Alibaba Group

IThe business

Alibaba runs the biggest online marketplaces in China and rents out cloud computing. In the twelve months to June 2026 it had revenue of RMB 1,045.0B and adjusted EBITA of RMB 64.9B, a margin of 6.2% C (calculated).⁵⁶⁷ In FY2025 it was 17.4%. The fiscal year ends on 31 March, so TTM on this page means the four quarters to 30 June 2026.

The company has changed its shape in a year. In the June 2026 quarter it merged its China and international commerce units with Freshippo into one Alibaba E-commerce Group, merged its cloud unit with its chip designer T-Head into AI Cloud and Compute Services, and pulled its model labs, the Qwen consumer app and the QwenWork agent into a new segment, AI Labs and Applications.⁶ The numbers below use that new layout where the company has recast it, and the old one where it has not.

The June quarter shows both halves. Revenue rose 9% to RMB 269.0B. Cloud grew 45%. The commerce core earned RMB 39.7B of adjusted EBITA, about what it earned a year earlier. But group adjusted EBITA fell 30% to RMB 27.3B, free cash flow was an outflow of RMB 44.7B, and capital spending was RMB 67.7B, up 75%.⁶

What it sells

• E-commerce Group. Taobao, Tmall, AliExpress, Alibaba.com, Freshippo and quick commerce. Customer management revenue, the fees merchants pay to advertise and sell on Taobao and Tmall, was RMB 82.5B in the June quarter, 31% of group revenue C (calculated). It fell 7%. The company changed how it books some merchant subsidies: they now reduce revenue instead of sitting in marketing costs. Without that change, it says, the line would have grown 1%.⁶ China quick commerce, delivery of food and other goods within the hour, grew 45% to RMB 53.3B.⁶
• AI Cloud and Compute Services. Revenue RMB 48.4B, up 45%, and the same growth for outside customers. Products tied to AI made RMB 12.4B of that, 26% C (calculated), and have grown by triple digits for twelve quarters in a row.⁶ Segment adjusted EBITA was RMB 5.6B, an 11.6% margin C (calculated), up from RMB 2.4B.⁶
• AI Labs and Applications. The Qwen models, the Qwen app and QwenWork. Revenue was RMB 3.3B and the adjusted EBITA loss was RMB 13.9B, four times the loss of a year earlier.⁶ The company says higher inference costs for the Qwen app drove the loss; the Qwen app also carries the user acquisition spending that raised sales and marketing costs in FY2026.⁶⁷
• All others. Alibaba Health, the media and entertainment group, Amap, games and other technology units. Revenue RMB 28.8B and a loss of RMB 3.3B in the quarter.⁶

Where the profit went

FY2026 is the year the profit moved. Adjusted EBITA fell 56% to RMB 76.4B, from RMB 173.1B. China e-commerce profit dropped 44%, from RMB 193.2B to RMB 107.5B, as Alibaba put money into quick commerce, user experience and technology. The losses in the units outside commerce and cloud went from RMB 9.5B to RMB 35.7B.⁷ Operating income under US GAAP fell 64% to RMB 50.2B.⁵ Net income to ordinary shareholders was RMB 105.9B, but it included a gain of RMB 66.1B on equity investments and RMB 8.4B on disposals, neither of which is operating profit.⁵ Non-GAAP net income, which strips those, was RMB 60.7B, down 62%.⁷

Alibaba also sold two big shops. Sun Art, the hypermarket chain, and Intime, the department stores, were sold in FY2025; the company says revenue grew 11% in FY2026 once their sales are taken out of the earlier year, against the reported 3%.⁷ I count the reported figures in the charts and use the lower growth rate when I talk about trend.

The things around the core

Alibaba owns 33% of Ant Group, the payments and finance company behind Alipay, on a fully diluted basis. It does not control it and records its share of profit one quarter late: RMB 1.56B in the June quarter.⁸⁶ It holds listed shares worth RMB 100.6B and private stakes carried at RMB 130.4B, as of 31 March 2026, plus RMB 203.2B in equity-method investments at 30 June.⁸⁶ On 27 April 2026 it announced a plan to move the Jiaxing data campus into a listed real estate fund in Shenzhen, which would take the buildings off its balance sheet while it keeps using them.¹⁶

The commerce business pays for everything. The page below is about what the money is being turned into, and about whether the owners get any of it back.

