥3,470 and the annualised return is still above 60%. If growth stalls to 20% annually and the fair multiple is 15, the target is roughly ¥2,200 and the return is around 39%. The margin of safety here is not in my precision; it is in how much has to go wrong before this stops working. That is the definition of an asymmetric setup.

6. Risks — and why the opportunity exists

A stock does not fall 61% from its high without reason. Here is what the market is pricing, and I think most of it is legitimate.

Leverage against fragile cash flow. ¥25.55bn of bank debt against ¥6.05bn of equity and negative ¥5.88bn of operating cash flow. If the wholesale transition takes longer than planned to convert receivables into cash, this becomes an interest-coverage conversation rather than a growth conversation. This is the single largest risk and it deserves the discount it is receiving.

Working capital structurally deteriorating. Own-EC collects instantly; wholesale collects in months. Moving to 70–80% wholesale means the cash conversion cycle lengthens permanently. Growth becomes cash-hungry rather than self-funding.

A credibility deficit. Missing guidance by 21% after a limit-down on the half-year print is the kind of thing a Growth Market small cap gets exactly one chance to do. Management has now used that chance.

Gross margin compression by design. The brand plan shows the ladder explicitly: 70% at Yunth down to 57% at BJC. Wholesale scale trades margin for volume. Whether operating leverage offsets it depends on whether the SELL–POS integration delivers the efficiency management believes it will.

Concentration and key-man exposure. Yunth is roughly a third of planned revenue. The company runs on approximately thirty people. Both the strategy and the execution live in one founder’s head.

Category fragility. Beauty D2C built on paid acquisition can unwind fast when platform economics shift. The Japanese market has produced clear precedents. This is precisely why my fair PE is 26 and not higher.

Integration and goodwill. BJC brings roughly ¥1.5bn of annual amortisation and, more importantly, a professional salon channel with an entirely different sales culture from performance-driven e-commerce.

Microstructure. TSE Growth listing, no sell-side coverage, thin liquidity, and margin-buy-only designation. The stock will overshoot in both directions.

Sentiment risk that runs the other way. On the Japanese retail forums and across X, sentiment on 247A is overwhelmingly positive — the Yahoo Finance board runs around 89% “strong buy.” That is not comfort; that is a warning. A stock down 61% with near-universal retail conviction has a shareholder base that has not yet capitulated, which means there may be more supply to come. My preference is to build a position gradually rather than assume ¥628 was the floor.

So why does the opportunity exist? Because three separate things landed within six months: a half-year profit miss that produced a limit-down, a full-year miss that confirmed it, and a large debt-funded acquisition announced days before the year-end. Any one would compress a multiple. Together, on an uncovered Growth Market stock with no institutional support, they produced a de-rating from 40x-plus to 8x.

The market is pricing a balance sheet accident. What I see is a deliberate, well-communicated decision to trade near-term margin for structural distribution advantage — a decision the same founder made in 2018, at smaller scale, and which produced the system the entire company now runs on.

7. Latest news and what I am watching

The next print is 14 August 2026 — first-quarter FY3/2027, the first quarter to consolidate BJC. This is the most important disclosure this company has ever made, and it lands four days from now.

What I will read first, in order:

1. Operating cash flow. Not revenue, not profit. Whether the ¥7bn receivable position converted, and where inventory sits. Everything else is secondary.

2. Wholesale mix. Management targets 70–80% from Q2. The Q1 trajectory tells me whether the pivot is on schedule.

3. Brand-level gross profit. The first quarter under the new KPI framework, and the first test of whether it is genuine transparency or a softer bar.

4. Whether the ¥56–60bn range narrows. A range that wide at the start of a year is an admission of limited visibility. Any narrowing is meaningful information.

Since the May results, the operational news flow has been consistently constructive: Straine’s nano-mist hair milk launched in late June, Yunth added a PDRN-and-astaxanthin serum line and a cushion foundation, the Deep Boost device extended Yunth into hardware, and Disney/Pixar and Alice in Wonderland collaborations kept shelf visibility high. BTS’s V continues as Yunth ambassador with national TV support — the international awareness asset that the 2.7%-to-20% overseas ambition depends on.

Tatsukawa has given interviews since the results and has reaffirmed the FY3/2029 targets in full. He has been direct that the profit shortfall was the SELL–POS investment and nothing else, and that revenue exceeded the doubling target while profit grew only 1.5x for that specific reason.

The stock, meanwhile, has done nothing. It bottomed at ¥628 on 3 June, traded to ¥765 in mid-July on elevated volume, and has drifted in the ¥790–850 range since. That flatness after a 61% decline is itself informative: forced selling looks finished, but nobody is willing to underwrite the balance sheet until they see a cash flow statement. Which is, of course, exactly what arrives on Thursday.

8. Conclusion

Ai Robotics is a company whose name is wrong, whose founder makes promises no rational analyst would underwrite, and whose balance sheet took on four times its equity in debt at precisely the moment its cash flow turned negative.

It is also a business that has tripled revenue in two years organically, sustains 73% gross margins and 40%-plus returns on equity, generates almost a billion yen of revenue per employee, has launched three successful consumer brands in three consecutive years, and is guiding to roughly double again — and it trades at eight times forward adjusted earnings.

Both descriptions are accurate. The investment question is which one the market is over-weighting.

My answer is the first. The leverage is real and I do not dismiss it, but it was deployed to buy a category leader with a 28% operating margin at 8.3x EV/EBIT — a good asset at a good price, financed badly. The negative cash flow is real, but it is the mechanical signature of a channel transition, not of a business losing money. The profit miss is real, but it stemmed from an identifiable, strategically coherent investment that management explained in advance and has not disowned.

Against a fair PE of 26 and deliberately conservative growth assumptions well below the company’s own plan, I arrive at roughly ¥5,000 per share on a three-year view — just under 85% annualised.

The investment cases, as I see them:

The base case is the wholesale transition completing on schedule, receivables converting, BJC integrating, and the multiple travelling from 8x toward the high teens or low twenties as visibility returns. That alone is a multi-bagger.

The upside case adds the overseas expansion — from 2.7% to a 20% revenue share — where J-Beauty brand equity and a globally recognised ambassador meet a marketing system built to scale acquisition efficiently in new markets. I carry this at zero.

The founder case, which I hold loosely and size accordingly, is that Tatsukawa builds what he says he will build: a portfolio of brands each exceeding ¥10bn, a marketing engine that repeats hits reliably, and a multiple that eventually reflects software rather than cosmetics. I do not underwrite it. I simply note that a founder who sold his first company eight months after launch and compounded revenue at over 100% for three consecutive years has earned the benefit of a small, patient allocation.

The bear case — and I want it stated as plainly as the rest — is that receivables do not convert, the debt burden forces an equity raise at a depressed price, the wholesale pivot compresses margin faster than volume compensates, and the shareholder base finally capitulates. In that scenario this is not a 50% drawdown, it is a permanent impairment.

That is why this is a position, not a conviction weighting. It is a leveraged Japanese nano-cap on the Growth Market with no analyst coverage, thin liquidity and a founder aiming at the moon. Sized correctly, that is an asymmetric bet. Sized incorrectly, it is a way to lose money in a business that was fundamentally sound the entire time.

For me, the arithmetic is straightforward. I am paying eight times earnings for a company doubling its revenue with 73% gross margins. The market is charging me almost nothing for growth and charging me a great deal for leverage. I have seen that trade misprice in both directions — and I think this time it is mispriced in mine.

I will be reading the cash flow statement on 14 August with more attention than I have given any Japanese small-cap disclosure this year.

Risk disclaimer and disclosure of conflicts of interest

This article represents my personal opinion and analysis as at 10 August 2026. It does not constitute investment advice, a recommendation, a solicitation, or an offer to buy or sell any security. It is not a financial analysis within the meaning of applicable securities regulation and does not meet the legal requirements for ensuring the impartiality of investment research.

Ai Robotics Inc. (TSE: 247A) is a Japanese nano-cap listed on the TSE Growth Market. It carries above-average risk, including but not limited to: substantial financial leverage relative to equity, negative operating cash flow, low trading liquidity, absence of sell-side analyst coverage, high revenue concentration in a single brand, key-man dependency, currency risk for non-yen investors, and elevated share price volatility. Historical figures and management guidance are no indication of future results. A total loss of capital is possible. Every investor must reach their own conclusions and, where appropriate, seek independent professional advice. All financial data is drawn from company disclosures and public market sources and is presented to the best of my knowledge; no warranty is given as to accuracy or completeness.

Conflict of interest: The stock discussed is a position held in the cost-efficient Haas Invest4 Innovation Fund(invest4.net). I therefore have a financial interest in the price development of this security. Positions may be increased, reduced or closed at any time without notice or subsequent publication.

Philipp Haas — investresearch.net

By Philipp Haas — investresearch.net

There are stocks you buy for the story, and there are stocks you buy for the arithmetic. Noah Holdings (NYSE: NOAH / HKEX: 6686) is one of the rare cases where the arithmetic is the story — and where the story the market is telling itself has drifted so far from the numbers that I find it hard to look away.

Let me put my cards on the table right at the start: I own a small, indirect position, so a conflict of interest exists and everything below is my subjective opinion, not investment advice. With that out of the way, here is the case in one paragraph.

Noah is the pioneer and one of the clear leaders in wealth and asset management for wealthy Chinese families — a structurally growing client base that is slowly migrating out of property and into professionally managed financial assets, increasingly across borders. The company earns money in two ways: cyclical distribution fees on the products it sells, and higher-quality recurring management fees on the roughly RMB 140 billion it manages in-house. It is founder-led, carries zero interest-bearing debt, and sits on cash and short-term investments of around RMB 5.1 billion — which, at today’s price near $10.40 and a market cap of roughly $680 million, is worth about as much as the entire company. You are, in effect, being handed the operating business for close to nothing. The stock trades at roughly 8x earnings and pays out 100% of its non-GAAP profit to shareholders. My subjective fair multiple is 16x, and on that basis I see an annualized return potential of around 55% over my holding horizon. The catch — and there is always a catch — is that this is a China ADR, with everything that entails.

That is the whole thesis. Now let me walk through why I think it holds up.

1. Product, business model, brand and moat

What Noah actually does. Think of Noah as something close to a private bank for China’s high-net-worth and ultra-high-net-worth families — except it started life independent rather than as the retail arm of a giant lender. On the wealth side, its relationship managers sit down with wealthy clients and build portfolios: private equity, private secondary and public-securities funds, insurance and comprehensive family-office services. Much of what they distribute is third-party product, which is exactly what you would expect from an advisory platform.

How it earns money — and why the mix matters. This is the part I want investors to internalize, because it is the difference between a mediocre business and a good one. Distribution revenue is transactional: you get paid when the client transacts, and that flow is cyclical, sentiment-driven and lumpy. The higher-quality half of Noah is its own asset-management arm — Gopher Asset Management onshore and Olive Asset Management offshore — where Noah manages money itself and collects a management fee every single year the assets stay on the platform. That is recurring, capital-light and far more predictable. It is the same reason I generally prefer asset gatherers with sticky management fees over pure brokers living off transaction commissions. The more of Noah’s revenue that migrates toward recurring Gopher/Olive fees, the more the whole enterprise deserves to be re-rated.

Brand. Noah has been doing this since 2005 and is genuinely a category pioneer in China. Several would-be competitors from the last decade have quietly disappeared; Noah is still standing, with roughly 468,000 registered clients and an established presence not just in mainland China but in Hong Kong, Singapore, Japan and key US hubs including New York, Los Angeles and Silicon Valley. That international footprint is not decoration — it is the spine of the current strategy, which management describes as becoming an AI-driven global platform serving Chinese families wherever they live.

Moat. I want to be honest here: this is a good business, not an unassailable one. The moat is a combination of soft factors — brand, a two-decade track record, a trained relationship-manager network, regulatory relationships and the trust of families who do not move their life savings casually. Those are real, and they compound. But they are replicable over time, competition is intensifying, and the total addressable segment, while growing, is not yet enormous. I would file Noah under “durable franchise with switching costs,” not “fortress.” What genuinely elevates the model is the pairing of that franchise with a scalable, capital-light fee engine and a balance sheet most competitors can only dream of.

2. The market: big, growing, cyclical — and politically loaded

The tailwind behind Noah is one of the largest wealth-migration stories on the planet. China has minted an extraordinary number of wealthy households over two decades, and for most of that period their default asset was property. That era is ending. Real estate has stopped being the automatic store of value it once was, the population is ageing, and — crucially — China as a system needs domestic savings to flow into its capital markets rather than sit idle in bricks. So even as the political direction of travel is often tighter and more inward-looking, the financial logic pushes toward deeper, more professional, more diversified capital allocation. That is the exact behavioural shift Noah is built to monetize.

There is a second, offshore leg to this. Wealthy Chinese families increasingly want a portion of their capital allocated outside the mainland — through Hong Kong, Singapore and beyond — for diversification and estate planning. Noah’s overseas build-out (ARK Wealth, Olive Asset Management, Glory Family Heritage) is precisely the vehicle for that demand.

Now the honest caveats. This market is cyclical — when Chinese risk appetite freezes, distribution volumes freeze with it, and we saw exactly that in the brutal 2021–2022 stretch. And it is not free from political intervention — this is China, and regulatory and geopolitical risk is a permanent feature, not a bug. What I would say is that Noah operates in the one corner of the Chinese economy that Beijing has a genuine incentive to support rather than suppress: a functioning capital market that channels household savings into productive assets. That is meaningfully different from sitting in the crosshairs of a crackdown.

3. Culture and management: founder-led and playing the long game

Noah was co-founded in 2005 by Jingbo Wang, who remains chairwoman today. I place a lot of weight on founder-led companies, because founders think in decades and treat the balance sheet as if it were their own money — because it largely is. Noah’s conduct bears this out. Through the ugliest years for Chinese equities, management did not chase growth for its own sake. Instead it did the unglamorous work: streamlining the domestic coverage network, cutting compensation costs, and deliberately reshaping the revenue mix toward higher-quality recurring fees and overseas expansion.

The clearest tell of shareholder-aligned management, though, is capital return. Noah is not hoarding its cash pile — it is handing it back. The board has committed to distributing 100% of non-GAAP net income for both 2024 and 2025 (a regular dividend plus a special dividend), on top of a running share-buyback program under which the company has already repurchased well over 13% of its shares. That is not the behaviour of empire-builders. That is management treating minority holders as partners. In a China ADR — a universe where governance scepticism is entirely warranted — that track record of actually paying cash out to shareholders is worth a great deal.

4. Financials, margins and recent developments

Here is where the “flat revenue, rising quality” picture comes into focus. For full-year 2025, net revenues were roughly RMB 2.6 billion — essentially flat year-over-year. If you stopped reading there you would shrug. But look one line down: operating profit rose about 22.5% to RMB 777 million, operating margin climbed to 29.8%, and non-GAAP net income grew 11.2% to RMB 612 million. The engine here is not the top line — it is discipline. Noah is squeezing far more profit out of the same revenue base.

That trend accelerated into Q1 2026. Net revenues of RMB 625.8 million were up only 1.8% year-over-year, but income from operations jumped 27.1% to RMB 236.4 million, pushing the operating margin to an eye-catching 37.8% — up from 30.3% a year earlier and among the highest quarterly levels the company has printed. Domestic revenue grew a healthy 25.9%, while overseas revenue fell 23.3% on weaker performance-based income — so the geographic mix swung back toward the mainland this quarter. Underneath the reported figures, active clients rose 21.8% and the total value of products transacted grew a striking 44.8% year-over-year, both signs that client engagement is recovering even before revenue fully reflects it.

On margins specifically: this is an asset-light business, so operating and EBITDA margins sit close together in the low-to-mid 30s%, and net income margins run in the low-20s% range. Return on equity is more modest than those margins suggest — and here is the subtle point — precisely because of that enormous cash balance. A pile of cash earning near-nothing drags reported ROE down; strip it out, and the return on the actual operating business is substantially higher. That is a “problem” I am very happy to own.

The one genuine blemish in the recent numbers is at the bottom line: GAAP net income in Q1 2026 fell 16.3% to RMB 124.7 million, and headline non-GAAP net income dropped as well — not because the core business deteriorated, but because of volatility in “equity in affiliates” and investment losses. Adjusted for that non-operational noise, underlying non-GAAP net income would have grown around 28%. It is a reminder that reported earnings here can be noisy quarter to quarter; you have to look at the operating line to see the real trajectory.

5. Valuation: cheap on earnings, absurd on cash

This is the section that made me pay attention in the first place.

Start with the headline multiple: at roughly $10.40 per ADS, Noah trades at about 8x earnings — and closer to 6x on a forward basis by some estimates. For a profitable, cash-generative, dividend-paying franchise leader, that is already inexpensive.

Then account for the balance sheet. Cash and short-term investments of roughly RMB 5.1 billion (about US$710 million) sit against a market cap of only about US$680 million, with zero interest-bearing debt. Read that again: the cash and liquid investments are worth approximately the entire market value of the company. On an ex-cash basis, the market is valuing Noah’s actual operating business — a two-decade franchise throwing off RMB 600 million-plus of annual profit — at something approaching zero. That is not a value stock; that is a mispricing waiting for a catalyst.

My framework here is a fair P/E, and for Noah I set that at 16x. That is deliberately conservative — for a Western asset manager with these margins, this balance sheet and this payout, you would routinely see multiples well north of that, and I am not even crediting the net cash in the multiple. A re-rating from ~8x toward 16x is roughly a doubling of the share price from the earnings multiple alone. Layer on a dividend yield that runs in the high-single digits (and, with the special-dividend top-up representing a full 100% payout, higher still in a full-payout year), plus even modest recovery in earnings, and the return math becomes very powerful. Running that through my model over my holding horizon, I arrive at an implied annualized return of roughly 55% in my base upside scenario.

