Alibaba completed a HK$80.0 billion placement on 26 August — 710 million new ordinary shares at HK$112.70, about US$10.2 billion gross, with 100% of the net proceeds earmarked for AI. The obvious thing to divide that by is the company's own AI revenue, and the obvious division is startling. Against the AI run rate Alibaba gave on its May call — RMB 35.8 billion annualised — the raise is 1.93 times everything the company's AI products earn in a year.
That figure was four months old and had already been replaced. In the June quarter, reported on 20 August, three days before the placement was announced, AI-related product revenue was RMB 12,376 million. Four times that is RMB 49.5 billion — the US$7.3 billion the chief executive quoted on the call — and it is 38.0% above the March quarter's RMB 8,971 million. On the current number the raise is 1.40x. On the number management put on the record for the September quarter, it is about 1.02x.
The multiple is not a fact about Alibaba. It is a fact about which quarter you divided by, and two quarters nearly halve it.
The points
- The raise, as completed. 710 million shares at HK$112.70 — HK$80.0 billion gross, HK$79.8 billion net — closed on 26 August. The dollar figure everyone quotes uses HK$7.8399, the rate in Alibaba's own pricing announcement.
- The denominator is reported, not estimated. RMB 8,971 million of AI-related product revenue in the March quarter, RMB 12,376 million in the June quarter, both on the face of the company's results announcements. Management's "annualised run rate" is four times the quarter, which is how RMB 12,376m reproduces the RMB 49.5 billion and US$7.3 billion quoted on the call.
- The only forward number Alibaba gave is on the call, not in the release. The September-quarter run rate "will approach US$10 billion" — about RMB 16.9 billion in the quarter, another 37% sequential step.
- Both itemised uses of the proceeds are infrastructure. The completion announcement splits them: approximately 60%, HK$47,871 million, to expand global computing infrastructure; approximately 40%, HK$31,914 million, to build hyperscale AI data centres and upgrade storage, databases and networking. Neither line names model development.
- The model is downloaded, not sold. Management said in August that the open-weight series has passed 3 billion downloads and 300,000 derivative models. Both are call-only figures.
- What that adoption is worth is disclosed, and it is a curve. Model and application services ARR, model-as-a-service included, was "over RMB 8 billion" in May, past RMB 16 billion in August, against a RMB 30 billion year-end target. All three are call-only.
The denominator, three times
The summary card above is the same $10.2 billion divided by three vintages of the same disclosure: 1.93x on May's run rate, 1.40x on August's, 1.02x on the one management described for September.
Every raise-to-revenue multiple works this way, which is why the comparison is weak on its own: capital raised to build multi-year capacity is not supposed to be a small multiple of current revenue, and no hyperscaler would pass that test either. What makes this one worth printing is the speed. A denominator compounding at 37 to 38% a quarter turns a two-year multiple into a one-year multiple in six months without the numerator moving at all. Anyone quoting a raise-to-AI-revenue ratio on Alibaba is quoting the date they looked it up.
Both named uses are machines
The 23 August announcements said only that Alibaba intended to use "100% of the net proceeds from the Equity Placement to invest in its full stack AI capabilities, including to expand and enhance its AI infrastructure." That is a sentence, not an allocation. The completion announcement on 26 August turned it into two line items, and they are the second table above: HK$47,871 million for global computing infrastructure "to meet accelerating customer demand", and HK$31,914 million for hyperscale data centres and a cloud upgrade "across storage, databases, and high-performance networking".
Neither is model work. That does not mean Alibaba has stopped paying for models — the AI Labs and Applications segment, which now houses Qwen, lost RMB 13,861 million of adjusted EBITA on RMB 3,338 million of revenue in the June quarter alone. It means the equity was raised against the part of the stack that has customers, and the part that has customers is compute. When the placement was announced we measured it against one quarter of capital expenditure, which was the only denominator available; the itemisation came three days later.
