"Now almost no card in Alibaba's servers sits idle."
That's Wu Yongming on the May 13 earnings call, explaining why Alibaba just printed the number US hyperscalers keep gesturing at but never quite naming: RMB35.8 billion in annualized AI product revenue, RMB89.71 billion for the quarter, an eleventh straight triple-digit-growth quarter.
And then, in the same filing, Alibaba printed the price of that number: free cash flow swung to a negative RMB17.3 billion, and adjusted EBITA fell 84%.
Last week the open question in this space was whether any hyperscaler—anywhere—could show inference monetization actually catching up to capex, or whether the whole industry was still selling futures. This week we have an answer, and it comes from China, not the US.
The answer is yes for one of them. Not yet for another. And a third isn't really answering the question at all.
Start with the concession, because it's real. These are hard revenue figures with disclosed mix ratios and dollar-denominated ARR—the kind of specificity Alphabet and Amazon still fold into "cloud growth" without breaking out. The Chinese Big Three put actual monetization numbers on the page. That's the setup, and it's genuine.
But hard numbers are not interchangeable. Read closely, the three disclosures measure three different things, and only one of them is what the headline says it is.
Take Baidu first, because it's the cleanest. Core AI-powered Business revenue hit RMB13.6 billion in Q1, up 49% year over year, and crossed 52% of Baidu General Business for the first time—up from 36% a year ago.
That mix crossing is the tell. A revenue line only overtakes the rest of a business when it's actually replacing it, not when it's being subsidized alongside it.
Underneath, AI Cloud Infra grew 79% to RMB8.8 billion, with GPU Cloud up 184%. And CFO Haijian He noted GPU cloud is structurally higher-margin than CPU cloud—stronger demand, tighter supply, higher barriers, better pricing power.
Crucially, Baidu's operating cash flow stayed positive at RMB2.7 billion. The segment table is drawn from internal management accounts, not audited—but it's laid out line by line, quarter over quarter, in a way you can actually check the mix against.
So Baidu is booking AI revenue while generating cash. That is monetization catching capex, delivered, not promised. It is the proof-point.
Baidu is booking AI revenue while generating cash. That is monetization catching capex, delivered, not promised. It is the proof-point.
Now Alibaba, where the same vocabulary describes something structurally different.
The RMB35.8 billion figure is annualized recurring revenue—an ARR run-rate—and the RMB89.71 billion quarterly line is AI-related product revenue sitting at 30% of external cloud revenue. Both are real. Both are also volume metrics: MaaS and inference throughput, priced per token.
Here's the coexisting reality. That RMB89.71 billion sits beside a negative RMB17.3 billion free-cash-flow quarter and an adjusted EBITA down 84%, to RMB5.1 billion. Group operating income flipped from a RMB28.5 billion profit a year ago to an RMB848 million loss.
The "All Others" segment alone posted a RMB21.1 billion operating loss, driven by user acquisition for the Qianwen app—which hit 166 million monthly actives but has no disclosed commercialization timeline. That's a mobile-internet subsidy playbook applied to compute-intensive infrastructure.
So Alibaba is booking AI revenue faster than almost anyone, and burning cash faster still. The revenue is accelerating; the money is leaving faster. That is not delivered monetization. That's a promissory note.
And the note has terms. Alibaba projects AI model-and-application ARR to cross RMB10 billion this June quarter and RMB30 billion by year-end—up more than tenfold from November. It projects AI to cross 50% of external cloud revenue within a year. All forward.
Which brings us to Wu Yongming's framing, the honest tell in the whole quarter. "AI is more like a manufacturing industry," he said. "To get more revenue, you must build two core factories—an AI training factory and an AI inference factory."
This is a better story than the hyperscalers tell, and it's worth understanding why. It reclassifies open-ended capex as factory capacity. Manufacturing capacity self-certifies: you can map plant to future output to revenue with some precision. "Burning cash for scale" on the internet model often never pencils out.
That's the point of "no card sits idle"—it's the utilization claim that turns spend into an asset instead of a loss. And Wu went further: capex will run "far beyond" the announced RMB380 billion, toward a compute base "ten times" the pre-boom 2022 level, chasing US$100 billion in combined cloud-and-AI revenue over five years.
But notice what the manufacturing frame requires to be true. It requires the RMB30 billion year-end ARR to land, the compute to stay fully utilized, and demand to keep outrunning supply so pricing holds—Alibaba raised AI compute prices 5% to 34% in April.
If those hold, RMB380 billion-plus isn't burning money; it's building product. If any slips—utilization dips, the ARR curve bends, pricing softens—then the same spend is exactly the open-ended capex yield-seeking investors are fleeing. The claim is unproven and priced on the bull outcome.
That's why Baidu's number reads as delivered and Alibaba's reads as a promissory note denominated in RMB rather than dollars.
Which leaves Tencent, whose "9% revenue growth on AI" is the most misread of the three—because it isn't a growth story at all.
Tencent's Q1 revenue was RMB196.4 billion, up 9%; Non-IFRS operating profit also grew 9%. But strip out the new AI products—Hunyuan, Yuanbao, CodeBuddy, WorkBuddy, QClaw—and Non-IFRS operating profit grew 17%, to RMB84.4 billion.
Read that again. The AI products didn't add 9%; they subtracted roughly eight points of operating-profit growth. Analysis puts the net AI cost Tencent absorbed this quarter at about RMB8.8 billion.
That's the actual shape of it: RMB8.8 billion of AI cost absorbed by a gaming-and-WeChat cash fortress that generated RMB56.7 billion in free cash flow. Tencent isn't monetizing AI at scale—Ma Huateng's own line is that the core business "provides ample cash flow support" for AI investment. The AI line is a cost the fortress can eat, dressed up as a growth line.
None of this means Alibaba is wrong and Baidu is right in some permanent sense. Alibaba's manufacturing bet may well pencil out; a full year from now the ARR and mix targets could vindicate the burn.
But investors should read the disclosures for what they measure today. Baidu has shown inference monetization catching capex, with positive cash flow and a 52% mix crossing to prove it. Alibaba has shown enormous inference volume and an enormous cash outflow to fund it, with the returns booked as a forecast. Tencent has shown it can afford the experiment.
The market is already pricing the distinction, if crudely—Alibaba's Hong Kong shares are down roughly 7% on the year, Tencent down 23%, while US AI names ran up over 15%.
The one number that will settle the argument is Alibaba's year-end ARR. Hit RMB30 billion with utilization intact, and the promissory note clears at par. Miss it, and "no card sits idle" becomes the most expensive sentence of the quarter.