I want to be clear-eyed: that is a subjective estimate, not a promise, and it explicitly assumes the market eventually stops applying a near-maximal China discount. If that discount never narrows, the cash and dividends still pay you to wait — but the re-rating leg does not fire. You are being compensated for patience either way; the upside is the option on sentiment turning.

6. Risks: why the opportunity exists at all

A stock does not trade at its cash balance for no reason. The discount is the thesis, so I owe you an honest accounting of what it is pricing.

The dominant risk is simply “China ADR.” That single label bundles several fears: geopolitical friction between Beijing and Washington, the tail risk around US-listing status for Chinese companies, opaque governance concerns that hang over the entire cohort, and the ever-present possibility of regulatory intervention. Investors have applied a blanket discount to Chinese ADRs, and Noah has been marked down alongside its peers regardless of its own conduct. Noah’s dual listing — NYSE for liquidity, Hong Kong (6686) as an alternative venue — partially mitigates the pure delisting tail, but it does not make the China risk disappear.

Layered on top is cyclicality. Noah’s distribution revenue rises and falls with Chinese risk appetite and asset prices. In a downturn you get hit twice — lower transaction volumes and lower asset values that shrink the fee base — which is exactly why the stock overshoots to the downside in bad years. That double-leverage cuts the other way in a recovery, which is part of the appeal, but it means this is not a low-volatility holding.

And then the softer risks I flagged earlier: a moat built on brand and trust rather than structural lock-in, rising competition, and a still-developing addressable market. None of these are fatal. But together they explain why a business this cheap has stayed this cheap — and why the opportunity is available to a patient buyer willing to underwrite China exposure that most of the market currently refuses to touch.

7. Latest news and earnings

The most recent reported quarter is Q1 2026 (released late May), covered above: flat revenue, sharply higher operating margin, noisy bottom line, and a domestic business visibly regaining momentum while overseas cooled off. Management’s guidance is to keep full-year operating margin comfortably above 30% while investing in globalization and AI, with three stated priorities — expanding overseas clients, deepening global asset allocation, and optimizing Olive’s asset-management revenue.

On capital return, the news flow has been steady and shareholder-friendly. The board approved a final dividend and a matching special dividend of RMB 306 million each — together roughly RMB 1.87 per share — for the 2025 year, paid out to ADS holders in early August 2026, representing that full 100% non-GAAP payout. The buyback has continued in parallel, with a fresh board mandate authorizing repurchases of up to 10% of shares. Sell-side sentiment has drifted toward the constructive end, with a modest “Buy” consensus and price targets clustered a little above the current quote — though, as is typical for this name, even good operational news has tended to produce muted or negative near-term share reactions. That gap between operational progress and market response is precisely the inefficiency I am trying to exploit.

8. Conclusion: why I find this interesting

Strip everything back and Noah offers a stack of investment cases that rarely appear together in one ticker:

The net-cash case. You are buying the operating business at close to zero, with a cash-and-investments buffer roughly equal to the market cap and no debt. That is a genuine margin of safety, not a slogan.

The scalability case. The Gopher/Olive fee engine is capital-light and recurring. As the revenue mix tilts toward management fees, margins and multiple should both grind higher — and the last two years of expanding operating margin on flat revenue show the operating leverage is already working.

The recovery case. This is a cyclical business trading near the bottom of its cycle. When Chinese and global-Chinese risk appetite returns, Noah gets a double kick: more transactions and a larger, appreciating fee base.

The structural case. Wealthy Chinese families are migrating out of property and, increasingly, seeking professional, cross-border, offshore-capable allocation through hubs like Singapore and Hong Kong. Noah is one of the few franchises positioned to serve exactly that demand, onshore and off.

The shareholder-return case. A founder-led board paying out 100% of profit and buying back stock is telling you how it thinks about minority holders.

Put a conservative 16x fair multiple on that, decline to credit the net cash, get paid a high-single-digit yield to wait, and you arrive at a return profile I find genuinely compelling — on my numbers, on the order of 55% annualized in the upside scenario. The price of admission is a willingness to underwrite China risk that most of the market currently won’t. That is the whole game: the same discount that creates the risk creates the opportunity. For an investor who can stomach the volatility and size the position accordingly, I think that trade is worth making.

Risk disclaimer

This article reflects my personal, subjective opinion and is not investment advice, a recommendation, or a solicitationto buy or sell any security. All figures are drawn from publicly available data and are subject to change; forward-looking estimates, including any “fair P/E” and expected-return figures, are subjective projections that may prove wrong. Investing in equities — and in Chinese ADRs in particular — carries the risk of substantial or total loss. Do your own research and consult a licensed advisor before making any investment decision.

Conflict of interest: Noah Holdings is a position in the Haas Invest4 Innovation Fund (invest4.net), the cost-efficient investment fund I manage, and I may hold the security personally or indirectly. A material conflict of interest therefore exists, and my views should be read with that in mind.

by Philipp Haas – investresearch.net

Let me start with a scene most of you will recognize.

You go to the doctor as a private patient. A few weeks later a bill lands in your mailbox — but it isn’t from the doctor. It’s from some company you’ve never heard of, with a name you can’t pronounce, and it wants its money. What happened in between? The physician doesn’t want to spend his afternoons chasing invoices, checking insurance codes, and arguing with payers. So he hands the entire billing machinery to a specialist and gets on with treating patients.

That specialist is exactly the kind of business CareCloud (Nasdaq: CCLD) runs — and here I want to correct a small but important misunderstanding, because I had it in my own head at first too. CareCloud is not buying the debt and collecting it like a factoring house. That’s a different, uglier business. What CareCloud does is run the revenue cycle for the provider: it codes the visit, files the claim with the insurer, chases the denials, reconciles the payments, and bills the patient for the remainder. It gets paid a percentage of what it successfully collects. So its interests are aligned with the doctor — the more CareCloud collects, the more CareCloud earns. It never takes the credit risk of owning the receivable. That distinction matters, because it turns what sounds like a grubby collections agency into a sticky, recurring, high-retention software-and-services business.

And that is the whole investment case in one sentence: an unloved US micro-cap that has spent a decade rolling up sleepy little billing shops, bolting its own cloud software on top, and — as of the last eighteen months — quietly wiring artificial intelligence through the entire stack. The market still prices it like a tired healthcare-services roll-up. I think it’s turning into something closer to a vertical SaaS platform. That gap is the opportunity.

At the current price of around $2.54 (Nasdaq close, 5 August 2026 — and note, they report Q2 this morning, more on that later), on my numbers I get to a fair P/E of 22 and a potential return of roughly 34% per year. For a US-listed name, that is a lot of upside, which is why I’ve built a starter position in the Haas Invest4 Innovation Fund.

Let me walk you through how I get there.

1. Product, Business Model, Brand and Moat

What does the company actually do? CareCloud sits at the plumbing layer of US healthcare. Its bread and butter is technology-enabled revenue cycle management (RCM) — the billing engine I described above — sold to everyone from a two-doctor family practice up to hospitals and health systems. On top of that core, it sells a suite of proprietary, cloud-based products: electronic health records, practice-management software, patient-experience tools, analytics, and increasingly a family of AI agents branded stratusAI. Since the 2025 acquisition of Medsphere it also plays in the inpatient software market, including a Black Book–ranked emergency-department information system, which meaningfully widened the pond it can fish in.

How do they make money? Two ways, and they reinforce each other. First, the RCM engine earns a slice of collections — recurring, usage-linked, and remarkably sticky, because once a practice’s cash flow runs through your pipes, ripping you out is a nightmare nobody volunteers for. Second, the software modules are cross-sold into that same installed base of 45,000-plus providers. The genius of the model is the cross-sell: land a small practice through a cheap billing contract, then upsell EHR, analytics, and now AI. That’s textbook vertical-SaaS land-and-expand, dressed up as a services company.

Brand and scale. Nobody outside US healthcare IT has heard of CareCloud, and that’s fine — this is a B2B infrastructure brand, not a consumer one. What matters is the trajectory: revenue has grown from roughly $23 million in 2015 to an expected ~$130 million in 2026, through more than twenty acquisitions, while the reputation among mid-market providers has quietly hardened into “the affordable, tech-forward alternative to the incumbents.” That’s a respectable brand within its niche, and it’s growing.

Moat. Here’s where I temper my enthusiasm honestly. The moat is real but narrow. The stickiness comes from switching costs — mission-critical, cash-flow-linked workflows that providers dread migrating — plus the accumulating data advantage as AI learns across 45,000 providers’ worth of claims. That’s genuine. But CareCloud is not the only roll-up in this space, the underlying RCM service is somewhat commoditized, and a determined larger competitor with deeper pockets could compress margins. So: a moat built on switching costs and integration friction, deepening as AI compounds, but not a fortress. Call it a two-metre wall that’s slowly getting taller.

2. The Market

The US healthcare revenue-cycle and health-IT market is exactly the kind of pond I like: enormous, structurally growing, deeply fragmented, and stubbornly non-cyclical.

• Big and growing. Americans keep getting sick, getting older, and generating ever-more-complex insurance paperwork. Billing complexity is a feature of the US system, not a bug that’s going away. Every layer of regulation, coding change, and payer friction is demand for someone like CareCloud.

• Fragmented — which is the point. There are thousands of tiny regional billing shops still running on ageing software and human labour. That fragmentation is the raw material for CareCloud’s whole strategy: buy them cheap, modernise them, plug in AI. The acquisition runway here is measured in years, not quarters.

• Non-cyclical. Healthcare demand doesn’t collapse in a recession. People delay buying a car; they don’t delay chemotherapy. Revenue tied to claims volume is about as recession-resistant as revenue gets.

• Political risk — the honest caveat. This is the one soft spot. Reimbursement policy, Medicare and Medicaid rules, and any structural reform of US healthcare financing all wash through this sector. A big enough change to how providers get paid would ripple into how much CareCloud collects on their behalf. It’s not a reason to stay away, but it’s a live variable I keep on the dashboard.

Net-net: a large, growing, defensive market with a decade-long consolidation opportunity, carrying a manageable dose of regulatory risk.

3. Culture and Management

This is a founder-led business, and long-time readers know how much weight I put on that.

The company was built by Mahmud Haq, who remains Executive Chairman — the classic long-horizon founder who has steered CareCloud through a genuinely turbulent decade rather than cashing out at the first bump. Day-to-day is now run by CEO Stephen Snyder, with A. Hadi Chaudhry — himself a former CEO — shifted into Chief Strategy Officer, which tells you the bench has depth and continuity rather than revolving-door churn. Norman Roth holds the CFO seat on an interim basis, the one org-chart line I’d like to see made permanent.

What I like about the culture: the capital discipline is real and repeatable. The acquisitions follow a consistent playbook — asset purchases, priced at roughly one times revenue or less, structured to be non-dilutive to common shareholders, and funded from operating cash flow. That is exactly how a serial acquirer should behave. They then went further and cleaned up the balance sheet: converting the bulk of the Series A preferred into common (killing over $7 million of annual dividend obligations) and fully redeeming 100% of the Series B preferred in May 2026 via a $50 million bank facility, stripping out another $3.2 million a year. That’s a management team deliberately lowering its cost of capital and simplifying the story for institutional investors. Founder-led, disciplined, thinking in decades — this is my kind of team.

I’ll save the one place where I think management stubbed its toe for the risk section, because it’s the single best explanation for why the stock is cheap.

4. Financials, Margins and Recent Developments

The numbers are where the transformation stops being a story and starts being a fact.

Full-year 2025 (the pivot year):

• Revenue: $120.5 million, up from $110.8 million in 2024 (~9% growth)

• GAAP net income: $10.8 million, up more than 37% year over year

• GAAP EPS: $0.10 — the first full-year positive GAAP EPS since the 2014 IPO. Read that twice. After a decade in the wilderness, this business crossed into sustainable profitability.

• Adjusted EBITDA: $27.5 million, a 23% margin, up around 14–15%

• Operating cash flow: $28.6 million (+38%); non-GAAP free cash flow: $20.5 million (+55%)

Margins, in the language I care about: a ~23% adjusted EBITDA margin, a net margin in the high-single digits (~7.9%), and — the number that really makes me sit up — a return on equity above 24%. For a company still viewed as a beaten-down services roll-up, a mid-twenties ROE is not what you’d expect. That’s the fingerprint of a capital-light, cross-selling model starting to hit its stride.

Q1 2026 kept the momentum: revenue $31.3 million, up 13% year over year, GAAP net income positive, EPS $0.05. The 13% top-line print is running ahead of the full-year guide, though a chunk is acquisition-fed.

2026 guidance: revenue of $128–132 million, adjusted EBITDA of $29–31 million, and — the headline — GAAP EPS of $0.20–$0.23, which more than doubles the 2025 result. That guided revenue range implies roughly 6–10% growth, which lines up almost exactly with the “7–10%” I had penciled in. The EPS doubling is the part the market seems to be sleeping on.

One under-appreciated detail from the calls: CareCloud grew revenue in 2025 while reducing headcount, explicitly crediting AI-driven automation. That is operating leverage arriving in real time, and it’s why I believe the margin trajectory has further to run.

5. Valuation: The Fair-P/E View

Here’s my framework, the same fair-P/E (faires KGV) approach I apply to everything.

The trailing multiple looks unremarkable at first glance — around 22x on the trailing $0.10 of GAAP EPS. So on the rear-view mirror, the stock looks roughly fairly valued, not cheap. But the rear-view mirror is the wrong instrument here, because earnings are inflecting hard.

I anchor on a fair P/E of 22 for this business. Why 22 and not more? Because I respect the checkered history — the revenue declines earlier this decade, the preferred-dividend arrears, the still-narrow moat all argue for discipline rather than a rich SaaS multiple. And why not less? Because a 45,000-provider installed base, 20%+ ROE, high-20s% EBITDA margins, real free cash flow, and an AI-plus-M&A growth engine deserve better than the low-teens multiple the market slaps on tired healthcare services. Twenty-two is my honest midpoint — the number that neither forgives the past nor ignores the transformation.

Now apply it forward. On 2026 guidance the stock trades at roughly 11–13x GAAP earnings, and on an adjusted-earnings basis the forward multiple slips toward single digits — which is the “single-digit P/E next year” thesis, correctly stated. If EPS doubles into the low-$0.20s this year and keeps compounding into the high-$0.20s over the following two years on the back of cross-sell, AI automation, and disciplined bolt-ons, then a 22x fair multiple on that forward earnings power points to a share price in the neighbourhood of $6 on a roughly three-year horizon.

Run that through the model and it spits out an annualised return of about 34%. The return here isn’t coming from multiple expansion — I’m assuming the multiple stays at fair value. It’s coming almost entirely from earnings growth, which is exactly the kind of return I trust most, because it doesn’t require the market to fall in love, only for the company to keep executing.

6. Risks — Why Is It This Cheap?

If it’s this good, why is it trading at $2.54 and not $6? Fair question. A stock is rarely cheap by accident, and CareCloud gives you several honest reasons.

The checkered past. This is the big one. Revenue actually declined across the earlier part of this decade — from roughly $140 million to $111 million by 2024 — as large health-system clients migrated off a legacy platform CareCloud had acquired. The company only just clawed back to full-year GAAP profitability in 2025. Investors who got burned on the way down are, understandably, slow to re-rate on the way back up. Memory is the most powerful discount mechanism in markets.

The aircraft. Here’s management’s stubbed toe. In among the sensible, one-times-revenue acquisitions, CareCloud set up a subsidiary to purchase an aircraft — justified as a way to visit current and prospective clients. Roth Capital’s analyst put it bluntly: it “doesn’t help,” and it raises fair questions about capital-allocation discipline and whether the acquisition machine is obscuring softer underlying organic growth. I don’t love it either. It’s a small dollar amount, but it’s a signal, and signals matter for a management team asking the market to trust its judgment. I’m watching it.

Serial-acquirer risk. Roll-ups live and die on integration. Buy enough companies and eventually you overpay, or you fail to integrate one cleanly, or the reported growth flatters over organic weakness. The discipline has been excellent so far; the risk is that it doesn’t stay that way.

Micro-cap reality. At a market cap around $100 million, this is small, thinly traded, and volatile. There’s a $60 million ATM equity facility on the shelf — though, to management’s credit, they’ve committed to only issuing shares at or above $5.00, which protects existing holders from dilution at today’s depressed price. Analyst price targets are all over the map, from the low-$2s to $8, and the consensus rating sits at Hold. This is a stock you size as a starter position, not a core holding — which is exactly what I’ve done.

The opportunity exists because of these warts, not in spite of them. The market is pricing the turbulent past and the aircraft misstep, and largely ignoring the doubling earnings and the AI optionality.

7. Latest News and Earnings

The timing here is almost too neat: CareCloud reports Q2 2026 this morning, 6 August 2026, before the open, with the call at 8:30 a.m. Eastern. Consensus is looking for roughly $0.07 of EPS on about $31.9 million of revenue. Given the last few quarters beat on the top line, and given the visible operating leverage from AI automation, this is the print I want to see confirm the thesis. As always, I care less about the headline number than about three things on the call: organic-versus-acquired growth, adjusted-EBITDA margin direction, and any hard metrics on AI adoption.

The other recent developments that matter:

• May 2026 — acquired Empower Healthcare & Compliance Partners, an asset purchase funded from operating cash flow, adding compliance and advisory services and — crucially — another recurring revenue stream to cross-sell into the 45,000-provider base.

