Update, August 23, 2026: the first open question below — whether capex has a stated ceiling — got a partial answer three days later, and it was not a ceiling. Alibaba proposed a $10.2 billion Hong Kong placement, all of it for AI, which buys about one quarter of the capital expenditure discussed here and adds 14.6% to the cash balance this piece said was good for roughly ten quarters.
Alibaba reported the June quarter before the US open this morning. Revenue was RMB 268,953 million (US$39,639 million), up 9%. Cloud — now reported as AI Cloud and Compute Services — grew 45%, accelerating again. Non-GAAP net income fell 38%, net income fell 75%, and free cash flow was an outflow of RMB 44,670 million (US$6,584 million) against an outflow of RMB 18,815 million a year earlier.
We published a preview last night that pre-registered specific bars. Most of them were cleared or missed cleanly, which is the point of writing them down. One of them was simply wrong, and that is the most useful thing on this page.
What we learned
- The cash bars were missed by a distance, not a whisker. The preview set two: trailing free cash flow improves on any print better than −$2,626M, and reaches zero at +$3,954M. The print was −$6,584M. Trailing free cash flow went from −$6.58B to −$10.54B — it did not improve, it deteriorated by almost exactly the amount the improvement was supposed to be.
- Our Rule of 40 grid was too narrow, and the reason is capex. The preview built a grid of plausible outcomes and said "the whole grid sits between +0.9 and +7.2". The quarter printed −1.9. The grid bounded revenue growth and free-cash-flow margin across a range of reasonable quarters, and Alibaba spent outside that range: capital expenditure rose 75% to RMB 67,678 million, which is 2.5× the quarter's entire adjusted EBITA. A band that cannot contain the actual result was a band drawn around the wrong variable.
- The AI burn now has its own segment, and it is larger than it looked. Alibaba recut its reporting this quarter, pulling the model labs, Qwen Consumer and QwenWork out of "All Others" into AI Labs and Applications. That line lost RMB 13,861 million of adjusted EBITA on RMB 3,338 million of revenue — a −415% margin, against a RMB 3,224 million loss a year ago. It was invisible inside a residual bucket until the quarter it got big.
- Revenue met expectations exactly and earnings missed all of them. Against LSEG's RMB 268,880m the print was +0.03% — in line. Non-GAAP EPS per ADS was $1.26 against a $1.60 consensus, a 21% miss, and below the bottom of the $1.49–1.94 band the estimates spanned. One line did precisely what was asked and the other missed every estimate in its range.
- The commercial thesis is working while the cash thesis is not. Cloud external revenue growth accelerated to 45%, AI-related product revenue reached RMB 12,376 million — a twelfth consecutive triple-digit quarter, and 25.6% of cloud — and cloud's adjusted EBITA rose 133% to an 11.6% margin. Quick commerce grew 45% to RMB 53,295 million. None of that is in doubt. What is in doubt is what it costs, and this quarter the answer was RMB 67.7 billion of capex and a goodwill impairment of RMB 4,458 million.
The denominator on the burn
Adjusted EBITA was RMB 27,329 million, down 30%. Capex was RMB 67,678 million. So Alibaba spent two and a half times its own operating profit measure on infrastructure in three months, and the release says why: procurement timing, more CPU compute for AI agents, and "higher pricing of a broad range of chip components."
Take the other denominator and it reads the same way: RMB 67,678m of capex against RMB 268,953m of revenue is 25.2% of everything the company billed in the quarter, and 1.40× the entire AI Cloud revenue line of RMB 48,437m. A quarter of revenue is not a rate — capex is lumpy and one period of procurement timing is not a policy — but it is the size of the thing that has to be funded before any of it earns.