IIThe moat

I think Alibaba has two protections of different quality. The marketplace is a wide one that has stopped widening: merchants and shoppers are already there, and the fees they pay still add up to RMB 82.5B a quarter.⁶ The cloud is a newer one, with real scale in China and no proof yet that the scale pays off.

Evidence it holds

The size of the fee base. Customer management revenue was RMB 343.9B in FY2026, up 5%, and the company says the take rate, its fee as a share of sales, rose in the year.⁷ A fee base that big, still growing against a weak consumer, is not a business that is leaking merchants.

The top shoppers stay. 88VIP, the paid membership for its highest-spending customers, grew by double digits to about 64 million members, and the company says it is working on retention.⁶ Members who pay a fee tend to concentrate their spending in one place.

Cloud share. Omdia ranks Alibaba Cloud first in China’s AI cloud market with a 38.1% share, from its 2025 report.⁶ That is the company’s citation of a third-party count, which I could not check against Omdia itself.

Evidence it narrows

It is paying to stand still. To hold quick commerce share, the company let sales and marketing rise from 14.5% to 23.9% of revenue in FY2026.⁷ The E-commerce Group made RMB 39.7B of adjusted EBITA in the June 2026 quarter, against RMB 40.0B a year earlier, on 4% more revenue.⁶ A protected business earns more as its sales rise. Here profit is flat and the spend is large.

Tencent is growing faster on the same customers. Tencent’s advertising grew 22% in the June quarter, while Alibaba’s customer fees fell 7% or grew 1% like for like.⁶¹¹⁹ Those are different pools of advertisers, but they are the same pool of marketing budgets.

The state. The US Department of Defense put Alibaba on its Chinese Military Companies list in June 2026. Alibaba says it is a mistake, that the list bars only the Defense Department from buying from it, and that it will take legal action.¹⁷ The 20-F adds that the department added Alibaba in February 2026 and then withdrew the list, and that there is no assurance the company will not be on it when it is republished.⁸ In Europe the company booked a provision for a EUR 550 million fine under the Digital Services Act in the June quarter.⁶

The marketplace is a wide protection with a flat profit. The cloud is a wide lead in China that I cannot yet price. The protection I am sure of is smaller than the one the share price once assumed.

IIIThe competitive advantage

Alibaba’s advantage is a stack: shoppers, merchants, a payments partner, a cloud, its own chips and its own models, under one roof. Almost no one else in China has all of it, and Tencent, the closest in market value, has the users and the cloud but not the marketplace. The question is whether the stack earns more together than its parts would apart.

Against Tencent

The table above puts the two on the same footing. Tencent earns an operating profit of RMB 258.5B on revenue of RMB 788.5B. Alibaba earns GAAP operating income of RMB 30.3B on RMB 1,045.0B.⁵⁶⁹ Revenue is a third larger and profit is a twelfth of it. Tencent also spends heavily, at 28% of revenue in the last twelve months against Alibaba’s 15%, but it still generated RMB 136.1B of cash after that spending, where Alibaba used RMB 77.5B C (calculated).⁹⁶

The market charges Tencent more for this. Tencent’s market value is US$444.7B against Alibaba’s $243.9B, so Alibaba is worth 55% of Tencent on 133% of the revenue C (calculated).⁹¹

Against Prosus

Prosus is not a rival. It is a Dutch holding company that owns 22.5% of Tencent and runs a set of consumer businesses outside China; its shares are a Tencent stake at a discount, not an Alibaba stake.¹⁰ I include it because it sits next to both in my notebook and holds Tencent shares. Prosus is the cheaper route to Tencent on my numbers, at 22% below its base value, and it does not carry the China commerce fight. Alibaba carries that fight and the cost of the AI build on one balance sheet.