• May 2026 — Analyst Day at Nasdaq MarketSite, where management laid out four themes I’ll be holding them to: an AI-first platform, a simplified common-stock capital structure, rising free cash flow, and a disciplined acquisition strategy.

• April–May 2026 — the capital-structure cleanup: the $50 million Citizens/Provident facility and the full Series B preferred redemption, lowering the cost of capital without diluting common holders.

• June 2026 — a credit-agreement amendment refining the financing terms.

The through-line of every one of these is the same: simplify the story, lower the cost of capital, feed the AI-and-M&A engine.

8. Conclusion: A Re-Rating Waiting to Happen

Let me bring it home.

Strip away the noise and CareCloud is a profitable, cash-generative, founder-led business in a large, defensive, fragmented market, trading at a low-double-digit forward multiple while its earnings are set to double. The base case doesn’t require heroics — it requires management to keep doing the unglamorous things it’s already doing: collect a cut of the bills, cross-sell software, buy small shops at one-times-revenue, and let AI quietly strip out cost. On my fair-P/E of 22, that base case is worth roughly 34% a year, driven by earnings rather than sentiment.

And then there’s the optionality, which is where this gets genuinely interesting.

The AI case. stratusAI isn’t a slide-deck buzzword here — the desk agent already automates a large share of inbound calls, and the company grew revenue while shrinking headcount. If the market ever decides to look at CareCloud as an AI-enabled healthcare-automation platform rather than a billing services roll-up, the multiple doesn’t stay at 22. Vertical-SaaS and AI platforms trade at multiples this stock can currently only dream of. That re-rating is the free call option you get on top of the earnings-driven base case.

The M&A case. A disciplined acquirer buying assets at one-times-revenue and making them “smarter, faster, more valuable” by layering AI on top is a compounding flywheel. Every acquisition that lands cheaply and integrates cleanly is accretive and strengthens the data moat. Do that twenty more times, and the story writes itself.

So the setup is: a solid, understandable base case with real downside protection from cash flow and a disciplined balance sheet, plus two genuine re-rating catalysts — AI-platform recognition and continued smart M&A — that the current price gives you essentially for free. That asymmetry is precisely why I’ve opened a starter position in the Haas Invest4 Innovation Fund. Not a table-pounding all-in — the checkered past and the micro-cap volatility earn it a small size — but a position I’m happy to build into as management keeps proving the transformation is real.

Boring billing company. Quietly becoming something more. Those are often the best ones.

Risk Disclaimer

This article reflects my personal opinion and is intended for information and educational purposes only. It is not investment advice, not a solicitation, and not a recommendation to buy or sell any security. I am not your financial adviser, and nothing here accounts for your individual circumstances, risk tolerance, or objectives. Stocks — and micro-caps like CareCloud in particular — can be highly volatile, and you can lose part or all of your invested capital. Figures, guidance, and prices are as of early August 2026 and may since have changed; always do your own research and, where appropriate, consult a licensed professional before investing.

Conflict of interest: CareCloud (Nasdaq: CCLD) is held as a position in the cost-efficient Haas Invest4 Innovation Fund(invest4.net). I therefore have a financial interest in the security discussed and am not neutral. Please keep this conflict in mind when reading.

— Philipp Haas, investresearch.net

Every few years I come across a company that is trying to build something that simply does not exist in Germany yet. SoFi Technologies is one of them. Imagine a single app where you keep your salary account, refinance your student debt, take out your mortgage, buy your ETFs, park your emergency cash at a competitive rate, get your credit card, and even shop for insurance — all under one roof, all mobile, all priced fairly. Everything is fragmented, and the moment you earn a bit more, the incumbents either can’t serve you well or charge you far too much. SoFi has solved exactly this problem in the United States, and its ambition is refreshingly large: to become the primary financial relationship — the Hausbank — for a generation of young, high-earning professionals.

I have wanted to write this piece for a long time. When SoFi came public via SPAC, it was too expensive for my taste and the profitability was still a promise. That promise has now been delivered. The company is GAAP-profitable for the eleventh consecutive quarter, it just posted record Q2 2026 results, and — crucially for a long-term investor — the stock sold off roughly 9% on the print despite an exceptional quarter. That combination, a high-quality compounder marked down on strength, is precisely the setup I look for. On my Fair-PE model I arrive at a fair multiple of 25 and an implied annualized return of about 24%. This is the story of why.

The short version of my thesis: SoFi is a vertically integrated digital bank with its own technology stack, a real banking licence, no branch network, a genuinely beloved brand, and a cross-selling flywheel that has just hit escape velocity. It attacks one of the largest, richest, most inefficient industries on earth — U.S. consumer finance — with a modern brand and a structural cost advantage. It is early, it is still small relative to the incumbents, and the market is currently paying more attention to a single quarter of guidance mechanics than to the durability of a 35%-growth franchise. I think that is a mistake, and I am positioned accordingly.

1. Product, Business Model, Brand and Moat

What SoFi does. SoFi started life refinancing student loans for graduates of top universities — think of it as a modern, digital-native version of what MLP once was in Germany: capture the young academic early, when their income curve is about to bend upward, and grow with them. From that beachhead it expanded relentlessly: personal loans, home loans, credit cards, checking and savings, a brokerage, and third-party insurance. Today SoFi describes itself as the “everything app” for money, and for a growing number of Americans it has become exactly that — the primary place they borrow, save, spend, invest and protect.

How it earns money. There are three engines. First, Lending — personal loans, student loans and home loans — which remains the largest profit contributor and produced record originations of $14.8 billion last quarter. Second, Financial Services — the bank accounts, the card, the brokerage, SoFi Plus subscriptions, interchange — which is where the flywheel lives. Third, the Technology Platform (Galileo and the Technisys core-banking system, now marketed as SoFi Tech Solutions), a genuine B2B business that sells the plumbing — issuing, processing, program management, core banking — to other fintechs, banks and consumer brands. This is the AWS analogy that gets thrown around: SoFi built the rails for itself, learned what a modern financial product actually needs, and now rents those rails to everyone else. Together, fee-based revenue now makes up 39% of the top line — a deliberate, less balance-sheet-intensive, less cyclical mix than a pure lender would carry.

The brand. This is where I get genuinely enthusiastic, because brand in financial services is criminally underrated. SoFi was named the #1 U.S. bank in Forbes’ World’s Best Banks list and took the #1 spot in J.D. Power’s 2026 study for DIY investing. Its unaided brand awareness reached an all-time high of 10% — still low, which to me is upside, not weakness. New customers arrive cheaply through “best-of” placements at the likes of NerdWallet and Bankrate, and increasingly through word of mouth, because — and this is the point incumbents can’t replicate — SoFi can credibly claim to put the customer’s interest first, with cheaper, better products. A bank people actually recommend to their friends is a rare animal.

The moat. SoFi occupies an unusual position: it owns its technology stack, operates with no branch overhead, and holds a national bank charter. That last part matters enormously — it lets SoFi fund loans with its own low-cost deposits (over $45 billion now, more than 90% of its funding base) rather than renting balance sheet from someone else. Pure fintechs have the modern tech but no charter and expensive funding; incumbents have the charter and cheap funding but 1970s technology and branch cost. SoFi is one of the very few players that combines both cost edges at once. Replicating that requires simultaneously building a beloved consumer brand, an owned tech platform, a deposit base and a regulated bank — a decade-long, capital-intensive undertaking. That is a moat.

2. The Market: Enormous, Growing, and Ripe for Disruption

I have a bias, and I’ll own it: I like companies that go after the very largest markets. When the vision is small, most of the upside is already priced in even for a good business. When the vision is Tesla-sized or Netflix-sized and management executes, the achievable market capitalization becomes extraordinary. U.S. consumer finance is one of those markets. Lending, deposits, payments, investing, insurance — the total addressable pools run into the trillions, and the incumbents are fat, slow and still enormously profitable.

That last point is the disruption thesis in a sentence. Just as traditional bricks-and-mortar retail was a soft target for e-commerce, the traditional U.S. bank is a soft target for a digital-native challenger, precisely because it is expensive and still earning very comfortable margins. There is an enormous pool of economics sitting inside legacy institutions that has never been properly contested with a modern brand and modern technology. To frame the runway: JPMorgan is worth well north of half a trillion dollars. SoFi’s market capitalization is roughly $20 billion. This is not a company that has run out of room; it is a company that has barely started, with the ability to keep launching new products into an ever-larger share of each customer’s financial life.

On political risk, U.S. financial services is regulated but stable and rules-based — SoFi navigated the student-loan payment moratorium and a brutal rate cycle without derailing. The genuine sensitivity here is macroeconomic, not political: the lending business is cyclical. If a deep U.S. recession sent credit losses sharply higher, that would hurt. My counter is twofold. First, SoFi deliberately underwrites to a prime, higher-income borrower — people earning six figures who, even in a downturn and even in an AI-disrupted labour market, tend to find work again faster than most. Second, the shift toward fee-based and platform revenue steadily dilutes that cyclicality over time.

3. Culture and Management: Internet DNA Meets Wall Street

Management is a big part of why I trust this one. CEO Anthony Noto is the rare executive who genuinely bridges two worlds. He was a Goldman Sachs partner who ran technology investment banking and helped take Twitter public; he was CFO of the NFL; and he served as COO and CFO of Twitter before taking the helm at SoFi. That combination — deep capital-markets and finance-industry fluency plus authentic internet-and-product DNA — is exactly what you want running a company trying to out-build both banks and fintechs. It also shows in the details: Noto has been a consistent open-market buyer of the stock, which is the kind of alignment I always want to see.

More telling than the résumé is the track record against the plan. SPAC-era projections are notoriously bullish, and most SPAC companies missed theirs by a mile. SoFi did not. It broadly hit its product roadmap and its profitability targets — through a rate-hiking cycle nobody in that original deck had modelled. Delivering GAAP profitability while interest rates were being marched several points higher is, frankly, an underappreciated feat of execution. This is a management team that thinks in years, not quarters, that keeps its promises, and that is building the flywheel patiently rather than chasing vanity metrics.

4. Financials and Margins

The Q2 2026 numbers (quarter ended June 30, 2026) are the best evidence for the thesis:

• Revenue: record GAAP net revenue of $1.22 billion, up 43% year-over-year (adjusted net revenue $1.21 billion, up 40%). Management raised full-year 2026 adjusted-revenue guidance to $4.75–$4.85 billion and lifted its revenue-growth guide to 32–35%.

• Members and products: 15.8 million members, up 35%, with a record 1.1 million added in the quarter; products rose 42% to 24.4 million. For the first time ever, SoFi added twice as many products as members — the clearest possible sign the cross-sell engine is working.

• Cross-buy: 51% of new products were opened by existing members, up from 35% a year ago. Products per member climbed to 1.54. This is the Financial Services Productivity Loop that Noto calls “escape velocity,” and it is the single most important slide in the deck.

• Lending: record $14.8 billion of originations, up 69% year-over-year — $10.7 billion personal, $2.7 billion student, $1.4 billion home.

On margins, which is where quality gets tested:

• Net income margin: GAAP net income of $156.6 million on $1.22 billion of revenue is a ~13% net margin — for a company still investing heavily to grow 40%.

• EBITDA margin: adjusted EBITDA of $357.8 million, up 44%, at a 30% margin. SoFi cleared the Rule of 40 for the 19th consecutive quarter, with a score of 70.

• Return on equity: tangible book value grew 56% year-over-year to $9.5 billion, or $7.34 per share. Annualized returns on that tangible equity are running in the high single digits today and — this is the key — climbing quarter after quarter as operating leverage kicks in over a largely fixed cost base. That is the pattern I care about most in a scaling bank: once revenue clears the fixed-cost hurdle, incremental economics turn sharply positive.

A note on how I value this business. I am wary of leaning on EBITDA for a bank; for financial institutions I care far more about net interest margin, return on tangible common equity, and the trajectory of book value. On that lens, net interest income rose 52% year-over-year to roughly $790 million, net interest margin held at 5.98%, and the balance sheet is a fortress — a total capital ratio of 18.8% against a 10.5% regulatory minimum, funded overwhelmingly by sticky, low-cost deposits. This company can finance its own growth without tapping the market.

One honest blemish: the Technology Platform segment. It was supposed to be a major growth and profitability driver, and instead it has been the laggard — revenue fell about 23% year-over-year and enabled accounts declined to roughly 135 million, largely because one very large client fully transitioned off the platform. Underlying, like-for-like growth is still positive and new partnerships are landing (Wyndham, Southwest debit rewards, and more brands in the pipeline), but I am not going to pretend this segment has executed to plan. It hasn’t. The offset is that Lending and Financial Services have more than carried the company — which is arguably a better outcome than the original script, since it means the consumer flywheel, not a lumpy B2B pipeline, is doing the heavy lifting.

5. Valuation: A Fair PE of 25 and ~24% Annualized

SoFi guides to full-year 2026 adjusted EPS of around $0.60. At a share price in the mid-$15s, that is roughly 25x this year’s earnings. For a legacy bank that would be absurd; for a business compounding revenue at 35% with expanding margins and a widening moat, it is not remotely demanding — and it is a world away from the pre-profit multiples this stock once carried.

Here is how my Fair-PE (faires KGV) framework handles it. I assign SoFi a fair multiple of 25 — appropriate for a high-quality, founder-caliber-led, platform-and-network-effect franchise with durable 25–30%+ growth, but with a deliberate haircut for the cyclicality of the lending book. I then apply that multiple to my three-year EPS path. If adjusted EPS compounds at roughly 30% from this year’s $0.60 base — conservative for a company this early in its S-curve, and management’s own framework points to at least 30% member growth and durable revenue growth well beyond 2026 — earnings reach around $1.00 within about three years. Twenty-five times that, plus the compounding of a tangible book value already growing north of 50% a year, lands me at an implied return of approximately 24% per annum.

I want to stress the margin of safety embedded in that number, because it is the same logic I applied to my Netflix position. First, the growth estimates I’m using are almost certainly below what a company still adding members at 35% will actually deliver — SoFi has beaten guidance for ten-plus straight quarters. Second, I’m buying below fair value even on nearer-term earnings, so I get paid for the multiple re-rating and the earnings growth and the book-value compounding. Whether I value SoFi as a fast-growing fintech or, more conservatively, simply as a bank on its book value and normalized earnings power, it does not screen as expensive. That is a comfortable place to underwrite from.

6. Risks and Why the Opportunity Exists

So why did a stock this good fall ~9% on a record quarter, sitting closer to its 52-week low of $14.88 than its high of $32.73? Understanding the sell-off is the whole opportunity.

The proximate trigger was guidance mechanics, not fundamentals. SoFi raised its revenue guidance but only reaffirmedits EBITDA and EPS targets, and a handful of analysts wanted the profit guide lifted too. A higher tax rate also clipped EPS by roughly half a cent. The market, as it so often does, decided that a beat which wasn’t a bigger beat was a disappointment. Several banks trimmed price targets by a dollar or two while keeping their ratings largely intact.

The deeper worry investors are chewing on is a narrative one: with 51% of new products now coming from existing members, some read a “shift to cross-buy” as code for “new-member growth is peaking.” I read the same fact as the flagship strength of the model — monetizing a base you already acquired cheaply is the good kind of growth, and member additions were still a quarterly record. There is also the shadow of a March 2026 short report (Muddy Waters) that alleged accounting and charge-off issues; notably, SoFi’s cash revenue has continued to track reported revenue closely, and the credit data has since improved rather than deteriorated.

That is the crux. On the actual credit numbers — the thing bears fear most — personal-loan net charge-offs improved to 3.7%, down 70 basis points sequentially, and 90-day delinquencies fell to 40 basis points. In other words, the market punished the stock on tone and guidance conservatism while the underlying credit and growth held up or got better. The macro backdrop did shift — SoFi now assumes one to two rate hikes in 2026 rather than the cuts it originally penciled in — but it delivered these results into that tougher environment, not away from it.

The genuine, non-narrative risks I do respect: a severe U.S. recession lifting credit losses, continued underperformance of the Technology Platform, and the reality that a lender’s earnings are inherently more cyclical than a pure software company’s. I size the position with that in mind. But “strong company, spooked market, unchanged long-term trajectory” is the textbook definition of where a patient investor wants to be shopping.

7. Latest News and Earnings

The July 29, 2026 print is the headline event, and I’ve covered the numbers above. A few forward-looking items from the call are worth flagging. SoFi Plus, now a paid subscription, has already passed 200,000 paid subscribers at roughly $24 million of annualized revenue — a nascent, high-margin recurring stream layered on top of the bank. SoFi Coach, the AI financial-guidance tool, logged nearly half a million conversations with 90%+ positive feedback, and management is leaning hard into AI both for member engagement and for engineering productivity. The Technology Platform’s reacceleration case rests on new brand partnerships — the Wyndham and Southwest debit-rewards programs, with additional “largest yet” brands signed and awaiting launch. And the capital position (18.8% total capital ratio, self-funded growth) means none of this requires diluting shareholders to finance it.

Net-net: revenue guidance up, profit guidance held, credit improving, capital fortress intact, flywheel accelerating. A very good quarter that the tape chose to sell.

8. Conclusion: The Modern Mobile Bank of America

Strip away the quarter-to-quarter noise and here is what SoFi is: a genuinely high-quality business — rare enough that I don’t find many — with an excellent product, a beloved and still-underpenetrated brand, a structural cost advantage almost nobody else can assemble, a management team that keeps its promises and thinks in decades, and revenue and earnings still growing far faster than a maturing peer set where growth has largely flattened. All of that sits on a strong, self-funding balance sheet, at a valuation that is reasonable on my Fair-PE model and outright cheap on tangible book value.