And the blended +9% hides which business is paying. Alibaba E-commerce Group grew 4% to RMB 205,862m, and that segment is 76.5% of consolidated revenue. Inside it, China e-commerce revenue fell 8% to RMB 110,900m, with customer management down 7% as reported — 1% growth on a like-for-like basis excluding the new business development program's contra-revenue — and the +4% only holds because quick commerce grew 45%. AI Cloud grew 45% off RMB 48,437m. So the majority of the company grew at 4% while a fifth of it grew at 45%, and the capex is going to the fifth.
Against a cash and liquid investment position of RMB 474,505 million, the company can fund this for a long time. That is the correct read and it is not the same as saying the spending is free. At this quarter's rate the position funds roughly ten quarters of outflow before anything else changes — a real constraint with a distant date on it, which is exactly the kind that stops being discussed until it is close.
What the segment recut does to our model
Our Alibaba model projects four verticals: Alibaba China E-commerce, Cloud Intelligence, AIDC and All others. Three of those are no longer reported lines. China E-commerce and AIDC were merged with Freshippo into Alibaba E-commerce Group; Cloud Intelligence was combined with T-Head into AI Cloud and Compute Services; AI Labs and Applications is new.
The model is not wrong about the business, but its verticals can no longer be reconciled to a disclosed segment, which is the standard this site holds them to. That is a rebuild rather than an edit, and it is not something this article does.
What to watch
- Whether capex has a stated ceiling. RMB 67.7B in a quarter, +75%, with no guided envelope. The first number management puts a bound on is the one that changes the cash case.
- AI Labs and Applications next quarter. A −415% margin on a line that just became visible is either a peak or a trend, and one more print distinguishes them.
- Cloud EBITA margin above 12%. The CFO called it 12% this quarter, up from 7.2% a year ago. That glide is the only thing currently paying for the build.
- The cost bank of goodwill. A RMB 4,458m impairment appeared with no prior-year comparative and is most of the gap between adjusted EBITA falling 30% and operating income falling 57%.
- Whether the estimate band narrows. Consensus for this quarter spanned $1.49–1.94 on non-GAAP EPS, 30% wide. A band that wide is analysts telling you they cannot model the adjustments.
Every reported figure — revenue of RMB 268,953m (US$39,639m), income from operations of RMB 15,161m, adjusted EBITA of RMB 27,329m, net income of RMB 10,444m, non-GAAP net income of RMB 20,715m, GAAP diluted EPS per ADS of RMB 3.71 (US$0.55) and non-GAAP of RMB 8.52 (US$1.26), free cash flow of −RMB 44,670m (US$6,584m), capital expenditure of RMB 67,678m, cash and other liquid investments of RMB 474,505m, the goodwill impairment of RMB 4,458m, the segment revenue and adjusted EBITA tables, AI-related product revenue of RMB 12,376m, and the 45% cloud growth — is from Alibaba's June quarter 2026 results announcement of 20 August 2026, captured in full in our June-quarter note. The SEC 6-K carrying the same document as Exhibit 99.1 had not appeared on EDGAR at the time of capture. The consensus revenue figure of RMB 268,880m is LSEG's, press-reported; the $1.60 non-GAAP EPS consensus and the $1.49–1.94 range are press-reported from third-party estimate feeds and are not series this site stores or verifies. The −$2,626M and +$3,954M bars, the −$6.58B trailing figure, and the "+0.9 to +7.2" grid are quoted from our own preview and were computed there from Alibaba's reported quarterly figures. Derived here: the −$10.54B trailing free cash flow, the 2.5× capex-to-adjusted-EBITA ratio, the 25.2% capex-to-revenue ratio, the 1.40× capex-to-AI-Cloud-revenue ratio, the 76.5% e-commerce share of consolidated revenue, the −415% AI Labs margin, the 25.6% AI share of cloud, the −1.9 Rule of 40 (quarterly revenue growth in dollars plus quarterly free-cash-flow margin), and the roughly ten quarters of funding at this quarter's outflow rate. The "higher pricing of a broad range of chip components" quotation is the release's own. No fair-value figure is quoted here because the segment recut has not been reflected in our model.