Where Alibaba wins

It wins on breadth and on the speed of its AI products. AI-related cloud revenue has grown by triple digits for twelve quarters, the Zhenwu chips have more than 650 outside customers, and the company launched a frontier language model, Qwen3.8-Max, in August with open weights.⁶ It wins on payments, through the Ant stake, and on cross-border trade, where AliExpress turned an operating profit in the June quarter.⁶ And it has a financing edge: cash and liquid investments of RMB 474.5B at 30 June against debt of RMB 267B C (calculated).⁶

Where it loses

It loses on profit per yuan of sales, as the table shows, and on pace of capital: Tencent’s spending is out of cash flow, Alibaba’s is out of the balance sheet and now out of new shares. It loses on the thing that is hardest to fix, which is a marketplace that must keep spending to hold share against delivery and video-commerce rivals. I did not build a clean table against PDD, JD and Meituan, because I did not open their filings for this page. The Alibaba numbers on this page come from Alibaba’s own filings and the comparisons to Tencent and Prosus come from my pages on them.

Alibaba has the broadest stack in China and the thinnest margin on it. The stack is what the business has going for it; the margin is why the price is where it is.

IVManagement

Alibaba has had a new top team since September 2023. Eddie Wu, a co-founder who ran technology, is chief executive; Joe Tsai is chairman; Toby Xu, a former PwC partner, is chief financial officer.⁸ They own very little of the company, they are paid by a formula, and they have been given a very large balance sheet to spend. I judge them on those three things.

How they are paid

For FY2026 the company paid its directors and executive officers about RMB 250M in fees, salaries and benefits, and granted awards over 28.1M shares, which is 3.5M ADSs.⁸ The cash bonus pool for management is a percentage of adjusted pretax operating profit. The board and the compensation committee decide how much of the pool goes to management; the rest goes to the Alibaba Partnership, a group of long-serving people who hold the right to nominate directors.⁸

That formula pays managers to keep profit high, and the AI build is pushing profit down. Either the pool shrinks, or the definition of adjusted profit leaves out some of the build, or the board decides the number each year. The 20-F does not say which. The equity awards matter more. The option table shows three grants to Mr. Wu: 16,000,000 shares (2,000,000 ADSs) from November 2023 at US$78.37, and two of 12,000,000 shares each from November 2025 at HK$116.70 and HK$152.69, as I read the table.⁸ The Hong Kong share closed at HK$97.05 and the ADS at $98.04. If I have read it right, the 2023 grant is worth about $39M at today’s price, and the two 2025 grants are worth nothing until the share rises 20% to 57%, so most of what he can earn from options depends on a rise.

How much of their own money is in

The directors and executive officers together hold 349.5M shares, 1.9% of the count, worth about $4.28B at the price.⁸¹ Mr. Tsai’s own economic interest is about 114M shares, or $1.40B; Mr. Wu holds 19.7M shares, about $241M. For a company worth $244B that is small. Mr. Wu’s stake is about 0.1% of the company. He carries the strategy and has little at risk beside his job.

What they did with capital over a cycle

I look at five years. In FY2022 to FY2025 Alibaba repurchased RMB 311B of its own shares, and the number outstanding fell from 21.70B at March 2021 to 18.47B at March 2025.¹⁵⁵ That is about RMB 96 for each share by which the count fell, net of shares issued to staff, which is a rough cost C (calculated). It began paying a dividend in FY2024 and paid RMB 33.7B in FY2026.¹⁵⁵ Then, with the build under way, repurchases fell to RMB 7.6B in FY2026 and US$162M in the June quarter, and on 23 August 2026 it sold 710 million new shares in Hong Kong for HK$80B, at HK$112.70 a share, 3.6% under a reference price and 9.0% under the five-day average.⁵⁶¹² HK$112.70 is RMB 96. So it paid about the same price to retire shares that it later got for issuing new ones. The new shares were sold 16% above the HK$97.05 close in Q4 2026 C (calculated).¹²² The sale was good for the company at the price; whether the buying before it was good depends on where the shares go from here.