The bull case, if it plays out, is simple and enormous: SoFi becomes the default primary bank for a generation of Americans — the modern, mobile, everything-app Hausbank — the way Netflix became the default for streaming and Tesla forced an entire industry to follow. When one company credibly targets a market that large and executes, the eventual market capitalization can be a multiple of today’s. Think about the strategic logic even from a competitor’s chair: this is the Mercedes-and-Tesla lesson in reverse. Mercedes held a stake in Tesla and sold it right before it compounded — strategically one of the great unforced errors, because owning even a slice of your most dangerous long-term disruptor would have let you watch the whole thing unfold in comfort. SoFi is that kind of disruptor to the U.S. banking establishment. Owning a position in it is, in part, owning insurance on the future of American finance.

There is also a personal reason this one resonates with me. SoFi’s mission — getting more people to make better financial decisions and giving them fair, low-cost products to do it — is not so different from what I try to do with my own research, my channel, and a cost-efficient fund. I like backing companies whose incentives point the same way as their customers’.

On my numbers, a fair PE of 25 and roughly 24% annualized potential, this is exactly the kind of high-quality compounder-on-sale I want to own for the long run.

Risk Disclaimer & Disclosure

This article reflects my personal opinion and is intended for information and educational purposes only. It is expressly notinvestment advice, a recommendation, or an invitation to buy or sell any security. Every investment in equities carries risk, including the total loss of capital; past performance and forward-looking estimates (including my Fair-PE model, growth assumptions and return projections) are no guarantee of future results and may prove materially wrong. Please conduct your own research and, where appropriate, consult a licensed advisor before making any investment decision.

Disclosure of a conflict of interest: SoFi Technologies (NASDAQ: SOFI) is a holding in the cost-efficient Haas Invest4 Innovation Fund (invest4.net), which I manage, and may also be held in my private accounts and Wikifolios. I can therefore benefit from a rising share price, and interests may conflict accordingly. All figures are based on publicly available information as of late July 2026 and may change without notice.

The 500-bagger nobody wanted to touch at 50x — and now can buy at 19x

Let me start with a confession. For years I looked at Vitec Software and did what most quality investors do with a genuinely great business trading at a genuinely absurd price: I admired it, I put it on a watchlist, and I waited. Vitec was one of those rare European names that had turned a modest sum into a fortune — a multi-hundred-bagger over two decades, arguably one of the best-performing stocks the continent has produced. Businesses like that are almost never cheap. The market knows what it owns, and it charges accordingly. At the peak, Vitec traded north of 45–50x earnings, and I simply couldn’t make the math work at that entry point.

Then something happened that I’ve learned to pay very close attention to over my career: a wonderful business fell out of favour. From its 2025 high near SEK 495, the shares have dropped roughly 60%. As I write, Vitec changes hands around SEK 223, the market capitalisation has shrunk from about SEK 20 billion to roughly SEK 9 billion (around €0.8 billion), and the trailing P/E has collapsed from the high-40s to under 20. The underlying business, meanwhile, kept compounding: sales up 15% in the latest quarter, EBITA margin back at 29%, earnings per share up 13%. That is the setup I want to walk you through — because when the fundamentals march north while the price marches south, the gap between the two is where returns come from.

My thesis in one paragraph: Vitec is the Nordic Constellation Software — a disciplined, founder-anchored serial acquirer of tiny, mission-critical vertical software businesses with ~85% recurring revenue and enormous switching costs. The market has de-rated it on three fears — a slower private M&A market, rising financing costs, and the fashionable worry that AI will hollow out legacy niche software. I think at least two of those fears are cyclical and one is overblown. On my numbers, a fair P/E of around 23 and mid-teens earnings growth point to roughly 24% annual return potential from here. This is precisely the kind of quality-on-sale situation I built my process around, and it’s why Vitec sits in the Haas Invest4 Innovation portfolio.

Let me build the case properly.

1. The business: forty-nine little monopolies under one roof

To understand Vitec you have to understand vertical market software (VMS). Forget the flashy, horizontal SaaS names everyone talks about. Vitec buys the unglamorous stuff — the software that runs a specific, narrow slice of a specific industry. Software for car dismantlers in the Netherlands. Software for pharmacies. For real estate agents. For energy and water metering in Swedish housing. For hairdressers, taxi firms, laundries, healthcare providers, banks, insurers. These programs have existed for ten, twenty, sometimes thirty years. They rarely change dramatically. Nobody loves them. And crucially — nobody can operate without them.

How they make money. Around 85% of Vitec’s revenue is recurring — subscriptions, licences, maintenance and support that renew year after year. The company generated roughly SEK 3.63 billion in net sales in 2025. The engine has two cylinders: a slow, steady organic cylinder (price increases and modest volume growth in the underlying niches, running around 4–6% a year) and an acquisitive cylinder (buying more of these little businesses and folding them in). Vitec doesn’t chase revenue for its own sake — it optimises for cash earnings, quietly raising prices, tightening costs, and cross-pollinating best practices across its subsidiaries.

The portfolio. Vitec now runs 49 subsidiaries — up from around 40 a couple of years ago. These are typically businesses doing SEK 50–150 million in sales, often 10–60 employees, each dominant in its own tiny pond. The beauty is the diversification: the single largest customer accounts for only about 8% of group sales, and no one vertical can sink the ship. If a startup disrupts one niche, or one industry hits a rough patch, the other 48 units carry on. That’s not a portfolio of bets — it’s a portfolio of small, boring, cash-generative monopolies.

If I score Vitec the way I score every business — product, business model, brand, moat — it lands close to world-class. The only marks I hold back are for the weak brand and the modest debt load. Everything else is elite.

2. The market: small, defensive, and delightfully dull

Nordic niche software is not a giant addressable market, and I won’t pretend otherwise. But that’s a feature, not a bug. Small niches are unattractive to natural buyers — too small for private equity, too small to IPO, too specialised for the big horizontal software players to bother with. That leaves Vitec as one of the very few credible acquirers, which is a wonderful position to occupy when you’re on the buy side. It also means the underlying niches face little competition; there’s rarely a well-funded rival racing to build a better mousetrap for, say, Finnish hotel administration.

The demand is structural and defensive. These industries — pharmacies, real estate, healthcare, energy metering, insurance — need their software regardless of the economic weather. That makes revenue remarkably acyclical. Political intervention risk is low and diffuse: spread across the Nordics plus the Netherlands, Poland and a few other geographies, with dozens of unrelated verticals, there’s no single regulator or subsidy that can move the whole business. And the total market keeps expanding as more of these industries digitise and as Vitec pushes outward from Scandinavia into continental Europe. Small, yes. But durable, growing, and beautifully insulated from the macro noise that batters more cyclical names.

3. Culture and management: the founder is still watching

This is where I want to correct a common misconception, because the leadership story at Vitec has changed and it matters.

Vitec was founded in 1985 in Umeå by two university researchers, Lars Stenlund and Olov Sandberg, who set out — I love this origin — to help Sweden through the energy crisis by writing software for property owners to monitor energy use, coded in Turbo Pascal on beige Ericsson machines. Stenlund ran the company as CEO for 36 years. But he is no longer the chief executive. In April 2021 he handed the reins to Olle Backman, previously Vitec’s CFO and the man who had been central to its acquisition machine. Stenlund moved up to Chairman of the Board, where he remains the largest shareholder by voting power through his Class A holdings. Co-founder Sandberg stepped back from operations years ago but is still a principal owner.

I actually think this is close to the ideal governance setup. You have a finance-and-M&A-native CEO in Backman running the operational, deal-making day-to-day — exactly the skill set this business needs — while the founder sits above him as Chairman, safeguarding the long-term philosophy and the culture he spent four decades building. Insider ownership sits around 12.5%, so management eats its own cooking. Backman himself has been buying shares personally through the downturn, which is the kind of behaviour I like to see when the crowd is selling.

The culture is Buffett-like and deliberately decentralised. M&A is run centrally from headquarters; the subsidiaries are left alone to run their businesses, rewarded on performance, and treated as a permanent home rather than a company to be stripped and flipped. Managers of acquired firms know Vitec won’t gut them — which is precisely why founders of these little niche businesses, often facing a succession problem with no other buyer, choose to sell to Vitec. The strategy is patient, disciplined, and measured in decades. That long-term orientation is the rarest and most valuable asset a serial acquirer can have.

4. Financials: the earnings are better than they look

Now to the numbers — and to a subtlety that trips up a lot of investors looking at Vitec for the first time.

Growth. Over the past decade Vitec has compounded revenue and earnings in the mid-teens, blending low-single-digit organic growth with acquisitions. Earnings per share have grown around 11% annually over the last three years even through a margin wobble. In 2025, net sales reached SEK 3,633 million with operating cash flow of SEK 1,110 million — a business that converts profit into cash beautifully, with cash conversion around 80%.

Margins — and the hidden-earnings point. On a reported basis, net margin is around 12–13% and EBITA margin runs in the high-20s (29% in the latest quarter). But look at the EBITDA margin and you see something different: roughly 36%, on EBITDA of about SEK 1.3 billion. Why the enormous gap between a 36% EBITDA margin and a 12% net margin? Amortisation of acquired intangibles. Every time Vitec buys a business, accounting rules force it to book and then amortise the acquired customer relationships and software over years. That amortisation is a real charge on paper but not a cash cost — it’s the accounting residue of past acquisitions, not money going out the door. The practical upshot: Vitec’s true cash earnings power is meaningfully higher than the reported net income and the headline P/E suggest. This is the classic serial-acquirer distortion, the same one that makes Constellation Software look optically expensive on GAAP earnings while gushing cash underneath.

Return on equity. Reported ROE currently sits around 8–9%, below its ten-year average near 15%. I don’t panic about this. Two things are depressing it: the recent margin dip (now reversing) and, more structurally, a balance sheet loaded with goodwill and intangibles from all those acquisitions, which inflates the equity base and drags the ratio down. On the cash actually deployed into the operating businesses, the returns are far healthier than the accounting ROE implies. Net debt to EBITDA sits around a manageable 1.9x — some leverage, which I flag as a genuine (if modest) risk, but nothing that keeps me up at night for a business this cash-generative.

Recent trajectory. After a soft patch in mid-2025 when margins slipped and the market took fright, profitability has turned back up. Management is executing on its long-standing target of gradual margin improvement, holding organic headcount flat while revenue grows — pure operating leverage. Earnings are once again outpacing sales, which is exactly the signature you want from a compounder finding its footing.

5. Valuation: paying 19x for something worth 23x — and growing

Here’s the part that turned my long-standing admiration into an actual position.

At roughly SEK 223 against trailing EPS near SEK 11.4, Vitec trades around 19x earnings — and closer to 18x on forward estimates for 2026. Set that against its own history, where the stock routinely commanded 40–50x, and against the Swedish software sector’s average forward multiple of about 23x. Vitec is now cheaper than its peer group and less than half its own historical multiple, despite being the same high-quality, mission-critical, recurring-revenue machine it always was.

I anchor my valuation on a fair P/E of 23 — roughly the software-industry norm and, in my view, a reasonable multiple for a business with 85% recurring revenue, high switching costs, a disciplined capital allocator at the helm, and a long acquisition runway. That’s a deliberately conservative fair value: I’m not asking the market to pay up for Vitec the way it did in its euphoric years. I’m simply assuming it re-rates from a de-pressed ~19x back toward a merely-normal ~23x.

Now stack the return drivers:

• Earnings growth. I model around 15% per year — low-single-digit organic plus the acquisition cylinder firing again as the private M&A market thaws. Remember, reported EPS understates true cash growth because of that amortisation drag.

• Multiple re-rating. From ~19x toward a fair ~23x adds a meaningful one-time lift as sentiment normalises.

• Dividend. A modest ~1.5% yield on top, growing over time.

Blend those together and the base case points to roughly 24% annual total return over a multi-year horizon. That is an unusually attractive expected return for a business of this quality — the kind of asymmetry that only shows up when a great company is temporarily unloved. Analysts, for what it’s worth, carry an average target around SEK 445, and independent intrinsic-value estimates cluster in the high-SEK-300s. I don’t need Vitec to return to its old glory multiple for this to work. I just need it to stop being cheap while it keeps compounding.

6. Risks: why is it down, and why does the opportunity exist?

I never present an investment case without being straight about why the price is where it is. A 60% drawdown is not nothing, and markets aren’t always wrong. Here is what has actually happened, and why I think it created opportunity rather than a value trap.

It was priced for perfection. At 45–50x earnings, Vitec had zero margin for error. Any stumble was going to hurt, and it did. Much of the fall is simply the air coming out of an over-inflated multiple — a de-rating, not a business collapse. That distinction is everything.

The mid-2025 margin scare. After a Q2 2025 report showed margins deteriorating, the stock dropped 16% in a single day. Rising financing costs — Vitec funds acquisitions partly with debt, and rates went up — pressured the model, and the market extrapolated the wobble into a trend. Margins have since recovered, but the fear lingered.

The M&A engine stalled. Vitec’s growth flywheel depends on buying businesses. Through 2025 and into 2026 the private M&A market cooled: sellers became hesitant, deals took longer, and in the most recent quarter no acquisitions closed at all. The market’s worry is legitimate — if the acquisition cylinder seizes up, the growth story weakens. But management has been explicit and disciplined: it is not overpaying to force deals, it has lowered its own price expectations, and it is keeping a full pipeline warm, with an appetite for larger targets when sellers get realistic. A temporary pause in dealmaking is very different from a broken model. This is cyclical, and cycles turn.

The AI fear. This is the fashionable one, and in my view the most overblown. The narrative goes: nimble AI startups will build better niche software cheaply and disrupt legacy providers, while AI generally lowers the barrier to writing software. It’s a plausible-sounding story that ignores how this market actually works. A competitor building a superficially similar product has always been possible — that’s not the hard part. The hard part is persuading a cautious, risk-averse customer to rip out the mission-critical system running their business and accept the migration risk, for a category of software nobody wants to think about. AI doesn’t solve that trust-and-switching-cost problem; if anything it makes the incumbent with the data and the relationships stronger. And Backman has flipped the narrative on the earnings calls: Vitec is using AI as an efficiency tailwind internally and as a way to enhance its products, not cowering from it.

So: the opportunity exists because a high-quality business got over-valued, then got hit by a real-but-cyclical M&A slowdown and margin scare, then got tarred with a structural AI fear I think is misplaced — and the multiple over-corrected on the way down. That combination is how you get a world-class compounder at 19x.

7. Latest news and earnings

The most recent data point is the Q2 2026 report, published in mid-July, and it’s a quietly strong one that the market largely shrugged off — the shares were trading near their 52-week low even as the numbers came in solid.

Net sales rose 15% to SEK 935 million, of which about 4% was organic and the rest acquisition-driven. EBITA grew 15% to SEK 271 million at a 29% margin, cash EBIT was up 18%, and EPS climbed 13% to SEK 2.98 (first-half EPS up 16% to SEK 5.47). Recurring revenue held at 85%. Half-year operating cash flow was a robust SEK 876 million. Management struck a cautiously optimistic tone, flagging rising customer activity in healthcare, the public sector, PropTech and among real estate agents, while noting a lag before that activity shows up in reported revenue.

Two housekeeping items worth understanding. First, Vitec changed how it reports parts of its transaction-based revenue (moving to “agent” accounting for two subsidiaries), which optically reduces reported net sales by around 8% but improves margins by roughly two points — and crucially has zero effect on cash flow, financial position, or EPS. It’s cosmetic, and it actually flatters the margin profile. Second, on capital allocation: the earlier part of 2026 saw two acquisitions close — Autonet, a Dutch vehicle-dismantling software business, and Infometric, a Swedish energy-and-water metering business — bringing in around 75 new colleagues. In 2025 the deals were Intergrip in the Netherlands and NMG in Poland. The pipeline remains active, with management specifically eyeing larger targets. The dividend for 2025 was set at SEK 3.68, paid in quarterly instalments. Headcount now stands near 1,870.

The takeaway from the recent reporting: the operating engine is humming, margins are expanding again, cash generation is strong, and the only genuine soft spot is the pace of acquisitions — which is a market-timing issue, not a quality issue.

8. Conclusion: quality on sale, and a fear I’m happy to take the other side of

Let me bring it home. Vitec Software is one of the highest-quality businesses in Europe: a decentralised, founder-anchored, disciplined serial acquirer of mission-critical niche software, ~85% recurring, diversified across 49 tiny monopolies, run by a capital allocator who’s buying his own shares while the crowd sells. For most of the last decade it was too expensive for me to own. It isn’t anymore.

The market has handed us a ~60% drawdown built on a cocktail of an over-stretched starting multiple, a cyclical M&A slowdown, a temporary margin scare, and a structural AI fear that I believe misreads how switching costs and customer trust actually work in this corner of software. Strip those away and you have a business still compounding earnings in the mid-teens, now available at ~19x against a fair multiple I put at 23x — a combination that frames roughly 24% annual return potential from here.

The clearest investment case is exactly the AI-fear-overblown one: if the market is wrong that agile AI disruptors will unseat entrenched, mission-critical incumbents — and I think it is — then the de-rating reverses, the acquisition engine re-accelerates as the private market thaws, and you’re left owning a proven compounder at a starting price its long-term history says is a gift. In downturns, capital tends to flee toward quality anchors like this, of which Europe has precious few. My discipline says: when a wonderful business goes on sale for reasons that are cyclical and sentiment-driven rather than structural, you buy it. That’s why Vitec is in the portfolio.

