Alibaba reported its June-quarter results on Thursday, and the two headline numbers point in opposite directions. Revenue rose 9% year over year to 268.95 billion yuan — about $39.64 billion — the fastest top-line growth the company has managed in years. Net income fell more than 75%, to 10.54 billion yuan.
The gap is not a demand problem. It is a spending decision. Capital expenditure in the quarter came to 67.68 billion yuan, up 75% from a year earlier, almost all of it going into AI infrastructure and model development. Free cash flow swung to an outflow of 44.67 billion yuan, against an 18.82 billion outflow in the same quarter last year. US-listed shares fell around 5% on the print.
What the money is buying shows up in one line. Revenue from AI Cloud and Compute Services rose 45% to 48.44 billion yuan. Alibaba Cloud’s external revenue growth — the number that strips out selling to itself — also accelerated to 45%. AI-related product revenue reached 12.38 billion yuan and has now grown at triple-digit rates for twelve consecutive quarters.
The adjusted numbers still fell
It would be easy to write the profit collapse off as depreciation and one-off charges. The adjusted figures argue otherwise. Non-GAAP net income fell 38% to 20.72 billion yuan. Adjusted EBITA fell 30% to 27.33 billion. Both of those measures are designed to look past exactly the sort of accounting noise a build-out generates, and both still went backwards by double digits.
That tells you the cost is landing in the operating line now, not just in future depreciation schedules. Chief executive Eddie Wu framed the quarter around “the improving commercialization of our full-stack AI capabilities” and pointed at the 45% cloud acceleration. He is not wrong about the direction. He is asking investors to fund the gap between a 45% growth rate and a 9% one for several more quarters.
Our take: This is the clearest version yet of the trade every hyperscaler is making, and Alibaba is making it with less margin cushion than its US peers. A 45% cloud growth rate is genuinely rare. But 9% group revenue growth cannot carry a 75% capex increase indefinitely — the arithmetic only works if cloud keeps compounding at something close to its current rate long enough to become the majority of the business. Watch the external cloud growth rate next quarter. If it holds near 45%, the spending was disciplined. If it slips toward 30% while capex stays elevated, this stops being an investment story and becomes a margin story.
The China angle
Alibaba is spending into a domestic market where compute is constrained by export controls, not just by budget. That makes the capex number harder to read than an equivalent figure from a US hyperscaler: some of it is capacity, and some of it is paying up for scarce supply. It also makes the twelve-quarter triple-digit streak in AI product revenue more meaningful — that demand is showing up despite the constraint, not because compute is cheap.
The other thing worth holding in view is that Alibaba is running two AI businesses at once. It sells cloud capacity, and it ships its own frontier-class Qwen models, increasingly with open weights. The second one is a customer-acquisition cost for the first. Model releases that do not obviously pay for themselves start to make sense when you read them as demand generation for 48 billion yuan of quarterly cloud revenue.
What to watch
- External cloud growth next quarter. 45% is the number that justifies everything else. Deceleration is the signal, not the absolute level.
- Whether capex growth moderates. A second consecutive 75% increase would mean the build-out is still front-loaded; a flattening would suggest capacity is catching up to demand.
- Free cash flow. A 44.67 billion yuan outflow is manageable once. The question is how many quarters of it the balance sheet is being asked to absorb.
- Adjusted EBITA. If it stops falling while capex stays high, the commerce business is covering the bill. If it keeps falling, it isn’t.
Alibaba just told the market it would rather have the cloud business in three years than the earnings today. That is a defensible choice. It is also the same bet a dozen other balance sheets are making at the same time, and the returns will not arrive for all of them.