The deals outside the core tell a mixed story. The company recorded goodwill impairments of RMB 10.5B in FY2024, RMB 6.2B in FY2025, RMB 9.5B in FY2026 and another RMB 4.5B in the June 2026 quarter, RMB 30.7B in total C (calculated).⁵⁶ It sold Sun Art and Intime in FY2025 and booked disposal losses on them; in FY2026 it booked gains on Trendyol.⁷ I give the capital allocation a mixed mark: strong on returning cash while the core was throwing it off, weaker on the acquisitions and now unproven on the build.

Management is paid for profit, owns little, and has moved from buying its own stock to selling it. The incentive to make the AI build pay rests mostly on options struck above today’s price.

VFinancials

Alibaba reports in renminbi under US GAAP, and the ten-year figures below come from five Forms 20-F: FY2018, FY2021, FY2024, FY2025 and FY2026, taking each year from the latest filing that covers it. FY2017 to FY2020 are as first reported; revenue before FY2019 has not been restated for later changes. Where the company changed a definition I say so under the chart.

Revenue grew from RMB 158B in FY2017 to RMB 1,024B in FY2026, a compound rate of 23.1% a year C (calculated). Almost all of it came before FY2023. Revenue was RMB 853B in FY2022 and RMB 1,024B in FY2026, a rate of 4.7% a year C (calculated). Operating income peaked at RMB 140.9B in FY2025 and then fell by two thirds.⁵

The lines tell the story in one picture. Operating margin was never below 8% until FY2026. Cash after capital spending sat above it until FY2024, fell below it in FY2025 and went negative in FY2026. After stock pay, the TTM line is -8.5% of revenue C (calculated).

Capital spending was RMB 126.9B in FY2026, 48% more than the RMB 86.0B of FY2025 and almost four times the RMB 32.9B of FY2024.⁵¹⁵ Operating cash flow was RMB 76.2B, less than half of FY2025’s RMB 163.5B, mainly because profit fell. Free cash flow on the company’s definition was an outflow of RMB 46.6B, against an inflow of RMB 73.9B.⁷ In the June 2026 quarter the outflow was RMB 44.7B, so the last twelve months are an outflow of RMB 72.5B on that definition C (calculated).⁷⁶ The company says the rise comes from quick commerce, user acquisition for the Qwen app and cloud infrastructure, and that the June quarter also reflected the timing of purchases, more processors for AI agents, and higher chip component prices.⁷⁶

The cash line shows the cost. Cash and short-term investments fell from RMB 511.1B in March 2024 to RMB 286.8B in March 2026, and borrowings rose from RMB 170.8B to RMB 260.1B C (calculated).¹⁵⁵ The company still has RMB 474.5B of cash and other liquid investments by its own definition, which adds treasury investments held in other lines, plus the HK$80B raised in August.⁶¹⁸ The RMB 54.1B of contracted capital commitments at 31 March 2026 is small beside the capital being spent.⁸ There is no large unbooked purchase commitment of the kind that sits behind some US cloud builds. The spending is a choice made quarter by quarter.

Quality of earnings

There are four sets of earnings numbers for FY2026: income from operations RMB 50.2B, net income to ordinary shareholders RMB 105.9B, non-GAAP net income RMB 60.7B, and cash after capital spending of −RMB 46.6B.⁵⁷ The first is a clean operating profit but includes goodwill impairments of RMB 9.5B. The second is large because Alibaba marks its investments to market and sold some at a gain. The third strips those out and also strips stock pay. The fourth is what was left in the till. I look at the last one and the first one.

Stock pay is down. Equity-settled stock pay was RMB 11.2B in FY2026, down from RMB 50.1B in FY2021, and the company says it has moved more of its awards into long-term cash incentives.⁵⁷ That is a better habit for the share count, and a worse one for cash: the cash incentive reduces operating cash flow. The tax charge is also odd in the June quarter: income tax expense was RMB 12.8B on pretax income of RMB 22.3B, a rate of 57% C (calculated), and the release does not give a full reason. It matters because my model taxes operating profit at 20%.⁶

Trend in the latest quarter

The June 2026 quarter is the first with the new segments. Cost of revenue rose from 55.1% to 61.8% of revenue, which the company ties to a bigger share of cloud and technology revenue and the new merchant program; product development rose from 6.1% to 8.4%; sales and marketing fell from 21.5% to 17.7% because of the accounting change that moved merchant subsidies into revenue.⁶ General and administrative costs rose from 3.0% to 4.7% on the European fine.⁶ Net: operating margin fell from 14% to 6% in a year.⁶ The figures show a business spending more on what it builds, and a segment table that says where: AI Labs lost RMB 13.9B and All others RMB 3.3B, while the E-commerce Group and Cloud earned RMB 45.4B between them.⁶

VIValuation

How this goes badly wrong

Before the numbers, the ways I lose money, most likely first.