Risk disclaimer

This article reflects my personal opinion and analysis and is not investment advice, a recommendation, or a solicitationto buy or sell any security. Investing in equities carries the risk of partial or total loss of capital. Share prices can be volatile, past performance is no guarantee of future results, and the assumptions in my valuation (earnings growth, fair multiple, expected return) are estimates that may prove wrong. All figures are drawn from publicly available data at the time of writing and may since have changed. Please do your own research and consult a qualified financial adviser before making any investment decision.

Disclosure: Vitec Software is part of the portfolio of the cost-efficient Haas Invest4 Innovation investment fund (invest4.net). I therefore hold a financial interest in the security discussed, which may give rise to conflicts of interest.

I have a soft spot for stories that go like this: a small company grows its top line by nearly 50%, flips from years of losses into a proper profit, keeps compounding — and the share price does the exact opposite of what the numbers suggest it should. That mismatch between the business and the ticker is where I make most of my money, and it is exactly what pulled me toward a tiny Japanese name almost nobody outside Tokyo has heard of: PRONI Inc., ticker 479A on the Tokyo Stock Exchange Growth market.

PRONI runs Japan’s answer to a software-and-service discovery platform. If you are a small or mid-sized Japanese company and you have no idea which accounting cloud, which HR tool, which web agency or which SaaS product to buy, you go to PRONI Aimitsu (”aimitsu” roughly means “get comparative quotes”), you describe what you need, and PRONI matches you with vetted providers and pulls in quotes. Think of it as a Japanese blend of G2, Capterra and OMR Reviews, with a lead-generation and matching engine bolted on top. The vendors pay to be found. The buyers save weeks of research. PRONI sits in the middle and takes a cut of a decision Japanese SMEs are being forced to make in their millions right now: how do we finally digitalize?

Here is the short version of the case. The stock peaked at ¥2,099 in January 2026, shortly after its IPO, and has since roughly halved to around ¥1,220. Over the same window the business grew revenue ~47% and delivered its first full year of net profit. The market handed me a ~30%+ grower, with a return on equity north of 30%, at a trailing P/E of about 11 and a forward multiple on 2026 earnings closer to 8. Against a fair multiple I would happily assign this profile — I use 21 as my fair P/E — the maths points to a possible return in the region of ~70% per year if the company keeps executing and the market eventually pays up. That is the prize. The catch, and there is always a catch, is that this is an illiquid, freshly-listed Japanese micro-cap trading under a genuine cloud of “will AI make software-discovery platforms obsolete?” I think that fear is overdone. Let me walk you through why.

(Full disclosure up front: the Haas Invest4 Innovation Fund holds a small position, so I am talking my own book. More on that at the very end.)

1. Product, business model, brand and moat

What they do. PRONI operates PRONI Aimitsu, a B2B procurement-and-matching marketplace covering a surprisingly wide spread of categories — IT production and systems, advertising and sales promotion, HR and back-office, general business management, BPO/outsourcing, and specialist services. The crown jewel, and the fastest-growing slice, is PRONI Aimitsu SaaS, the arm dedicated specifically to helping companies pick IT products and SaaS to drive their digital transformation (”DX,” the term you cannot avoid in corporate Japan). More recently they layered on PRONI AI(launched February 2025) to help SMEs prep sales negotiations, plus tools like the “Industry DX Ultimate Navigator.” So they are not passively watching the AI wave — they are surfing it.

How they earn money. This is a classic two-sided marketplace. Buyers (SMEs) use the platform for free to describe a need and receive matches and quotes. Suppliers (SaaS vendors, agencies, outsourcers) pay for qualified leads and visibility. PRONI monetizes the introduction. The beauty of this model is that it is asset-light and structurally high-margin: there is very little cost of goods, so the entire game is about acquiring buyer demand cheaply (largely via content and search) and monetizing it against a base of paying vendors. You can see this in the cost structure — the dominant expense line is sales, marketing and G&A, not any factory or inventory.

Brand. For a company this size, the brand punches above its weight in its niche. “Aimitsu” is a recognized destination for Japanese businesses looking to compare and buy, the platform picked up an Advanced Business Model Award at the ASPIC Cloud Awards in 2024, it carries ISO 27001 information-security certification, and it keeps landing in the Japanese business press as one of the up-and-coming DX names. It is not Rakuten. But in the specific corner of “help me choose software,” it has real mindshare, and that mindshare is compounding as traffic and vendor count grow together.

Moat. Let me be honest rather than promotional here, because the moat is the crux of the whole debate. The durable advantages are (1) a two-sided network effect — more vendors make the platform more useful to buyers, more buyer demand attracts more paying vendors — and (2) a growing proprietary dataset on what SMEs want and which providers actually convert, which improves matching over time. Those are real. But the moat is not a fortress. Switching costs for a one-off procurement decision are low, the “aimitsu” concept is somewhat generic, and the entire discovery layer sits downstream of search traffic. Which brings us straight to the bear case, so let’s not pretend otherwise: if AI assistants let an SME simply ask a chatbot “which accounting SaaS should I buy and connect me to three vendors,” where does that leave the matchmaker? My answer — and I’ll develop it later — is that curation, trust, vetting and a live vendor network are not the same thing as a search box, and PRONI is building the AI layer itself rather than being blindsided by it. But anyone who tells you the moat here is wide is selling you something.

2. The market

The tailwind is the whole reason I am interested. Japan is home to millions of small and mid-sized enterprises, and by developed-world standards a shocking number of them are still running on paper, fax and spreadsheets. The government has spent years pushing “DX” as a national priority, and — more powerfully than any policy — Japan’s brutal demographic labor shortage is forcing companies to automate simply to survive with fewer workers. Every one of those digitalization decisions is a moment where a business owner has to choose software and a service provider, and most of them have no idea how. That is PRONI’s addressable moment, and it is expanding structurally, not cyclically.

On the risk axis: this is about as politically insulated a business as you can find. It is not exposed to tariffs, export controls, price regulation or geopolitics in any meaningful way — it is a domestic software marketplace. There is some cyclicality, because SME IT budgets are discretionary and get trimmed in a downturn, but the secular shift from analog to digital is a stronger force than the business cycle over any reasonable holding period. If anything, a labor crunch makes the automation case more urgent when times are tight.

3. Culture and management

I like founder-led companies where the founder still has skin in the game, and PRONI qualifies. Norio Kuriyama founded the business back in 2012 and remains a Representative Director with a meaningful equity stake of roughly 3.9% — real money for an individual, and real alignment. Day-to-day leadership sits with Daisuke Shibata, Representative Chairman & CEO, who has been on the board since 2018 and in the top seat since October 2023, and who also holds shares directly.

What I read from the corporate actions is a management team behaving like long-term operators rather than IPO-and-cash-out promoters. They ran a 1-for-10 stock split in September 2025 to broaden accessibility ahead of listing, they went public in December 2025 raising roughly ¥3 billion — capital that cleared the historical deficit and funds growth rather than lining founders’ pockets — and they have kept shipping product (PRONI AI, the DX navigators) instead of coasting. They also strengthened governance around the listing with external directors and auditors. It is a young public company with a short public track record, so I hold this view with humility, but the pattern is entrepreneurial, product-driven and reinvestment-oriented. That is the culture I want behind a compounder.

4. Financials, margins and recent developments

This is where the story gets loud. Here is the multi-year trajectory (fiscal years ending December, JPY):

• FY2023: revenue ¥1.68bn, net income −¥729m

• FY2024: revenue ¥2.20bn (+31%), net income −¥270m

• FY2025: revenue ¥3.23bn (+47%), net income +¥533m

So in two years revenue nearly doubled while the company crossed from deep losses into a genuine, ~16.5% net margin. That is textbook operating leverage in a marketplace: once buyer demand and vendor spend clear the fixed cost of running the platform and the marketing machine, incremental revenue drops toward the bottom line fast. You can see it in the quarterly cadence too — a recent quarter printed record revenue around ¥927m, and quarterly net income jumped from roughly ¥67m to ¥177m sequentially. Momentum is accelerating, not fading.

On quality of returns, return on equity sits north of 30% — an unusually high figure for a company this early, and a signal that the model doesn’t need much capital to grow. The business carries no dividend (correct at this stage — every yen should be reinvested), is essentially asset-light, and while PRONI doesn’t cleanly break out EBITDA, backing into it from the net margin and minimal depreciation puts the EBITDA margin plausibly in the low-to-mid 20s% range, with room to expand as that heavy marketing/G&A line scales more slowly than revenue. Treat that EBITDA figure as my estimate rather than a reported number.

The recent developments that matter: the first full year of profitability (FY2025), continued 40%+ growth into early 2026, the roll-out of the company’s own AI tools, and — crucially for a value-conscious buyer — a share price that fell while all of this improved.

5. Valuation — the fair-P/E case

Now to the number that made me open a position. I value businesses like this on a fair P/E framework: what multiple should a durable ~30%+ grower with 30%+ ROE and a structural tailwind trade at, and what return do I earn getting from today’s multiple to that fair one while earnings compound underneath me.

Start with the starting point. On my 2026 earnings estimate — assuming growth normalizes toward 30–40% with modest margin expansion — PRONI is trading at a forward P/E of roughly 8. (On a trailing basis it’s closer to 11; the “8” is the forward figure, and it is the honest anchor for a growth company, so that’s what I use.) For this profile I assign a fair P/E of 21. That is not aggressive — at 21x on ~30% growth you are paying a PEG below 1, which for a high-ROE, asset-light compounder is conservative if anything.

Here is the engine of the return. If I buy at a forward P/E of ~8, earnings grow at ~30% a year, and the market re-rates the multiple from 8 toward my fair 21 over roughly three years, the total return compounds from two sources at once — the earnings growth and the multiple expansion. Run that: earnings up ~2.2x over three years, multiple up ~2.6x, and the combined effect implies a share price several times higher, which annualizes to a possible return in the neighborhood of ~70% per year. That is the mechanical output of the fair-P/E model, and it is why a boring-sounding matchmaker can be a genuinely exciting risk/reward.

I want to be crystal clear about what that 70% is and isn’t. It is a scenario, not a promise. It rests on three assumptions stacked on top of each other: that growth stays around 30%+, that the starting multiple really is ~8 on 2026 earnings, and that the market eventually agrees with my fair value of 21. Knock any one of those over and the number comes down hard. But even if you haircut every assumption — say growth slows to 25% and the re-rating only reaches 15x — you are still looking at a return profile most large caps can’t touch. The asymmetry is the point.

6. Risks — and why the opportunity exists at all

The single most important question in any deep-value situation is: why is it cheap? If you can’t explain the discount, you are probably the sucker. Here I can explain it, and the explanation is what gives me conviction.

Why the stock fell. Three forces stacked up. First, post-IPO gravity: the stock spiked to ¥2,099 in the euphoric first weeks of trading in January 2026 — a level that priced in perfection — and then did what freshly-listed small caps almost always do, which is give a lot of it back. Second, the AI-disruption fear: 2026 has been the year the market started asking whether AI chat assistants will disintermediate exactly this kind of software-discovery and matching platform, and PRONI got tarred with that brush. Third, the small-cap-Japan derating: thin liquidity, low float, minimal analyst coverage and a −0.3 beta mean this thing trades on its own weather system and can be sold off hard by very little volume. Stack a ~50% AI-narrative haircut on top of an IPO give-back on top of a tiny float, and you get a stock that halved while the business grew 47%.

The real risks going forward, in order of how much they keep me honest:

• AI disruption. The genuine long-term threat. If LLMs become the default front door for “which software should I buy,” the discovery layer erodes. My counter: matching, vetting, trust and a live paying-vendor network are a different product than a search box, and PRONI is building its own AI rather than being run over by someone else’s. But I hold this position because the risk is real and priced in, not because it’s absent.

• Micro-cap fragility. Illiquidity, a short public history, potential lock-up expiries, and forecast risk that comes with a company that has only just turned profitable.

• Traffic dependence. A meaningful chunk of buyer demand flows through search; changes to Google’s algorithm or the rise of AI answer-engines could raise customer-acquisition costs.

• Competition and low switching costs. Others can and do target the same SME DX budget, and buyers aren’t locked in.

• Key-person and small-team risk, plus, for us as euro investors, yen FX exposure.

None of these are disqualifying. All of them are the reason I can buy a 30%+ grower at 8x forward.

7. Latest news and earnings

The most recent full-year print (FY2025, to December) is the headline: revenue ¥3.23bn up ~47%, and the first clean year of net profit at ¥533m. Into early 2026 the quarterly figures showed record revenue and a sharp step-up in quarterly net income, so the growth engine did not stall after the IPO — it kept accelerating. On the product side, the launch of PRONI AI and the DX navigator suite over the past year tells me management is leaning into the AI transition offensively.

The tone I pick up in the wider conversation — on the small-cap-Japan corners of X/Twitter and in the Substack circles I read and write in — is a tug-of-war I find very familiar. One camp sees “Japanese marketplace + AI-disruption risk + halved chart” and won’t touch it. The other camp, the one I sit in, sees a profitable, 47%-growing, high-ROE business that the market repriced on a narrative rather than on its numbers, and treats the drawdown as the opportunity. When the crowd is arguing about the story while the financials quietly compound, that is usually my favorite kind of setup.

8. Conclusion — why PRONI is interesting, and the investment cases

Strip everything back and PRONI is a simple proposition: a founder-aligned, asset-light, ~30–47% growing marketplace, with 30%+ ROE and its first profits on the board, riding a decade-long structural tailwind (Japanese SME digitalization forced by demographics), that the market has repriced from a P/E in the mid-20s to a forward P/E around 8 on the back of an AI fear I believe is overstated.

I see a few distinct ways this plays out:

• The re-rating case. Growth simply continues, the AI fear fades as PRONI demonstrates it can coexist with (and use) AI, and the multiple drifts back toward my fair 21. This is the ~70%-a-year scenario, and it needs nothing heroic — just execution and time.

• The compounder case. Even with no re-rating, a business growing earnings 30%+ with high ROE that keeps reinvesting is worth a lot more in five years than today. You get paid by the growth alone.

• The M&A case. A profitable, sub-¥10bn platform with a strong niche brand, real data and a two-sided network is exactly the kind of asset a larger Japanese IT or media group — or a global marketplace player — might decide is cheaper to buy than to build.

The thing that could break the thesis is equally clear: if AI genuinely hollows out the software-discovery layer faster than PRONI can adapt, growth rolls over and the cheap multiple turns out to be a value trap, not a bargain. That is the bet. I’ve sized it as what it is — a high-conviction idea in a high-risk wrapper — and I think, at these levels, the odds are on my side.

Risk disclaimer & disclosure

This article reflects my personal opinion and analysis for information and educational purposes only. It is not investment advice, not a recommendation, and not a solicitation to buy or sell any security. I am not your financial advisor, and nothing here is tailored to your personal circumstances.

PRONI Inc. (TSE Growth: 479A) is a small, freshly-listed, illiquid Japanese micro-cap. Such stocks carry above-average risk, including sharp volatility, wide spreads, limited liquidity, currency risk for non-yen investors, and the real possibility of substantial or total loss of capital. Forward-looking statements, estimates, “fair P/E” figures and return scenarios are assumptions that may prove wrong; actual results can differ materially. Financial data is drawn from publicly available sources and may contain errors or be out of date — please verify independently before acting.

Conflict of interest: PRONI is a position in the Haas Invest4 Innovation Fund (invest4.net), the cost-efficient investment fund I manage, and I am therefore indirectly invested in this stock. This creates a conflict of interest, as I may benefit from a rising share price. Do your own research and make your own decisions.

It started, as a surprising number of my investment ideas do, with a phone call from my father-in-law.

Petrol prices had crept up again over the summer, and he’d started doing the arithmetic that half of Germany seems to be doing right now: maybe the next car — or at least the second car, the one you use to nip into town and spare the good one — should just be electric. He’d been looking around, and the name he mentioned wasn’t Tesla, wasn’t VW, wasn’t even BYD. It was Leapmotor.

I’ll admit I had to sit up. I’ve been hunting for years for a sensible second-car EV — nothing fancy, no luxury pretensions, just something cheap and cheerful to run local errands. And here was a brand I follow as an investor showing up unprompted in a family conversation as a product. That’s usually the first sign that something is quietly crossing from “China story” into “European reality.”

So let me lay out the case, because I think Leapmotor is one of the more interesting risk/reward setups in my universe right now — and, full disclosure up front, it’s the only pure electric-vehicle maker I currently hold in the Haas Invest4 Innovation Fund.

The short version: this is a company growing deliveries at roughly 60% year-on-year, that turned its first-ever annual profit last year, that has Stellantis as a strategic anchor and a European production route to sidestep import tariffs — and it trades at a market capitalisation of around €5–6 billion with something like €2.5 billion of net cash sitting on the balance sheet. Strip the cash out and you are paying a low-to-mid single-digit multiple of the earnings this business is capable of producing. In an EV sector that has burned an enormous amount of investor capital, that combination is rare enough to demand a proper look. My Fair-PE model — I’ll walk through the maths later — points to a fair multiple of 21x and an implied return in the region of 58% per year. Let me earn that number.

1. Product, business model, brand and moat

What the company does. Leapmotor (Zhejiang Leapmotor Technology, HKEX: 9863) designs, builds and sells battery-electric and range-extender vehicles. Founded in 2015 and headquartered in Hangzhou, it now fields a genuinely broad line-up organised into four families: the A-series and B-series entry models, the C-series family SUVs and sedans, and the newer D-series premium pieces, topped and tailed by the tiny T03 city car. The model my father-in-law had his eye on — the A10 (sold internationally under the name B03X) — is a compact SUV built on the company’s newest A-platform, launched in China in spring and now opening for orders in Europe. Think of it as the segment that a BMW X1 buyer might cross-shop, only at a fraction of the price.

