What Does Alibaba Do? The China Cloud Conversion Model

Alibaba converts e-commerce cash into China’s largest AI cloud: it gets paid when an external customer is billed, while the chip that would fill the rack is still the stuck step.

HYPERSCALERS · Data as at 20 Aug 2026 · Next update: after September-quarter 2026 results
Affects: Hyperscalers · China · Semiconductors

It is a hyperscaler on this layer: a buyer large enough to pull chips, halls and power. Unlike Amazon, Microsoft or Google, it does not print a merchant-cloud remaining-performance-obligation. The earn is still mostly commerce. The spend is cloud. See who the hyperscalers are.

In short: Alibaba sells cloud and AI compute on top of a still-larger e-commerce group. In the quarter ended 30 June 2026, AI Cloud and Compute Services was RMB48.437bn of RMB268.953bn revenue, up 45%. Cloud adjusted EBITA was RMB5.628bn. Capex was RMB67.678bn. Free cash flow was an outflow of RMB44.670bn. On 22 Sep 2026 the company said global shortages across the AI data-centre supply chain limit how fast it can scale. Chips are the stuck step.

Education only, not advice to buy or sell any security. Figures are from company filings and company releases dated as shown. Outlook is labelled outlook. Secondhand reports are labelled reported.

Numbers that matter

Sources: Alibaba Group June Quarter 2026 results, Form 6-K exhibit 99.1, 20 Aug 2026; company PDF of the same release; Alibaba Cloud Apsara release, 22 Sep 2026. Status chips are Second Order’s read. The 20 GW figure is a company target, not a print.

Our arithmetic. Cloud was 18.0% of group sales (48,437 ÷ 268,953 = 0.180). AI-related products were 25.6% of the cloud segment (12,376 ÷ 48,437 = 0.256). Cloud adjusted EBITA margin was 11.6% (5,628 ÷ 48,437 = 0.116), matching the 11.6% spoken on the 20 Aug 2026 call. Capex was 1.40× that quarter’s cloud sales (67,678 ÷ 48,437 = 1.40). Operating cash flow RMB22.945bn less PPE purchases of RMB67.660bn plus a RMB45m buyer-protection adjustment is −RMB44.670bn, matching the 6-K reconciliation. On the same call, management said AI-related products were 35% of external cloud and that four times this quarter’s RMB12.376bn is about RMB49.5bn. Our check: 12,376 × 4 = 49,504. If 12,376 is 35% of external, external cloud is about RMB35.360bn (12,376 ÷ 0.35). That 35% split is spoken, not a 6-K line.

What does Alibaba actually do?

Three steps, in order.

Earn. Light. Bill.

1. It earns on commerce — and now, more, on cloud. In the quarter ended 30 June 2026 the group recast its segments: Alibaba E-commerce Group, AI Cloud and Compute Services (Cloud Intelligence Group plus T-Head), AI Labs and Applications, and All others. E-commerce was RMB205.862 billion. Cloud was RMB48.437 billion. Labs and apps were RMB3.338 billion. The commerce invoice still dwarfs the cloud invoice. See AI capex.
2. It lights halls, and it designs some of the silicon that goes in them. Capex was RMB67.678 billion, up 75%. The 6-K names three reasons: procurement-cycle swings, more CPU-compute for expected agent demand, and “higher pricing of a broad range of chip components.” T-Head’s Zhenwu chips, including the M890, are in commercial use via Alibaba Cloud at more than 650 external customers. On 22 Sep 2026 the company unveiled the Zhenwu V900 — three times the M890, mass production in Q1 2027 — and a supernode of up to 500,000 cards. That is a plan, not this quarter’s recognised hour. See GPU as a service and cash capex.
3. It bills external customers for cloud and AI products — and uses a slice internally. Total cloud revenue and revenue from external customers both grew 45%. AI-related product revenue of RMB12.376 billion was the twelfth consecutive quarter of triple-digit year-over-year growth. Alibaba cites Omdia’s “AI Cloud Market: China — 2025” for a 38.1% share of China’s AI cloud market. That share is Omdia’s, as cited by the company, not a 6-K audit of invoices. See remaining performance obligation for the backlog print Alibaba does not give.

How does Alibaba make money?

It gets paid when a cloud customer is billed, and when a merchant or consumer is billed on Taobao, Tmall, AliExpress and the rest of the e-commerce group. Three things decide how much of that cash becomes a live rack:

• The cloud print — RMB48.437 billion, 18.0% of sales. External growth and total growth both +45%. AI-related products RMB12.376 billion.
• How fast chips arrive and substitute — the 6-K’s “higher pricing of a broad range of chip components”; T-Head as the in-house substitute. On the 20 Aug 2026 call, management said deploying proprietary chips in place of commercially procured GPUs is one of three ways to shorten AI-capex payback.
• Whether models and apps consume the same cash — AI Labs and Applications lost RMB13.861 billion of adjusted EBITA on RMB3.338 billion of sales. Cloud made RMB5.628 billion. E-commerce made RMB39.749 billion.