The build does not pay. Cloud grows, but the spending to serve it stays high for years, and the prices of chips and memory keep rising. If cloud capital spending stays at about 60% of cloud revenue until year eight and then falls to 35%, my base falls to $37 an ADS, below my bear case.⁶ This is the most likely way for the investment to fail, because the June quarter’s spending was already 140% of cloud revenue C (calculated).⁶

The marketplace gives way. Customer fees fell 7% in the June quarter and the E-commerce Group’s profit was flat on 4% more revenue.⁶ If the margin of commerce and other businesses is three points lower than my base in every year, the base is $80.

The structure or the funding fails. Alibaba is a Cayman company that controls its China businesses through contracts. It sits on a US Defense Department list again after being added and then withdrawn in the same year, and in August it raised equity to fund the build.⁸¹⁷¹⁹ I cannot price a delisting or a sanction, so I leave a wider margin of safety: 35% instead of 25%. In my base case the build is paid for from cash, not new equity: the cumulative owner cash burn peaks at RMB 119B, about 46% of the RMB 259B of net cash E (estimate).

What an owner can take out

Owner cash is cash left after stock pay and after the capital needed to run the business. In the twelve months to June 2026 that was −RMB 88.4B: operating cash flow of RMB 78.5B, less RMB 156.0B of capital spending and RMB 10.9B of stock pay C (calculated).⁵⁶ Reported net income to ordinary shareholders over the same period was RMB 73.3B and non-GAAP net income RMB 47.9B. So the gap between what the accounts show and what an owner could take out is about RMB 162B C (calculated). In FY2025, before the build, owner cash was RMB 63.6B against net income of RMB 129.5B, or 49% of it C (calculated).⁵

The run-rate answer depends on what counts as the capital needed to stand still. Take the June 2026 quarter’s segment profits, multiply by four, subtract stock pay and tax at 20%, and treat the depreciation of about RMB 47B a year as the capital needed to stand still. Owner cash then is about RMB 78B a year. If the loss of AI Labs is treated as growth spending and left out, it is about RMB 122B E (estimate).⁶ On the market value of RMB 1,623B that is a yield of 4.8% to 7.5%. The Treasury ten-year yield was 4.94%.²⁰ The honest summary is that today’s owner cash is negative and the stand-still cash is not much above a bond yield, so the price depends on growth that has not yet paid.

What I can’t judge

I cannot tell whether the AI Labs loss is an investment that will fade or a permanent cost of staying in the market. I treat it as fading, from RMB 56B in year 1 to nil by year 15, and I show what changes if it does not: with no loss at all the base is $110, and with twice the loss $87.⁶ I cannot judge the market value of Ant or of the private stakes, so I use a haircut of 40% on carrying value; with the investments worth nothing the base is $81 an ADS, and at the company’s own marks with no haircut $108.⁸ I cannot judge how long chip prices stay high. I show the range and not a single number.

I also tested the base against the latest numbers, and the tests point down. The group’s income tax was RMB 34.0B over the last twelve months C (calculated)⁵⁶, against RMB 17.7B in the model’s year 1 E (estimate). At a 25% tax rate the base is $93, and at 30% it is $87. The model also starts capital spending at RMB 210B in year 1, below the June quarter’s pace of RMB 271B a year; if that pace lasts two years the base is $93, and three years $90 E (estimate). I did not move the base for these, because one quarter is noisy and the model runs on yearly paths. But the tests I ran fall below the base, not above it, so I read the base as a value with more room to fall than to rise, inside a range of roughly $87 to $111.