How they make money. Overwhelmingly by selling cars, plus components, charging and after-sales services. But the interesting part of the model isn’t the revenue line — it’s the cost line. Leapmotor is unusually vertically integrated, and its whole philosophy is what management calls full-stack in-house development. Its architecture shares an extraordinary ~88% of parts across models, which is how a company this young can spin up a dozen nameplates without drowning in complexity. That parts-sharing discipline is the real engine of the story: it’s what lets them undercut rivals on price and claw their way to profitability at the same time.

Brand. Still small in absolute terms, but the trajectory is what matters. In China, Leapmotor has become the No. 1 among the new-energy start-ups by deliveries — ahead of the Nios and XPengs it’s usually lumped in with — and it crossed 1.5 million cumulative deliveries in June, a milestone it reached just eight months after passing one million. In Europe the brand is being carried in on Stellantis’s dealer and financing rails, which is precisely why it’s showing up as leasing offers (including headline promotions around €49 a month for the little T03) rather than as a name you’d have recognised two years ago.

Moat. Let me be honest about this, because it’s the part that requires the most humility. Cars are not software; there is no structural moat here of the kind you’d find in a payments network or an exchange. What Leapmotor does have is a cost moat and a technology-heritage moat. The cost moat comes from vertical integration and parts-sharing. The heritage moat is more subtle and, I think, underappreciated — which brings me to the founder.

2. The market

The market Leapmotor plays in is enormous, still growing, and — this is the crucial nuance — bifurcating. Chinese NEV penetration keeps climbing, but the results are no longer uniform. In the most recent monthly data, the young, product-cycle-fresh makers (Leapmotor among them) set records while two established rivals actually went backwards year-on-year. The market is rewarding fresh product over incumbency, and Leapmotor is currently on the right side of that line.

Two honest caveats. First, this is a cyclical, capital-hungry, brutally competitive industry. Nobody makes truly spectacular through-cycle returns building mass-market cars — I said as much to my father-in-law, and I’ll say it to you. Second, it is politically exposed. EU tariffs on China-built EVs are the whole reason the European production route matters so much (more below), and any EV story today is a bet on the direction of trade policy as much as on the product. That said, I actually think we’re heading into a cleansing of this sector — Volkswagen is cutting jobs hard, capacity is coming out on the legacy side, and a chunk of those lost sales is migrating to Chinese manufacturers both inside China and, increasingly, abroad. Even Tesla is broadly flat in China. Consolidation tends to reward the low-cost survivors, and Leapmotor is building itself to be exactly that.

3. Culture and management

Here’s the piece I find genuinely compelling.

Leapmotor’s founder and chairman, Zhu Jiangming, is not a first-time entrepreneur playing with venture money. Back in the 1990s he co-founded Dahua Technology — today one of the world’s largest video-surveillance and imaging companies. I want to correct a common assumption here: he did not fully cash out and walk away from that business. He remains a Dahua vice-chairman and shareholder; what he did was step back operationally to pour himself into Leapmotor. So this is a proven operator with real skin in a prior success, choosing to concentrate rather than diversify.

Why does that matter beyond the usual “we like founders” reflex? Because Dahua’s DNA is imaging, embedded software, chips and algorithms — and Zhu deliberately carried that DNA into a car company. Leapmotor developed its own automotive-grade intelligent-driving chip in-house. Its whole engineering culture came out of surveillance-grade embedded systems, not out of a battery lab. That’s a structurally different starting point from BYD, whose genius is fundamentally chemistry and batteries.

I’ll come back to why I think that origin story is a feature and not a footnote. On the governance side, the shareholder register is reassuring for a Chinese small-cap: the founder group holds around 22–23%, Stellantis around 19%, and FAW took a 5% strategic stake. Long-term strategy is stated plainly — one million units this year, and a stated long-term ambition of four million. Founders who commit to numbers that specific tend to organise the whole company around hitting them.

4. Financials, margins and recent developments

The financial inflection is the reason this is investable rather than merely interesting.

Growth. 2025 was the breakout: nearly 600,000 vehicles delivered, revenue up around 101%, and — for the first time ever — a full-year profit (net income of roughly CNY 540 million). That momentum has carried straight into 2026. First-half deliveries came in at about 356,000 units, up roughly 60% year-on-year, with the second quarter up around 84% and June setting an all-time monthly record. Overseas deliveries have grown to more than 12% of the mix — the international flywheel is spinning up.

Margins — the operating-leverage thesis. Let me not oversell the current state of profitability, because it’s thin. Gross margin sat around 14–15% last year, but net margins are still barely above breakeven and return on equity is in the low single digits. This is a company that is just past its profitability inflection, not one harvesting mature margins. The entire bull case is that these margins scale: as volume marches toward a million units, as the higher-priced overseas mix grows, and as that 88% parts-sharing architecture spreads fixed costs across ever more cars, net margin should expand from near-zero toward mid-single digits. On this revenue base, that is the difference between CNY 540 million of profit and the CNY 5 billion management is targeting for this year. Small changes in margin, multiplied by a doubling of volume, produce very large changes in earnings. That’s the whole game.

Recent developments. The product cadence has been relentless — refreshed C10, C11 and C16 SUVs moved onto an 800-volt architecture with up to 660 km of range and starting prices around CNY 126,000 (~€16,000); the D99, the company’s first premium MPV, launched into a segment Leapmotor had never touched; and the A10/B03X opened for European orders. Assembly has also begun outside China — a Stellantis plant in Malaysia is now building the C10, with the B10 to follow.

5. Valuation — the Fair-PE case

Now to the number that matters, and I’ll show my working because that’s the whole point of my Fair-PE (Faires KGV) approach: project a few years of earnings, apply a justified multiple, and let the implied annual return fall out.

Start with what you’re paying. At a share price in the region of HK$37, the market capitalisation is roughly €5.5–6 billion. Against that sits about €2.5 billion of net cash. So the enterprise value — what you’re actually paying for the operating business — is only around €3–3.5 billion. On the company’s own CNY 5 billion profit target for this year, you’re paying a high-single-digit multiple on the market cap and something closer to 4–5x on a cash-adjusted basis. Even on more conservative sell-side estimates, the forward multiple is only in the low teens. For a business compounding deliveries at 60%, that is cheap however you cut it.

Here’s how I get to a fair value. I’m prepared to pay a fair PE of 21x. That is deliberately below what I’d assign to a great software compounder — I’m discounting hard for the ugly economics of the car industry and for the China-listing risk — but well above a no-growth multiple, because the growth here is real and the cost structure is genuinely differentiated. Apply that 21x multiple to the normalised earnings power I expect this business to reach as the margin-scaling and volume story plays out over roughly a three-year horizon, and the fair value lands multiples above today’s price. Bridging from ~HK$37 to that fair value over three years implies an annualised return of about 58%.

Let me be clear about what that 58% is and isn’t. It is not a promise; it’s the output of a model whose single most important input is margin expansion delivered alongside volume growth. If Leapmotor executes toward its targets, the re-rating and the earnings growth compound together and you get a number like that. If margins stay pinned near zero, you don’t. That sensitivity is the investment decision.

6. Risks — why the opportunity exists

A stock does not trade this cheaply by accident, so let me explain the price rather than explain it away. Leapmotor shares are down substantially over the past year — roughly a third or more from their highs — and there are legitimate reasons.

• Sector sentiment. The market has been badly burned by EV names and is pricing the whole cohort for a price war and margin compression. Leapmotor gets tarred with that brush regardless of its own trajectory.

• Margins are thin today. The entire thesis rests on future margin expansion, and the market is refusing to pay for a promise. Fair enough — that skepticism is exactly what creates the entry price.

• Political and tariff risk. EU tariffs on China-built EVs are the sword hanging over every one of these companies. The European production route mitigates it but doesn’t eliminate the headline risk, and trade policy can turn on a tweet.

• China-listing discount. Hong Kong-listed Chinese equities carry a persistent governance and geopolitical discount that has, frankly, been justified more often than not over the past decade.

• A founder-heritage wrinkle worth naming. Dahua, the founder’s prior company, has faced Western scrutiny and sanctions over surveillance uses. It’s a separate company, but it’s part of the honest risk picture around the founder’s background, and I won’t pretend it away.

The way I hold this in my head: the opportunity exists because of these risks, not in spite of them. You are being handed a 60%-grower at a single-digit cash-adjusted multiple precisely because the market is unwilling to underwrite Chinese EV margins and Chinese governance. My job is to size the position accordingly — this is a conviction idea, not a bet-the-fund idea.

7. Latest news and earnings

The most recent quarterly numbers showed the tension in the story in miniature: revenue growth continued and overseas sales hit records, but gross margin ticked down on product mix and scale effects, with management guiding to a gross margin in the low teens for the following quarter while reaffirming the full-year million-unit and CNY 5 billion profit targets. In other words — grow now, margins to follow. The next set of results is due around late August, and it’s the read on margin trajectory, more than the delivery headline, that I’ll be watching.

The strategic news flow, meanwhile, has been almost uniformly constructive. The Stellantis relationship is deepening rather than cooling: the partners are building cars together in Europe (with a Spanish plant secured for EU production), assembly has started in Malaysia, and there’s been open talk of extending the partnership’s reach further still. On the balance sheet, the AGM slate passed cleanly, preserving management’s flexibility. And on product, the European roll-out of the A10/B03X and the aggressive summer incentives in China tell you this is a company pressing its advantage while rivals retrench.

The single most important structural fact remains the Stellantis architecture. Leapmotor International — the vehicle for selling and building Leapmotors outside China — is a joint venture controlled by Stellantis at 51% to Leapmotor’s 49% (not the even 50/50 split that’s sometimes quoted). That control matters: it’s what gives a European giant the incentive to hand over plants, dealers and financing — and it’s what lets Leapmotor build inside Europe and step around the tariff wall.

8. Conclusion — why I own it

Pull it together and here’s the picture. A founder-led company with a genuinely differentiated cost structure and an imaging/software heritage, growing deliveries around 60%, freshly profitable, anchored by one of the world’s largest automakers, with a European production route that neutralises the biggest political risk to the story — trading at a mid-single-digit cash-adjusted multiple of the earnings it’s built to produce.

I keep coming back to one comparison. In China, Leapmotor’s volumes are now in the same conversation as Tesla’s. Tesla is worth something like two hundred times as much. Yes, Tesla is a different animal — energy, autonomy, a global brand, a software franchise. I’m not saying the gap should close to zero. I am saying that when you can buy a fast-growing Chinese manufacturer at roughly a tenth of the valuation of its US counterpart, and that valuation gap has been a decade in the making, you are being paid to wait for even a partial catch-up. I think the next ten years may see more of that catch-up than the last ten did.

And there’s an underappreciated optionality I want to leave you with. Everyone frames the EV race as a battery race — and batteries are exactly the part I’m least excited about: capital-hungry, increasingly commoditised (you can now buy excellent cells from CATL, Samsung and others quite cheaply), and unlikely to deliver another revolutionary leap absent a genuine solid-state breakthrough. Leapmotor’s centre of gravity sits somewhere more interesting — in software, chips and imaging, the layer I think grows relatively more important from here. A company that came up through embedded systems and built its own driving chip is, to my mind, better positioned for a software-defined-vehicle world than one whose crown jewel is chemistry. Add the possibility of a distinct European sub-brand, eventual buybacks or a larger dividend off that cash pile, and a deeper Stellantis tie-up, and you have several ways to win beyond the base case.

That’s why Leapmotor is the one EV maker in the Haas Invest4 Innovation Fund — and why, when my father-in-law asked whether he should buy one, I told him I’d been quietly thinking the same thing about the stock.

Risk disclaimer & conflict of interest

This article reflects my personal opinion and is intended for information and educational purposes only. It is expressly not investment advice, nor a recommendation or solicitation to buy or sell any security. Equities — and small-cap, foreign-listed, single-market emerging-market equities such as Leapmotor in particular — carry substantial risk, up to and including the total loss of capital. Currency risk (HKD/EUR), liquidity risk, political and tariff risk, and the governance risks inherent in Hong Kong-listed Chinese companies all apply here. Figures and prices cited are approximate, reflect the situation at the time of writing, and can change quickly; please verify current data yourself before acting.

Conflict of interest: Leapmotor (HKEX: 9863) is a position in the cost-efficient Haas Invest4 Innovation Fund (invest4.net), which I manage, and may also feature in my Wikifolio portfolios. I therefore have a financial interest in the security discussed. Please do your own research and, where appropriate, consult a licensed advisor before making any investment decision.

The story, and why I own it

Every once in a while I come across a company that most people in my audience have never heard of, run by a founder most people have never heard of either, and yet the ambition on the table is so large that if it works, it changes the world. Caris Life Sciences (NASDAQ: CAI) is one of those companies.

Let me start with the thing that matters most to me, because it frames everything else. If you ask almost anyone — investor or not, medtech specialist or complete layperson — what the holy grail of modern medicine is, most will land on the same answer: beating cancer. And within that, there are really two grails. One is the cure. The other, arguably even more valuable in terms of lives saved, is early detection. Because the brutal arithmetic of oncology is simple: the earlier you find it, the better the odds. A cancer caught at stage 1 is often a manageable problem. The same cancer caught at stage 4 is frequently a death sentence.

Caris is betting on the second grail — detection and precision — and it is doing it the way I actually want to see it done in the AI era. I have watched a lot of capital in the last two years chase “AI” companies that are, when you strip away the narrative, commodity providers. Compute is fungible. Foundation models are increasingly fungible. What is not fungible is a proprietary, decade-deep dataset of molecular information paired with the machine-learning models trained on it. That is a genuine moat, and it is the kind of AI business I would much rather own than yet another interchangeable infrastructure story bought at a stratospheric multiple.

So here is my one-paragraph investment case. Caris is a profitable-at-the-cash-flow-line, fast-growing molecular diagnostics platform, sitting on one of the richest clinical-genomics datasets in the world, run by a founder who has already built and sold a multibillion-dollar healthcare company and who has poured his own fortune into this one. The stock has fallen roughly 60% from its post-IPO high and now trades below its IPO price, which is precisely why the opportunity exists. My base-to-bull framework points to a fair P/E of around 30 and a possible return in the neighborhood of 35% per year — with meaningful additional optionality if their multi-cancer early-detection test becomes a standard of care. It is a small, deliberately speculative position in my Haas Invest4 Innovation fund, sized so that the downside is survivable and the upside can be a multiple. Let me walk you through why.

1. Product, business model, brand and moat

What they actually do. Caris describes itself as an AI “TechBio” company, which is marketing language, but the substance underneath is real. The core product is molecular profiling: you take a patient’s tumor tissue or blood, run comprehensive sequencing on it — whole exome and whole transcriptome, and in newer assays whole genome — and out the other end comes an incredibly detailed molecular portrait of that specific cancer. What is driving it, what mutations it carries, which approved or trial therapies it is most likely to respond to. In plain terms: they help oncologists figure out what kind of cancer this is and how best to treat this particular human.

The product family has expanded well beyond a single test. There is MI Profile, the tissue-based flagship. There is Caris Assure, the blood-based (”liquid biopsy”) platform. There is a whole-genome precision platform, a minimal-residual-disease test to monitor whether cancer is coming back, and Caris ChromoSeq for blood cancers, which recently cleared a key reimbursement hurdle. And then there is the crown jewel of the pipeline, Caris Detect, the multi-cancer early-detection blood test — the one aimed squarely at that early-detection grail.

How they earn money. The overwhelming majority of revenue — call it roughly 97% today — comes from clinical molecular profiling: labs and oncologists ordering tests, reimbursed by payers. The company completed around 52,800 clinical therapy-selection cases in a single recent quarter. The second, smaller leg is biopharma partnerships: pharmaceutical companies paying Caris to use its platform and data in drug development and companion-diagnostic work. That biopharma slice is small in absolute terms right now, but it is strategically the most interesting, because it is where the data gets monetized rather than just the test.

The brand. Within oncology, Caris has quietly become a serious name in comprehensive tissue profiling and therapy selection — the kind of reputation you only earn by being clinically credible for years. It is not a consumer brand, and that is a feature, not a bug (more on that below). Volume growth of roughly 15–22% in cases, quarter after quarter, tells me clinicians keep coming back.

The moat. This is the heart of my thesis. Caris has been collecting molecular data for well over a decade, linking genomic profiles to real clinical and outcome data at enormous scale. In an AI-driven world, that longitudinal, outcomes-linked dataset compounds in value every single year, and it cannot be replicated by simply buying more compute or copying a model. A new entrant can rent the same GPUs; it cannot rent thirteen years of proprietary, outcome-annotated cancer genomes. That is the difference between a commodity and a moat.

The contrast I keep coming back to is 23andMe. Remember the DNA-testing hype? The problem with the consumer model was that it was largely a one-and-done novelty purchase — you spit in a tube once, learn you are 12% Scandinavian, and never buy again. That company filed for bankruptcy in 2025 and was ultimately sold off for a fraction of its former valuation. Caris is the opposite model: it sells into the clinical and pharma system, where the same patient is profiled and monitored repeatedly, where payers reimburse, and where the data has direct medical and commercial value. Same raw ingredient — DNA and molecular data — completely different, far more durable business.

2. The market

The market is, frankly, enormous, and it is one of the rare cases where I think “gigantic total addressable market” is not hand-waving.