On 20 Aug 2026 Alibaba printed RMB268.953 billion of revenue, up 9%. Income from operations was RMB15.161 billion (6% margin, from 14%). Adjusted EBITA was RMB27.329 billion, down 30%, “primarily attributable to the investment in technology, partly offset by the improved operating results in our Cloud business.” Net income was RMB10.444 billion. Property and equipment, net, rose to RMB312.497 billion from RMB282.699 billion at 31 Mar 2026. Deferred revenue and customer advances were RMB79.439 billion current plus RMB4.682 billion non-current — not an Azure-style remaining-performance-obligation.

Who gets paid. Alibaba, on the external cloud invoice and on commerce. Chip, server, construction and power vendors, on the RMB67.678 billion of capex. T-Head is paid inside the group when its die is used instead of a merchant GPU. Money sticks at energised, accepted accelerator hours billed to an external customer — not at a 45% growth headline, not at a 20 GW target, and not at an unfinished chip. See cash capex and which capex lines to watch.

The RMB380 billion three-year AI-and-cloud infrastructure plan (announced February 2025; management on 20 Aug 2026 said about RMB190 billion had been spent by 30 June) is outlook, not this 6-K’s cash line. So is the 22 Sep 2026 target of more than 20 GW by 2032.

What limits it?

Chips first. Unfinished substitution second. Watts in China third.

1. A spent yuan is not a live card. The 6-K says capex jumped on procurement cycles, extra CPU capacity, and higher chip-component prices. Wu, on 22 Sep 2026: “Global shortages across the AI data center supply chain are currently limiting the speed at which we can scale our compute infrastructure.” Cash is not the bind: liquid investments were RMB474.505 billion. See the China Alibaba 20 GW Pulse.
2. Own silicon is a plan until Q1 2027. Zhenwu is already in production use at 650-plus customers. The V900 — the chip the company called its next training-and-inference processor — is scheduled for mass production and commercial release in Q1 2027. Until then, Alibaba still buys a mix. Heterogeneous clusters (the 6-K’s phrase) are an orchestration cost, not a solved step.
3. Power still gates the hall. China’s official AI compute count and the power used by data services are not moving together; our China compute-versus-power Pulse is the dated read. A 20 GW 2032 target is a megawatt clock, not a 2026 invoice. See speed-to-power.
4. Labs consume what cloud earns. Cloud adjusted EBITA of RMB5.628 billion does not cover the RMB13.861 billion Labs loss. E-commerce of RMB39.749 billion does. If commerce thins, the funder — not the chip — becomes the bind.
5. There is no RPO conversion print. Amazon, Microsoft and Google convert remaining performance obligations as halls light. Alibaba’s June 6-K does not. Deferred revenue of about RMB84.1 billion (79.439 + 4.682) is not a substitute. See RPO, explained.

The honest uncertainty is conversion. A 45% cloud print, a RMB67.7 billion capex print, a Q1 2027 V900 clock, and a 2032 20 GW clock are different facts.

Who else does this?

Roles, not a league table.

Alibaba sits on the China conversion seat: commerce in, cloud hours out, silicon mixed. See neocloud vs hyperscaler.

What would change this view?

The read on the model weakens if:

• September-quarter 2026 results print cloud growth well inside 45%, or AI-related products fall off triple-digit year-over-year
• Capex drops while Wu’s “global shortages” language disappears — demand, not supply, would then be the bind
• V900 slips past Q1 2027, or Zhenwu customer counts stop being printed
• E-commerce adjusted EBITA no longer covers the Labs loss plus the capex residual
• A dated remaining-performance-obligation, or an Oracle-style megawatt-and-utilisation print, appears
• The China power Pulse shows watts, not chips, as the tighter clock
• Next GPU attach is late because packaging or HBM is short — the stuck step would then sit in semiconductors

It stays intact while external cloud holds around 45% growth, AI-related products stay a rising share of that invoice, capex stays large against cloud EBITA, V900 stays a 2027 clock, 20 GW stays a 2032 clock, and chips remain the layer’s stuck step in China.