What the price implies

At $98.04 Alibaba’s market value is $243.9B, RMB 1,623B or HK$1,914B, on the 19.902 billion shares in issue, which are about 0.5 billion more than the 20-F counts as outstanding, as I read the balance sheet’s treasury-share line; per-share values use the same count.¹³⁴ I value the New York ADS. The Hong Kong share, counter 9988, closed at HK$97.05, which is $98.96 an ADS at HK$7.8457 to the dollar, 0.9% above the New York price because Hong Kong closes first.¹²⁴ In renminbi the ADS is RMB 652.

Take out net cash and investments after my haircuts, RMB 486B or $29 an ADS, and the operating business is priced at RMB 1,136B.⁶¹²⁸ That is 17.5 times the last twelve months of adjusted EBITA of RMB 64.9B, or 10.4 times the June quarter at an annual rate of RMB 109B, before stock pay and tax C (calculated). On the same pre-tax basis as the Tencent page, which takes only net cash out, Tencent was priced at 11.4 times operating profit.⁹ Alibaba’s equivalent, on GAAP operating income of RMB 30B, is not meaningful: it is 45 times C (calculated).

The market is paying for the commerce business and a share of the cloud, and little for the AI Labs losses. The AI Labs loss of RMB 55B a year at a 10% rate is a cost of roughly RMB 550B of value if it lasts for ever C (calculated), a third of the market value. That is why the loss matters more than it looks.

Run my model backwards

Hold my base case and ask what the price needs. Cloud capital spending from year six on has to be 23.6% of cloud revenue, against 22% in my base; or the cloud revenue path has to be 97.5% of my base, which gives year-10 cloud revenue of RMB 744B against 770B; or the discount rate has to be 10.05% instead of 10.00%.⁶⁸ The price is within 1% of my base, so each single change is small. That is a statement about how finely my model reads: it cannot tell a price of $98 from a value of $99.

The more useful result comes from the one-at-a-time table below. Faster cloud growth adds little to the value, because each yuan of cloud revenue costs capital; what matters is how much capital cloud needs and what margin it earns. The commerce margin and the AI Labs loss matter about as much as the cloud’s growth.

The discount rate

I use 10.0%, a dollar hurdle that matches my Tencent page so that the two can be compared. The Treasury ten-year yield on 30 October was 4.94%; adding Damodaran’s implied equity risk premium of 4.20% at a beta of 1 gives 9.1%.²⁰²¹⁹ I add about a point for the China structure and the uncertain spend. I discount renminbi cash flows at that dollar rate and convert at today’s RMB 6.6529 to the dollar, which assumes the renminbi does not rise. It has risen: it was 6.9004 at the end of March 2026.⁴²² If it keeps rising, values in dollars are higher; a 5% weaker renminbi cuts them by 5%.

One point on the discount rate moves my base by 12% to 19%. At 9% the base is $116 and at 11% it is $86.²⁰²¹

Bear, base and bull

The model runs fifteen years, from July 2026 to June 2041, in renminbi. Revenue has three pieces: cloud and compute, commerce and other businesses, and the losses of AI Labs. The cloud margin is before depreciation and after stock pay. Depreciation is built from the capital spent: about RMB 47B a year on today’s assets, running off over years 6 to 10, and new spending over six years. Cash tax is 20% of operating profit. Each year’s owner cash flow is the margin before depreciation, less tax, less all cash capital spending. I start from the June 2026 quarter at an annual rate for margins and from the twelve months for revenue. Year 1 cloud revenue is RMB 235B against RMB 173B in the last twelve months.⁶⁸

Bear. Cloud growth slows to 8% in year five, spending stays high, commerce margins slip to 12.5% and AI Labs keeps losing money for ten years. The commerce and cloud businesses are worth about what the investments are worth, and the ADS is worth $41.

Base. Cloud grows 36% in year 1 and fades to 7% by year 10, reaching RMB 770B; its capital spending falls from 78% of its revenue to 22%; commerce margins ease from 15.5% to 14.5%; AI Labs losses fall from RMB 56B to nil. I first ran the cloud at 38% in year 1 fading more slowly and got $111; I then compared the year-10 cloud revenue of about RMB 911B, or US$137B, to Amazon’s cloud business and cut the path, which gave $99. So the base is a judgement within a range of roughly $87 to $111, not a point.