Comprehensive molecular profiling in oncology is already a large, growing market as it becomes standard of care for more and more cancer types. But the real prize is early detection. Think about the logic: if there were a reliable blood test that could catch many cancers early, from a single vial of blood, at an acceptable price and — crucially — with a low false-positive rate, who wouldn’t take it? Almost everyone would. That is not a niche; that is potentially a screening market spanning the adult population.

The false-positive point deserves emphasis, because it is where a lot of early-detection technologies quietly fail. If your test lights up ten times and nine of them are false alarms, you have not helped anyone — you have created enormous downstream cost, unnecessary follow-up procedures, and a lot of human anxiety. A test is only genuinely valuable if it combines real sensitivity with high specificity. That is exactly the bar Caris is trying to clear.

On the market’s character: healthcare diagnostics is relatively non-cyclical — people get cancer regardless of the business cycle, and cancer testing is not something patients defer because the S&P had a bad month. It is defensive demand in the best sense. The genuine external risk is not the economy; it is political and regulatory — specifically reimbursement. In the US, molecular diagnostics pricing runs through frameworks like PAMA and Medicare fee schedules, and payer coverage decisions can move revenue meaningfully. So the honest read is: the market is huge, structurally growing, and largely cycle-proof, but it is not free from government intervention. Reimbursement policy is the swing factor, and I keep a close eye on it.

3. Culture and management

This is the part that turned my head first, because I put enormous weight on founders who have both proven they can build and have their own money on the line.

The founder and CEO is David Dean Halbert. Back in 1987 he founded a pharmacy-benefit manager called AdvancePCS, took it public, grew it into a Fortune 250 company with well over $15 billion in annual revenue, and sold it to Caremark in 2004 in a deal valued at roughly $7.5 billion. So this is not a first-time founder hoping his story works out. This is someone who has already run the full cycle — build, scale, monetize — at multibillion-dollar scale, once.

What makes it more than a résumé is the why. Halbert has spoken openly about his mother’s death from cancer being the driving force behind Caris. He founded the company in 2008, funded it heavily with his own capital through the lean years, and remains a large shareholder — he became a billionaire on paper when the company went public. When a founder with that track record puts his own fortune behind a mission he cares about personally and stays in the chair to run it, the alignment between management and shareholders is about as good as it gets.

On strategy, the tell is long-termism. They are deliberately spending today — on the pipeline, on the early-detection launch, on sales-force expansion — while still holding the cash-flow line positive. That is a management team investing for a decade, not managing to a quarter. That is exactly the temperament I want behind a “big vision” company.

4. Financials, margins and recent developments

Here I want to be precise, because this is where I would gently correct some of the looser talk floating around about Caris, including a few of my own earlier back-of-the-envelope numbers.

Growth. It has been spectacular, and I want to be honest about why. Full-year 2025 revenue came in around $812 million, up roughly 97% over 2024’s ~$412 million. In the most recent quarter, revenue was about $216 million, up roughly 79% year over year, with the molecular-profiling engine up about 85%. But a large chunk of that eye-popping growth was not volume — it was average selling price (ASP). Clinical ASP jumped materially as reimbursement and collections improved, with tissue ASP up around 70%. Case volume grew a more sober 15% or so in the quarter. That distinction matters enormously for valuation, and I will come back to it in the risks section. Management guides to roughly $1.0–1.02 billion of revenue in 2026, implying a much calmer ~23–26% growth rate — which, notably, is right in the range I would want to underwrite going forward anyway.

Profitability and margins — read this carefully. Gross margin has expanded impressively, to around 65% from the high-40s a year earlier. The company is now generating positive adjusted EBITDA (roughly $26 million in the recent quarter, its fourth straight positive quarter) and positive free cash flow (around $22–23 million in the quarter, even after paying out annual bonuses). Operating income turned positive, and the GAAP net loss narrowed to essentially breakeven.

But — and this is the correction I owe you — on a full-year GAAP basis Caris is not yet net-income profitable. 2025 carried a large reported net loss, inflated by IPO-related and non-cash items, and the trailing net-income line is still deeply negative. So the accurate statement is not “already profitable” in the plain-vanilla sense; it is “already generating cash, adjusted-EBITDA positive, and standing right at the doorstep of GAAP breakeven.” That is a genuinely important milestone for a company at this growth rate — it means they are largely funding their own ambition rather than burning shareholders’ money — but it is not the same as a mature earnings stream, and I will not pretend it is.

Balance sheet. Strong, and stronger than I had assumed. The company ended the recent quarter with over $825 million in cash, equivalents and marketable securities, against a $400 million term loan (refinanced this year on better terms). So net cash is comfortably positive. This is the single biggest reason the financial probability of failure here is, in my view, low: with that liquidity plus positive cash generation plus a billionaire founder-owner, this company is not going to be forced into a bad corner.

5. Valuation

Valuing Caris on a trailing P/E is meaningless today — GAAP earnings are hovering around zero, so the ratio is not defined in any useful way. That is exactly the kind of situation where lazy screens spit out “not profitable, avoid,” and where patient investors occasionally get paid. So I do two things.

First, a sanity check on sales. Netting cash against debt, enterprise value sits around $4.5–4.8 billion. On 2026 guided revenue of ~$1.0 billion, that is roughly 4.5–4.7x forward EV/Sales. For a business growing 25%+ with 65% gross margins, a compounding data moat, and a genuine shot at a screening-scale product, sub-5x forward sales is not a demanding multiple. It is not statistically “cheap” the way a bank at 6x earnings is cheap, but it is far from the crazy multiples I have watched other AI names carry.

Second, my fair-P/E framework, which is how I actually anchor the decision. The way I look at a company like this is to ask what its normalized earnings power looks like a few years out, once the current investment phase converts into profit, and what multiple that earnings stream deserves. Given 25%+ revenue growth, high and expanding gross margins, the defensive character of diagnostics, and the optionality on top, I assign Caris a fair P/E of 30. That is a premium multiple, but a defensible one for a durable, high-margin, secular grower with a moat.

Run that forward. If Caris compounds revenue at roughly a quarter per year and scales its margins toward the levels a dominant molecular-diagnostics platform should eventually earn — which the 65% gross margin makes entirely plausible — and I apply a fair P/E of 30 to that future earnings power, then against today’s depressed share price in the mid-teens, my base-to-bull case points to an annualized return on the order of 35% per year over my holding horizon. I want to be transparent that this outcome leans on two assumptions doing real work: continued 25%-ish top-line growth, and meaningful net-margin expansion as the platform scales. Neither is guaranteed. But neither is heroic for a business already generating cash at this stage.

And then there is the part the model cannot really capture: early detection. My valuation deliberately does not bake in Caris Detect becoming a standard-of-care screening test, because that outcome is binary and uncertain. If it happens, the numbers above are far too conservative. That is the free option I am buying.

6. Risks — and why the opportunity exists at all

I never want to present only the bull case, so let me explain the elephant in the room: the stock has been cut roughly in half over the past year and sits about 60% below its 52-week high, trading below its June 2025 IPO price of $21, down in the mid-teens. If the business is so good, why is the chart so ugly? There are several honest reasons, and each is also a reason the opportunity exists.

IPO mechanics. Caris only went public in June 2025. Newly public stocks are volatile in both directions, and the post-IPO lockup expiration late in 2025 unleashed insider and early-investor selling — supply hitting the market for reasons that have nothing to do with the underlying business. Add thin trading and momentum-driven selling into a widely watched earnings date, and you get exaggerated moves.

Valuation de-rating. The stock ran up to $42.50 shortly after listing on pure enthusiasm. A lot of the subsequent decline is simply air coming out of an over-inflated initial price, not a deterioration in fundamentals.

The ASP question — the serious one. This is the risk I take most seriously. Because so much of the recent revenue growth came from ASP and favorable reimbursement rather than volume, the bear case is straightforward: those year-over-year comparisons get much harder as the company laps the reimbursement step-up, and if pricing normalizes, headline growth could decelerate sharply. The market may be pricing in that deceleration before it shows up in the reported numbers. This is a legitimate concern, and it is why the guided ~25% growth — not the reported ~79% — is the number I underwrite.

Still-heavy GAAP losses. As covered above, the trailing net loss is large, and for investors who screen on GAAP profitability, that is a red flag and a source of ongoing skepticism about the path to real earnings.

Reimbursement and regulation. The government-policy swing factor from the market section is a live, permanent risk, not a hypothetical.

So the opportunity, as I see it, is a classic mismatch: technical and sentiment-driven selling pressure (lockups, momentum, an over-eager IPO price) stacked on top of a real-but-manageable fundamental question (ASP durability), against a business that is compounding volume, generating cash, and sitting on a strategic dataset. Notably, management authorized a share buyback near these lows — a company that was capital-constrained right up to its IPO now choosing to retire its own stock is, to me, a meaningful signal about how insiders view the price. And I will admit I take some quiet comfort from the fact that at least one legendary investor has been building a position here too; I am not the only one who thinks the market has overshot.

7. Latest news and earnings

The most recent reported quarter was strong on almost every operational metric: ~$216 million in revenue (+79%), gross margin up to 65%, positive adjusted EBITDA for the fourth consecutive quarter, positive free cash flow, and a GAAP net loss down to essentially nothing. Management reaffirmed full-year 2026 guidance of roughly $1.0–1.02 billion. On the earnings call, the leadership spent real time addressing the reimbursement and ASP framework directly — a sign they know exactly where the market’s anxiety sits and are trying to get ahead of it rather than paper over it.

On the pipeline, the news flow has been genuinely busy: Caris ChromoSeq secured MolDX reimbursement approval, opening blood cancers as a new clinical market beyond solid tumors; the company launched MI Clarity, an AI-driven breast-cancer recurrence-risk tool; and it brought in-house PTEN IHC testing. It also refinanced its debt on better terms and set up a share-repurchase program.

The headline pipeline event, though, is Caris Detect, the multi-cancer early-detection blood test, which moved to commercial launch in mid-2026 through a consumer-health distribution partnership. The most-watched data point so far showed roughly 60% sensitivity for early-stage (stage 1–2) cancers at about 99% specificity in a high-risk cohort. I want to be measured here: that is an encouraging, high-specificity result from a case-control study — exactly the low-false-positive profile the market needs — but it is not yet a completed large prospective trial, and it is not in guidance. It is optionality, not a certainty. The next scheduled catalyst is the Q2 2026 earnings report in early August, which is precisely the collision point between the bull and bear views on ASP durability. I will be watching the volume line and the ASP commentary far more closely than the headline growth number.

8. Conclusion

Let me bring it back to why I own this, in plain terms.

There are only a handful of what I would call genuine world-class businesses attacking one of the largest, most defensible, most meaningful problems in the entire market — the early detection and precision treatment of cancer. Caris is one of them, and right now the market is handing it to me at below its IPO price, at under 5x forward sales, because of lockup selling, an over-heated initial valuation, and a real-but-quantifiable question about pricing durability.

Against that, I am buying: a compounding proprietary dataset that gets more valuable every year and cannot be bought with compute; a business already generating positive free cash flow with 65% gross margins on the doorstep of GAAP profitability; a fortress balance sheet with net cash; and a founder who has done this before at multibillion-dollar scale and has his own money and his mother’s memory riding on the outcome. My framework, at a fair P/E of 30, points to something like a 35% annualized return if they simply execute the base plan — and a genuine multiple if Caris Detect turns early cancer screening into a routine blood test.

That last scenario is the one that gets me out of bed. A single vial of blood that reliably catches cancer early, across many cancer types, at scale — that is not just a good investment case, it is one of the few things in medicine that could change survival statistics for millions of people. The base case pays me for owning a high-quality, growing, cash-generative diagnostics leader. The bull case is a genuine gamechanger. The downside, thanks to the profitability, the cash and the founder, I judge to be contained. That asymmetry — limited downside, capped by real cash flows; enormous, world-changing upside — is exactly the shape of bet I want in the innovation sleeve of a portfolio.

It is a big nail they are trying to hammer. But I would back this founder to swing at it. That is why Caris is a small, speculative healthcare position for me — sized for the uncertainty, held for the vision.

Risk disclaimer

This article reflects my personal opinion and analysis and is expressly not investment advice, nor a recommendation to buy or sell any security. All figures are approximate and as of late July 2026; market data, prices and company fundamentals change constantly, and you should verify current figures yourself before making any decision. Caris Life Sciences is a small, deliberately speculative position and carries above-average risk, including the risk of significant or total loss — the company is not yet consistently profitable on a GAAP basis, its recent growth has been heavily influenced by reimbursement and pricing dynamics that may not persist, it operates in a market exposed to regulatory and reimbursement changes, and its most exciting product remains clinically unproven at scale. Forward-looking statements, including any return or fair-value estimates, are assumptions that may prove wrong. Please do your own research and, where appropriate, consult a licensed financial advisor.

Disclosure: Caris Life Sciences is held as a position in the cost-efficient Haas Invest4 Innovation investment fund (invest4.net), which I manage, and may also be held in my Wikifolios. I therefore have a financial interest in the security discussed.

Roughly a decade ago I published an analysis of Netflix. The company was worth about $40bn, it was pushing past 70 million members, Reed Hastings was still running it, and it was earning barely $120m in net income while burning cash on content. I applied a fair P/E of 32, looked at the price, and concluded the stock was too expensive. I kept only a small residual position in my founder-led Wikifolio and said I did not expect significant further gains.

Today I am doing something I have not done in a long time: I am putting a US large cap back into the Haas Invest4 Innovation Fund. And it is the same name. Netflix. Only this time the arithmetic is inverted. The company is now worth about $292bn, generates roughly $51bn of revenue, guides to a 31.5% operating margin, throws off around $12.5bn of free cash flow — and trades at under 20x this year’s consensus earnings. Ten years ago I demanded a fair P/E of 32 for a company that was not making money. Today I apply a fair P/E of 26 to a company that has an ROE above 40% and a global monopoly on premium on-demand entertainment.

The investment case in three sentences. Netflix is the most scalable content business ever built: you finance a Korean thriller once and monetise it in 190 countries forever, and generative AI is now measurably lowering the cost of that one-time production. The stock has fallen roughly 45% from its high and de-rated to under 20x earnings because of a bruising, ultimately abandoned Warner Bros. bidding war, decelerating viewing-hour growth, and a management decision to disclose less, not because earnings power broke. On my fair P/E of 26 applied to a 2029 EPS estimate of about $5.15, I get a target of roughly $134 and an expected return of 24% per annum — which for a US large-cap tech franchise of this quality is, frankly, unusual.

This is my analysis, not investment advice.

1. Product, Business Model, Brand and Moat

What the company actually does

Netflix sells one thing: the right to watch, on demand, whatever you want, whenever you want, on whatever screen you happen to be holding. That is it. There is no cloud division, no advertising conglomerate, no theme park. In an era where every other mega-cap has become a holding company of unrelated bets, the simplicity is a feature.

The catalogue spans licensed film and television, an enormous and growing slate of own-produced originals in dozens of languages, live events, video podcasts, mobile games and — increasingly — vertical short clips. Profiles, algorithmic recommendation, cross-device resume and household sharing are the product mechanics that were novel when I first wrote about the company and are now simply the industry standard, because Netflix set it.

How they earn money

Two engines, one of which is still small and about to matter a great deal.

Subscriptions. In the US the ad-supported tier now costs $8.99 a month, Standard $19.99 and Premium $26.99, after an across-the-board increase in March 2026 — the second hike in just over a year, averaging around 11% across the product suite. Note the drift: when I first covered the stock, the subscription cost €7–9 and I described it as cheap relative to pay TV. It is now three times that at the top tier, and churn remains the lowest in the industry. That is what pricing power looks like when you measure it over a decade rather than a quarter.

Advertising. This is the part the market is still mispricing. Netflix expects ad revenue to roughly double to about $3bn in 2026, has more than 4,000 advertisers on the platform, and in markets where the ad tier is available more than 60% of new sign-ups choose it. Management has been explicit that there is a gap between average revenue per member on the ad tier and on the standard tier, and that closing that gap is the objective. A $3bn ad business inside a $51bn revenue company is a rounding error today; at $10bn it is a re-rating.

Brand

Netflix is one of a handful of brands that became a verb. It is the default noun for “watching something at home” across most of the developed and much of the developing world. Roughly 330 million paying households — 325 million was the last milestone the company confirmed, in January 2026 — sit behind that brand. It is not growing at the rate it did in 2015, but it is not eroding either, and it carries pricing power that Disney+, Peacock and Paramount+ have repeatedly failed to demonstrate.

The moat — and how it has changed

When I first wrote about Netflix, I flagged a genuine weakness: dependence on the studios. Netflix had to buy the good films, the studios knew it, and the studios had every incentive not to let Netflix become a monopolist that dictated terms.

That risk has been almost entirely engineered away. Netflix is now itself one of the largest content producers on earth, spending north of $20bn a year — up about 10% in 2026, ahead of the 8% average of the last five years but below the 14% average of the last decade. The dependency has reversed: legacy studios now need Netflix’s distribution more than Netflix needs their libraries.

The moat today rests on four pillars:

1. Scale economics on content. Fixed cost, near-zero marginal distribution cost, 190 countries. A hit produced in Seoul or Madrid amortises across the entire global base. No competitor has that denominator.

2. The data feedback loop. More members → more revenue → more content spend → better hit rate → more members. I described this cycle a decade ago as a strategy; it is now an established fact of the industry’s structure.

3. Recommendation and retention. Industry-leading churn is the single most under-discussed asset on the balance sheet — except it is not on the balance sheet at all.