The chief executive said as much when he was asked where the value sits. His answer on the August call was that near-term value is in chips and AI cloud infrastructure, and that charging for model access by the API call is "a short-term transitional approach" rather than the endgame.
The free model has a price list, and it is the serving
Qwen's weights are published. Thomson Reuters adapted them into Thomson-1, its own model, to lean less on a frontier lab — the company's own announcement, and its chief technology officer's framing of it in the press. Nothing in that transaction pays Alibaba. Reuters has reported that Alibaba plans to attach revenue-sharing terms to its next open-weight release, aimed at large firms that resell the model as a service — but the rate is not set, the company has announced no such licence, and today the weights are free to self-host.
What is not free is running them at scale on Alibaba's machines, and that is the line management chose to quantify. Model and application services ARR doubled from RMB 8 billion in May to RMB 16 billion in August, with a RMB 30 billion target for year-end. At the June quarter's rate, RMB 16 billion is 32% of the whole annualised AI product line, and the $10.2 billion raise is 4.3x it.
So the two curves in this story point the same way. Adoption of the model is free and enormous. Revenue arrives one layer down, in serving and in cloud — which is exactly and only what the placement is itemised to buy.
What it does to the cloud line we project
Our forward model for Alibaba already carries this placement: it was rebuilt on the June-quarter segments after the raise closed, so the cash and the share count are in it and the funding event itself moves nothing. It also carries no line for AI-related product revenue. That disclosure sits inside AI Cloud and Compute Services, which the model projects as a single growth driver at 9.7% a quarter — our assumption, not a company statement.
Cross the two and the arithmetic is uncomfortable. AI product revenue was RMB 12,376 million of the segment's RMB 48,437 million in June, or 25.6%. Take management's September figure at face value and that becomes about RMB 16,963 million. Hold our 9.7% and the whole segment comes to RMB 53,135 million, which leaves the non-AI remainder at RMB 36.2 billion against June's RMB 36.1 billion: growth of 0.3% in a quarter, which is to say none.
One of those two numbers is wrong. Either AI Cloud grows faster than we have it, or the traditional half of Alibaba Cloud has stopped growing while the AI half takes the segment. The raise does not settle it, because the raise is funding capacity for the half that is compounding. The September print does.
What to watch
- The September-quarter AI product line. RMB 16.9 billion in the quarter is what "approach US$10 billion" annualised means. It is the only forward figure Alibaba has given, and it was given on a call.
- Whether model-serving ARR reaches RMB 30 billion. RMB 8 billion in May, RMB 16 billion in August, and the target implies another near-doubling in four months.
- What the non-AI half of Alibaba Cloud does. If the AI line steps 37% and the segment does not accelerate past our 9.7%, the remainder is flat — and that would be the more important disclosure of the two.
- The licence on the next open-weight release. A revenue-sharing term would be the first time Qwen adoption paid Alibaba directly rather than through a meter. Nothing has been announced.
- September-quarter capital expenditure. The gross raise is about one June quarter of it, and the completion announcement says all of it buys infrastructure.
Sources and provenance: the placement's size, price, proceeds, completion date and the itemised use of net proceeds are from Alibaba's 23 and 26 August 2026 announcements, and the HK$7.8399 conversion is the pricing announcement's own. The March and June 2026 quarter figures — AI-related product revenue, the AI Cloud and Compute Services segment, the RMB 6.7851 convenience rate — are as Alibaba reported them, captured in our March-quarter and June-quarter notes. Everything attributed to management above — the annualised run rates, the September US$10 billion, the serving-ARR figures and target, the download counts and the remarks on API pricing — was said on the calls and appears in no release; Alibaba publishes no transcript of its own and the English is an interpreter's. Thomson Reuters' adapted Qwen model is that company's own announcement; the reported plan to attach revenue-sharing terms to the next open-weight release is Reuters, citing two people, and Alibaba has announced nothing. Every multiple and sequential rate here is our arithmetic, and the 9.7% quarterly cloud growth is our own assumption.