Bull. Cloud grows 41% in year 1 and keeps above 18% in year 5, margins reach 36% before depreciation, capital spending falls to 20% of cloud revenue, and AI Labs reaches break-even by year 10. The ADS is worth $149.

Weighted 25%, 50% and 25%, the cases give $97 an ADS against a price of $98.04.¹ Held to the base, the price returns 10.0% a year; in the bear it returns 5.8% and in the bull 13.2% C (calculated). A holder is paid 10% a year only if the base happens, and the Treasury bond pays 4.94% with no risk of that kind.²⁰

The next best use of the same dollar

Alibaba is not the only way to own a Chinese internet business. On my own numbers, Tencent is 24% below its base of $65 and has a margin-of-safety price of $45, close to its price of $49; Prosus is 22% below its base of $49 and holds Tencent through a discount.⁹¹⁰ Alibaba is 1% below its base and 53% above its margin-of-safety price of $64. Tencent carries similar structure risks and a similar build, but earns RMB 136.1B of cash after the spending where Alibaba uses RMB 77B.⁹⁶ A dollar in Alibaba buys less gap between price and value than a dollar in the other two. That is my own arithmetic on my own models, not a ranking.

Quality and price

These are two questions. As a business, Alibaba has a wide customer base, a leading cloud and real engineering; its profit margin is thin and falling and its cash flow is negative. As a purchase at $98.04, the price is about my base value and above my margin-of-safety price. A good business bought at a fair price is not the same as a good business bought cheaply, and I do not think the price here leaves room for the things going wrong that I list above.

Multiplying the odds

The base case needs five things at once: cloud keeps growing at least 30% for three years (I give that 70%), capital spending per yuan of cloud revenue falls as modelled (65%), the commerce margin holds at about 14% to 15% against quick-commerce rivals (70%), AI Labs losses shrink to a third of today’s by year six (60%) and no sanction, delisting or structural break interrupts it (90%). Multiplied, that is 17% E (estimate). These odds are my guesses, but they show why a fair value near the price is not a safe one: even generous odds on each step leave about one chance in six that all of them come right. The base is a weighted result, not the most likely single path.

The price is about what the base case is worth. That is a fair price for a business that needs five things to go right, and it is not a cushion.

Where this could be in 5 and 10 years

Value in a later year is my model rolled forward. It is the present value at that date, at 10%, of the owner cash still to come, plus the net cash and investments left after the early burn, per today’s 2,487.7 million ADS. Owner cash is paid out once it turns positive.

For the base to hold at five years, cloud revenue has to reach about RMB 483B and the build has to be paying: owner cash turns positive in year 4 and is RMB 57B in year 5. For ten years it needs cloud near RMB 770B at a 33% margin before depreciation, and capital spending near 25% of cloud revenue. Owner cash is then RMB 140B a year E (estimate).

The cases differ on three numbers. Cloud growth in year 10 is 4% in the bear, 7% in the base and 9% in the bull. Capital spending in year 10 is 29%, 25% and 23% of cloud revenue. The AI Labs loss in year 10 is RMB 15B, RMB 12B and RMB 4B. Owner cash in year 10 comes to RMB 48B, RMB 140B and RMB 222B E (estimate).

The most likely way the base is wrong is capital spending. If cloud keeps needing 34% of its revenue from year six, today’s value is $64 and the value in 2036 is $94, not $170 E (estimate). The base returns 8.3% a year, under the 10.0% earlier on this page, because here I hold the net cash and investments flat and they earn nothing; the earlier figures run to the end of the model and these stop at year 10.

These are estimates of value, not forecasts of the share price. The price can sit above or below value for years.