Practical takeaways

• Model on this layer: Earn → Light → Bill
• June quarter 2026: RMB268.953bn sales, RMB48.437bn cloud (18.0%, +45%), RMB5.628bn cloud adjusted EBITA (11.6%)
• Capex RMB67.678bn (+75%); FCF outflow RMB44.670bn; cash RMB474.505bn
• Labs lost RMB13.861bn of adjusted EBITA; e-commerce made RMB39.749bn
• Clocks: Zhenwu V900 in Q1 2027; more than 20 GW by 2032 (targets)
• Stuck step: chips, not the cloud invoice and not cash on the balance sheet
• Next hard read: September-quarter 2026 results (company typically prints in November; date not yet posted as of 26 Oct 2026)

FAQ

What does Alibaba do?

Alibaba runs China’s largest AI cloud and a still-larger e-commerce group. It bills external customers for cloud and AI products, spends tens of billions of yuan a quarter on infrastructure and its own T-Head chips, and uses some of that compute internally. It is a hyperscaler on this layer because of that spend, not because it prints a US-style remaining-performance-obligation.

What does Alibaba do in simple terms?

It sells shopping and cloud. Shopping still brings in most of the cash. Cloud is what is growing at 45%. The limit is getting enough chips into the halls.

Is Alibaba Cloud profitable?

Yes on the dated adjusted-EBITA print: RMB5.628 billion on RMB48.437 billion of cloud revenue in the quarter ended 30 June 2026 (20 Aug 2026), an 11.6% margin on our arithmetic. Group income from operations was RMB15.161 billion. Free cash flow was an outflow of RMB44.670 billion after RMB67.678 billion of capex. That is not a forecast.

How is Alibaba different from AWS or Azure?

Amazon and Microsoft convert remaining performance obligations as halls light. Alibaba’s June 6-K does not print that backlog. Its cloud invoice is smaller in dollars (US$7.139 billion this quarter) and sits inside a commerce group. It also designs its own accelerators through T-Head. See What Does Amazon Do? and RPO, explained.

What is Alibaba’s capex?

The June 2026 quarter printed RMB67.678 billion (US$9.975 billion), up 75% from RMB38.676 billion (20 Aug 2026). Management has described a three-year plan of RMB380 billion, with about RMB190 billion spent by 30 June 2026 (spoken on the 20 Aug 2026 call). Those plan totals are not this 6-K’s cash line.

Should I buy Alibaba shares?

We do not make buy or sell calls, and nothing here is a recommendation. What we can tell you is what the company does, how it gets paid, and which figures would change the picture. The decision is yours.

Keep reading on the map

• The layer: Hyperscalers — capex pull and the delivery bind
• The term: Hyperscaler, explained · Who are the hyperscalers?
• The mechanism: AI capex · Cash capex · GPU as a service · Speed-to-power
• The contrast: Remaining performance obligation · Neocloud vs hyperscaler
• The peers: Google · Amazon · Microsoft · Meta · Oracle
• Where it happens: China · 20 GW by 2032 · Compute up 146%, power 35%

The members cut on the long clock is already live: Where does the stuck step move?. This page stays the free Stack read of the Alibaba model.

How we check this

Figures come from company filings and company releases, each dated below. The 35% external mix, RMB49.5bn run-rate, RMB380bn plan and RMB190bn spent-to-date are spoken on the 20 Aug 2026 call. The 20 GW target, V900 dates and the “global shortages” sentence are from the 22 Sep 2026 company releases. Omdia’s 38.1% share is cited by Alibaba. Arithmetic is shown in full. Outlook is labelled outlook.

Last reviewed: 26 Oct 2026 · Next review: after September-quarter 2026 results print cloud growth against the 45% June print and capex against the RMB67.678bn run-rate

Spot an error? Tell us; we will correct it here. Editorial standards →

Sources

1. Alibaba Group, June Quarter 2026 results (Form 6-K exhibit 99.1), 20 Aug 2026 — revenue, cloud, AI-related products, EBITA, capex, FCF, cash, PPE, deferred revenue, Zhenwu 650+ customers, Omdia 38.1% as cited
2. Alibaba Group, same results, company PDF, 20 Aug 2026
3. Alibaba Group, June Quarter 2026 results page, 20 Aug 2026
4. Alibaba Group, earnings webcast archive, 20 Aug 2026 — spoken 35% mix, ~RMB49.5bn run-rate, RMB380bn plan, ~RMB190bn spent
5. Alibaba Cloud, Apsara full-stack AI roadmap, 22 Sep 2026 — 20 GW by 2032; Zhenwu V900 Q1 2027
6. Alibaba Group, Eddie Wu Apsara keynote page, 22 Sep 2026 — global supply-chain shortages limiting scale
7. China Pulse, 20 GW by 2032 — our dated map of the same company prints
8. China Pulse, Compute up 146%, power 35% — watts versus installed compute