4. Emerging: a proprietary production-technology stack. More on this below, because it is the newest and least-priced part of the moat.

How replaceable is Netflix? A competitor could copy the interface in a quarter. Replicating a $20bn annual content budget financed by 330 million subscribers, with the retention profile to sustain it, would require somebody to spend a decade and roughly $200bn. Paramount and Warner Bros. are about to attempt something adjacent to that, with $90bn of debt attached. I would rather own the incumbent.

2. The Market

Size and growth

The global market Netflix competes in is not “streaming.” It is human attention allocated to video entertainment — a market measured in the trillions of hours annually and several hundred billion dollars of consumer and advertising spend. Netflix’s ~$51bn of revenue is a low single-digit share of that. Linear television is still being dismantled in most of the world, and each percentage point of viewing time that migrates is incremental addressable revenue.

Two structural growth vectors remain intact. First, geography: penetration in Asia and Latin America is a fraction of what it is in the US and UK, and both regions grew revenue at double digits in the latest quarter, with LATAM at 21%. Second, monetisation: the ad tier converts the vast pool of price-sensitive households from non-customers into low-ARPU customers who can be re-priced upward over time.

Political interference

Moderate but real, and rising. The European Union imposes local-content quotas and investment obligations. Several countries levy digital services taxes on streaming revenue. Content is inherently political, and the last two years have brought recurring political noise in the US around foreign film production and around media ownership concentration. The Paramount–Warner Bros. Discovery merger is itself working its way through regulatory and litigation processes. None of this is existential for Netflix — it is a cost of doing business — but it is a permanent friction, and it argues against paying a premium multiple.

Cyclicality

Low, and this is a key part of why I am comfortable at large-cap weight. A $9–20 monthly subscription is one of the last discretionary items a household cancels; Netflix demonstrated this through 2022 and 2023. The advertising leg is genuinely cyclical and will move with the ad market — but at $3bn of a $51bn revenue base, that cyclicality is currently well contained. If ads scale to $10bn, the cyclicality of the business rises, and that is worth watching.

3. Culture and Management

Reed Hastings no longer runs Netflix, and as of June 2026 he is no longer even on the board. He stepped back from co-CEO in January 2023, served as executive chairman, and announced in April 2026 that he would not stand for re-election. Ted Sarandos and Greg Peters have been co-CEOs since 2023 — Sarandos on content, marketing and legal, Peters on product, technology, advertising, finance and games.

I want to be explicit about this, because in my earlier analysis the founder-led argument was the entire reason I held a residual position in the Wikifolio for founder-led companies. That argument no longer applies. Netflix is now a professionally managed large cap. Investors buying it as a founder-led business are buying a stock that no longer exists.

What replaces the founder argument?

First, the culture actually survived the founder. Hastings’ contribution — and he said as much on the way out — was less any single decision than a high-performance culture of freedom and responsibility that others could inherit. The co-CEO structure has now run for three and a half years through a password-sharing crackdown, an advertising build-out from zero, a move into live events, and an $83bn attempted acquisition. That is a stress-tested management team, not a caretaker one.

Second, the capital allocation has been genuinely entrepreneurial — including when it meant walking away. Netflix agreed to buy Warner Bros. and HBO Max for nearly $83bn in December 2025. When Paramount Skydance escalated to roughly $111bn for the whole of WBD, Netflix had four business days to counter. It declined, calling the transaction no longer financially attractive, and characterised the asset as something it would have liked at the right price rather than needed at any price. It collected a $2.8bn termination fee for the trouble.

I cannot overstate how much I like this. The easiest thing in the world for a management team with a bruised ego and a falling share price is to win the auction. Sarandos and Peters let the trophy asset go, took the cash, and immediately deployed $4.7bn into buying back their own stock in Q2 — the largest buyback quarter in the company’s history, with $27.1bn of authorisation remaining. Capital discipline under public pressure is the rarest quality in large-cap management, and Netflix just demonstrated it in the most visible way possible.

Third, the founder is voting with his own money. Hastings reportedly bought 794,250 shares in May and June 2026, at depressed prices, after announcing his board exit. He no longer has a seat, a title or an obligation. He bought anyway.

Strategic time horizon: long. Content spend is being raised into a falling share price. The AI production stack was built by acquisition and internal labs rather than rented. Ads were launched knowing they would dilute reported ARPU before they accreted to it. None of this is quarter-management.

4. Financials, Margins and Recent Developments

Growth

2026 net income and EPS include the $2.8bn Warner Bros. termination fee. Excluding it, underlying 2026 EPS is roughly $3.00.

Note the 10-for-1 stock split effected in November 2025 — all per-share figures here are post-split, which is why a stock that was quoted in four figures is now quoted at $70.

The pattern is exactly what I described in the German note that prompted this piece: still double-digit growth, with margins expanding. Revenue has compounded at roughly 15% over three years and is guided to about 13% this year, decelerating gently to around 11% next year. Operating margin has gone from 20.6% to a guided 31.5% in three years — nearly eleven points of expansion. That is operating leverage on a scalable content base doing precisely what the model says it should.

Margins and returns

• Operating margin: 31.5% guided for 2026; consensus has it reaching roughly 36% by 2028.

• Gross margin: ~50.8% expected in 2026, up from 46.1% in 2024.

• Net margin: 24.3% in 2025; roughly 25% underlying in 2026 (about 30% reported, inflated by the termination fee).

• Return on equity: 41.3% in 2025, up from 26.3% in 2023 and 35.2% in 2024. Rising equity base and rising ROE simultaneously is the signature of a genuinely high-return business.

• ROIC: around 29–30%, top quartile of the sector.

• Free cash flow: roughly $12.5bn guided for 2026, on a ~$292bn market cap — a free cash flow yield of about 4.3%.

A word on EBITDA, since I am usually asked. For Netflix, EBITDA is close to meaningless. Content amortisation runs at roughly $17bn a year and is not a non-cash accounting artefact you can wave away — it is the depreciation of the company’s actual product. Add it back and you get an “EBITDA margin” north of 65%, which flatters the business absurdly. I use operating margin and free cash flow. Anyone quoting you an EV/EBITDA multiple on Netflix is quoting you a number with no economic content.

Recent developments

Q2 2026, reported 16 July, was operationally fine and received terribly:

• Revenue $12.56bn, +13% year on year (+12% FX-neutral), a rounding error below consensus. Every region grew double digits — 10% in US/Canada, 21% in Latin America, with EMEA passing $4.0bn and both LATAM and APAC passing $1.5bn in a quarter for the first time.

• Operating income $4.19bn, margin 33.4% versus 34.1% a year earlier — ahead of the company’s own guidance, with the year-on-year dip explained by content amortisation being front-loaded into the first half.

• Net income $3.40bn, diluted EPS $0.80 versus $0.72, a penny ahead of consensus.

• Free cash flow $1.53bn in the quarter, down about a third, driven by cash tax timing and the Warner Bros. termination — full-year FCF guidance unchanged at roughly $12.5bn.

• Full-year revenue guidance narrowed to $51.0–51.4bn; operating margin target reaffirmed at 31.5%.

• Q3 guidance light on both lines: revenue $12.86bn against consensus near $13.0bn, EPS $0.82 against $0.84.

• Record $4.7bn of buybacks in the quarter.

• Viewing hours: more than 97 billion in H1 2026, up 2%, an incremental 1.5 billion hours — a slight acceleration from 1.5% growth in 2025.

• The company will publish its engagement report annually from 2027 rather than alongside quarterly earnings.

The stock fell as much as 9% after hours and has since ground to new 52-week lows. Two misses landing in the same release — revenue and EPS guidance — plus a reduction in disclosure, is about as bad a combination as a market can be handed in a single evening.

5. Valuation: A Fair P/E of 26 and 24% Per Annum

Here is how I get there.

Where the multiple is today. At around $70 per share and consensus 2026 EPS of $3.56, Netflix trades at 19.7x current-year earnings. On my underlying 2026 estimate of roughly $3.00 — stripping out the Paramount termination fee, which is emphatically not recurring — it is about 23x. On 2027 consensus EPS of $3.82 it is 18.3x.

For context, the forward multiple has run at roughly 52x (2021), 32x (2022), 32x (2023), 39x (2024) and 31x (2025). Today: under 20x.

The earnings bridge to 2029.

The assumptions are deliberately unheroic: revenue compounding at roughly 11%, driven by pricing, modest membership growth and advertising scaling toward $8–10bn; operating margin reaching 36% by 2029, which is where consensus already sits for 2028; and a share count declining around 2.5% a year given $27bn of remaining buyback authorisation deployed into a depressed stock.

Applying the fair P/E.

Fair P/E: 26 × 2029e EPS $5.15 = fair value ~$134 From ~$70: +91% total, or ~24% per annum over three years

Why 26 and not 32, as I applied a decade ago? Because the business is now mature, professionally rather than founder-managed, growing at 11% rather than 30%, and operating in a market where the incumbent competition just consolidated. And why not 20, as several bears would argue? Because a 40%+ ROE, a genuine global moat, sub-industry churn, a barely-monetised advertising asset and a structurally scalable cost base are worth a premium to the market. 26 is where I land, and it is a multiple Netflix has traded above in every single year of the last decade.

For a US large-cap technology franchise, 24% per annum is an unusual return profile. I would normally have to go to a Japanese small cap or a Central Asian fintech to find it. That is precisely why it is interesting.

6. Risks — and Why the Opportunity Exists at All

If a business this good is available at under 20x earnings, something must be wrong. Here is my honest list.

The engagement problem is real. Viewing hours grew 2% in the first half of 2026 while revenue grew 13%. Read that again. Almost all of Netflix’s growth is now coming from price and advertising, not from people watching more. Management argues that engagement should be assessed on quality, variety and quantity together, and that live events punch far above their share of viewing hours in sign-ups and monetisation. That is a defensible argument. It is also exactly the argument a company makes when the simple number is not cooperating.

The disclosure problem compounds the engagement problem. Netflix stopped reporting subscriber numbers after Q1 2025. It will now report engagement annually rather than quarterly, on the stated logic of keeping focus on revenue and operating profit. I understand the reasoning and I think it is strategically sound. I also think the market is entitled to punish it. When a company reduces disclosure precisely as the metric in question decelerates, a multiple discount is not irrational — it is the correct price for reduced information.

Short-form video is eating attention. The bear case articulated most sharply on the sell side is that TikTok, Instagram, YouTube Shorts and Snap are doing to streaming what streaming did to linear television — particularly among younger viewers. Netflix’s response is vertical clips, video podcasts and creator partnerships. Whether a company built on 50-minute prestige drama can win a format war it did not choose is an open question, and I do not pretend to know the answer.

The M&A overhang has not cleared. The Warner Bros. saga cost management months of attention and left investors uncertain about strategy. Comcast is now spinning off NBCUniversal and Sky; Fox has bought Roku; Lionsgate is effectively for sale; the Paramount–WBD combination is still working through regulatory approval. Netflix says it prefers organic growth. Every time a large asset comes loose, the market prices in the risk that Netflix changes its mind. A disciplined acquirer is an asset; an undisciplined one destroys 20% of the equity in an afternoon.

Content amortisation and the FCF optics. Free cash flow fell about a third in Q2 on tax timing and the terminated deal. The full-year guide is unchanged, but content amortisation front-loading makes the reported first half look worse than the business is.

Currency, taxes and regulation. FX-neutral growth was a point below reported growth this quarter, which happened to help. It will not always. Digital services taxes, EU content quotas and media-ownership politics are permanent frictions.

And the departure of the founder. I do not think it is a business risk. I think it is a sentiment risk, and it has been priced as one.

So why does the opportunity exist? Because all of the above are second-derivative concerns — the rate of change of engagement, the quality of disclosure, the risk of a future acquisition — while the first derivatives are unambiguously good: 13% revenue growth, 31.5% operating margin, 41% ROE, $12.5bn free cash flow, record buybacks. Markets de-rate on narrative and re-rate on numbers. Netflix has lost the narrative and kept the numbers. That is my favourite kind of setup, and it is a nearly exact structural rhyme with what I look for in the small caps I usually write about — with the difference that here the balance sheet risk is negligible.

7. Latest News and Earnings

Beyond the Q2 print itself, five developments matter for the thesis:

1. Generative AI moved from experiment to infrastructure. Netflix disclosed that generative AI workflows were used in roughly 300 titles in 2026 — up from essentially one show in mid-2025 — concentrated in post-production but extending into concept development and previsualisation. The concrete data point: a documentary series contained around 17 minutes of AI-enhanced footage that Sarandos said was produced twice as fast and at roughly half the cost of the conventional alternative, enabling sequences the production could not otherwise have afforded. The stack is built on three pillars: InterPositive, the AI production-tools company co-founded by Ben Affleck that Netflix acquired in March 2026 for up to $600m; Eyeline, its visual effects research group; and an internal animation lab. Peters described the three units as now working in concert on production speed. Sarandos has been careful to frame AI as a tool for creators rather than a replacement for them — sensible positioning after 2023, and also, I suspect, sincere.

2. The $2.8bn cheque cleared. Paramount Skydance paid the termination fee. Netflix ended a bruising eight-month M&A process with cash, no debt, no integration risk and no regulatory review — and immediately turned the proceeds into stock.

3. Record capital return. $4.7bn of buybacks in Q2, the largest quarter ever, with $27.1bn of authorisation remaining after the board added $25bn in April. At current prices this is high-return capital allocation, not financial engineering.

4. Balance sheet housekeeping. On 22 July Netflix issued $1bn of 5.25% senior notes due 2036, refinancing debt maturing later this year. Investment-grade, unremarkable, and worth noting only because a company under this much narrative pressure raising ten-year money at that coupon is a company the credit market is entirely relaxed about.

5. The consolidation wave is creating sellers. Comcast’s NBCUniversal/Sky spin-off, the Fox–Roku deal and Lionsgate’s availability mean that for the first time in years, premium studio and library assets are genuinely in play. Analysts are broadly sceptical that Netflix pursues NBCU — the regulatory and structural obstacles are formidable, and management has said organic growth is the preference. But the option exists, it is free, and Netflix has just proven it will not overpay to exercise it.

Sell-side price targets came down after the print — Baird to $90, Bernstein to $95, Pivotal all the way to $70 — but the distribution remains constructive: 51 analysts, consensus “Buy”, average target $95.28, and not a single sell rating. The most notable feature of the current tape is the divergence between prediction-market pessimism and analyst constructiveness, which usually resolves in favour of whoever is looking at the cash flows.

8. Conclusion: Why I Am Buying

I want to be clear about what I am not arguing. I am not arguing that Netflix is about to reaccelerate to 20% growth. I am not arguing that the engagement concerns are fabricated. I am not arguing that the market is stupid.

I am arguing that a business with a 41% return on equity, 13% revenue growth, eleven points of operating margin expansion in three years, the lowest churn in its industry, a barely-monetised advertising asset and $12.5bn of annual free cash flow should not trade at under 20x earnings — and that when it does, the reason is almost always narrative rather than arithmetic.

Four things could take this from “cheap” to “materially undervalued”:

Lower production costs through AI. This is the newest and, in my view, the most underappreciated line in the entire thesis. Netflix spends over $20bn a year on content. If generative AI and the InterPositive stack take even 5–10% out of that cost base over five years while holding quality constant, that is $1–2bn dropping toward operating income annually — on a business currently earning $16bn. Nobody’s model has this in it, because Netflix has only just started quantifying it. This is exactly the kind of structural margin lever that shows up as “unexplained” operating leverage two years from now.

Scalability. The oldest argument in the file and still the best one. Produce once, distribute to 330 million households in 190 countries, forever. The marginal cost of the 331 millionth subscriber is close to zero. No competitor has the denominator to match Netflix’s content budget per subscriber, and the gap widens every year.

Buying content studios — cheaply, or not at all. The consolidation wave has created motivated sellers and a rival carrying roughly $90bn of debt. Netflix has $2.8bn of found money, an investment-grade balance sheet, and a management team that just publicly demonstrated it will walk away from an $83bn deal on price. That combination — the capacity to buy plus the discipline not to — is worth something, and the market is currently pricing it as a liability.

Advertising. $3bn in 2026 heading toward a multiple of that, inside a company where 60%+ of new sign-ups in ad-enabled markets pick the ad tier. This is a second business being built inside the first, at scale, with 4,000 advertisers already on the platform.

Netflix is the first US large cap to enter the Haas Invest4 Innovation Fund in a long time, and I have built an initial position rather than a full one. I want to see engagement stabilise and the Q3 print land before adding. But at under 20x earnings, with a fair P/E of 26 and an implied 24% annual return, this is the kind of asymmetry that does not usually attach itself to a $290bn company — and when it does, it does not last.

Ten years ago I got Netflix wrong by insisting on a premium multiple for a business that had not yet earned it. I would rather not now get it wrong in the opposite direction, by refusing to pay a fair multiple for a business that has.

Risk Disclaimer and Conflict of Interest

This article is not investment advice, not a recommendation to buy or sell, and not a personalised financial analysis. It represents my personal opinion and assessment as of 24 July 2026, based on publicly available information, company filings, the Q2 2026 shareholder letter and earnings call, and market data as of the date of publication.

Equity investments carry substantial risk, including the risk of total loss of capital invested. Share prices can fall as well as rise, and past performance is not indicative of future results. The projections, fair P/E multiples, earnings estimates and expected returns presented here are model-based assumptions that will not materialise exactly as described and may prove entirely wrong. Netflix is exposed to competitive, technological, regulatory, currency and execution risks as outlined in Section 6 of this article, and its shares have already declined roughly 45% from their 52-week high, demonstrating the volatility involved.