VIIThe case for the price

What most people miss

First, the balance sheet carries a lot. After my haircuts, net cash and investments are $29 an ADS, 30% of the price C (calculated).⁶¹²⁸ Second, AI-related cloud products are already RMB 12.4B a quarter, 26% of cloud revenue, after twelve quarters of triple-digit growth.⁶ Third, the E-commerce Group still earns RMB 39.7B of adjusted EBITA a quarter, about RMB 159B a year, or 9.8% of the market value each year before tax and before the AI losses C (calculated).⁶ Last quarter AI Labs absorbed 35% of that profit C (calculated).⁶

The bull case

The case is made by the company. Eddie Wu said on 20 August 2026 that the improving commercialisation of the AI stack drove a strong quarter, that cloud’s external revenue growth accelerated to 45%, and that AI product revenue grew by triple digits for a twelfth quarter.⁶ Toby Xu said that synergies and AI monetisation give the company more strategic and financial flexibility to keep investing.⁶ The company also said it raised HK$80B to "extend the Company’s global AI leadership" and will spend all the net proceeds on AI.¹⁹ I did not find a named outside investor whose published view on Alibaba I could check against a filing, so I have not quoted one.

Against reality

If cloud grows 36% in year 1 and fades as in my base, if its spending per yuan of revenue falls from 78% to 22%, and if AI Labs shrinks its loss from RMB 56B to nil by year 15, then the price of $98.04 is possible, because it equals my base. Today’s numbers are further away than that sounds. Capital spending in the June quarter was 140% of cloud revenue, the AI Labs loss was RMB 13.9B in a quarter, and customer fees fell 7%.⁶ The case works if cloud growth stays near 45% while the spending per yuan drops by about a third in a year. The record so far is the opposite: spending rose 75% in a year.

What a 10x would take

A tenfold rise from $98.04 is $980 an ADS and a market value of $2.44 trillion, 5.5 times Tencent’s market value on its page.⁹ That is 25.9% a year for ten years. At 20 times owner cash flow it needs about $122B of owner cash a year, RMB 811B, where today it is negative. At a 20% owner cash margin revenue must reach RMB 4,057B, 3.9 times today and a growth rate of 14.5% a year; at 25% it is RMB 3,245B and 12.0% a year C (calculated). Revenue grew 4.7% a year over the last four fiscal years.⁵ I think a tenfold move is not plausible within ten years. My bull case is $149, 1.5 times the price.

Where I come down

The price moved me less than the cash. The business is good and the cloud is real, but the spending is large and the owners are not getting paid yet. What did not move me is the headline growth, because capital eats most of it. What would raise my values is evidence on three lines: the cloud margin staying up as revenue doubles, the capital spending per yuan of cloud revenue falling, and the AI Labs loss turning down.

VIIIRisks, and what would change my view

Risks

• Capital intensity. The business spent RMB 126.9B in FY2026 and RMB 67.7B in one quarter. If it does not fall, the base is about $37.⁵⁶
• The commerce core. Profit fell 44% in FY2026 and was flat in the June quarter on higher revenue. The company is paying to hold quick-commerce share.⁷⁶
• The state and the structure. The CMC listing, VIE contracts, US outbound rules and a European fine are all live. The company says the CMC listing does not stop it doing business as usual, but the 20-F says designation can bring wider consequences.¹⁷⁸
• Funding. It sold 3.7% of the company in August. A second raise at a lower price would dilute holders unless the build were already paying.¹²
• Governance. One class of shares with one vote each, but the Alibaba Partnership has nomination rights over directors, and the Hong Kong listing rules treat it as a weighted-voting structure.¹⁶⁸ Insiders own 1.9%.
• Model risk. The model reads the price and the base as equal. A small error in the cloud’s capital needs, which I cannot observe, moves the value more than the cloud’s growth does.
• Currency. The renminbi rose 3.7% against the dollar in seven months.⁴²² A reversal would cut dollar value.

What would change my view

Up. Capital spending per quarter falling toward RMB 50B while cloud growth stays above 35%. The AI Labs loss below RMB 12B a quarter. Customer management revenue growing again like for like. A buyback restarting without a placing.

Down. Cloud growth under 30% with capital spending still above 100% of cloud revenue. The E-commerce Group’s quarterly adjusted EBITA below RMB 35B. A second equity raise. A sanction rather than a list.