"Almost no card sits idle." That's Eddie Wu, Alibaba's CEO, reportedly describing—per remarks in an earlier earnings call—the utilization of every AI accelerator inside the company's data centers—and it's the single most important sentence in an earnings report the market couldn't figure out how to read.
Here is the scoreboard from that same report. Cloud revenue up 45%. Net income reportedly down 75%. Capital expenditure up 75%. Shares down roughly 5% shortly after the open.
Four numbers from one document, and they cannot all be good news. The job of this piece is to figure out which one you should actually trust.
Four numbers from one document, and they cannot all be good news. The job is to figure out which one you should actually trust.
Last week, we were still hunting for AI's missing revenue—the thread from 08/14, where the industry kept spending like the demand was real but couldn't produce the receipts. This week, a Chinese hyperscaler finally put the number on the table.
Start with the disclosure that matters, because everything else hangs off it. For the first time in a filing, Alibaba reportedly broke out AI revenue as a share of external cloud: reportedly around 30% to 35%. Not a vibe, not a "fast-growing segment"—a percentage of a line item.
Alongside it: annualized recurring revenue reportedly past RMB 35.8 billion, and triple-digit year-over-year growth for what the CEO reportedly described as the twelfth consecutive quarter. Twelve quarters. That's three years of a business more than doubling, every quarter, without a single deceleration into merely-fast territory.
Citi's analysts flagged the obvious point—this is the first time investors get a clean look at how much Alibaba is spending on AI and how the products are actually progressing. Before this, you were guessing. Now you're not.
But a disclosure is just a claim until demand is priced. So here is the pricing test.
Alibaba reportedly raised prices on its AI computing and storage products by as much as 34%. And orders kept coming. Management's explanation is almost clinical: because AI now completes more complex work for customers, acceptance of higher API prices stayed high, and demand stayed hot. Supply, in their telling, still can't fully meet it.
That is the part the missing-revenue thread was waiting for. Not just growth—growth that survives a price hike. You can fake a revenue line with discounts and land-grab pricing. You cannot fake a 34% increase that the market absorbs while asking for more.
Wu's "almost no card sits idle" is the same evidence from the cost side. Idle GPUs are the tell of speculative buildout—capacity bought on faith, waiting for demand that hasn't arrived. Full utilization means the capex is chasing real orders, not the other way around.
So the demand is real. That's settled. Which makes the profit collapse the interesting problem, not the disqualifying one.
Capex jumped 75% to RMB 67.7 billion—about $10 billion—driven by uneven customer purchase timing, more CPU-compute capacity, and higher prices across a broad range of chip components. Net income fell 75%. Free cash flow, reportedly positive at RMB 18.8 billion a year ago, swung to negative RMB 44.7 billion.
The naive read is that these two facts fight each other: how can demand be booming while the business bleeds? The correct read is that they're the same fact. Verified demand is precisely what forces the margin-destroying spend. You don't pour $10 billion a quarter into infrastructure for demand you're unsure exists—you do it because the cards are full and customers are still knocking.
The bull would stop there and call it a J-curve: eat the capex now, harvest fat software margins later. That's the story every AI capex defender tells.
Then read the 2030 guidance, and the story changes.
Alibaba's own target is reportedly RMB 100 billion in external cloud revenue by 2030—at a 20% gross margin. Sit with that number. Twenty percent. Not 60%, not the 70%-plus that made enterprise software the best business model of the last two decades. Twenty.
That is the quiet admission buried in the guidance. The destination isn't a rerating back to software economics. Even at RMB 100 billion of scale, the company is telling you the gross margin stays thin.
This isn't a crater the business climbs out of. It's the shape of the business. Alibaba is guiding you toward a permanently heavier company—one that looks less like a software vendor and more like a compute utility, where you win on volume and utilization, not on the gorgeous incremental margins software investors were trained to underwrite.
Management does hold out threads of margin relief. Wu notes that self-developed T-head chips still make up a low share of deployment inside Alibaba Cloud, and that as their penetration rises, the improvement to gross margin will be significant. The MaaS business, they say, should have a very positive effect on margins in coming quarters. And Wu expects AI to cross 50% of external cloud revenue in about a year.
Fine. But the 2030 number is the number they chose to anchor to, and 20% is 20%. If the chip substitution and MaaS mix were going to drag the model back toward software margins, that's not the destination they'd publish.
Now hold Alibaba up against Meta, because the mirror is almost too clean.
Meta's quarter: revenue reportedly up 28% to $60.8 billion, costs and expenses up 55% to $42 billion, free cash flow cratering to $784 million from $8.55 billion a year earlier. Same structure as Alibaba—top line growing fast, spending growing faster, cash flow collapsing under the weight of infrastructure.
Steve Eisman called Meta's spending math "astonishing." Zuckerberg's answer rhymes with Wu's: "We are now at a point where our investments in A.I. are accelerating every major part of our core business."
Two companies, two continents, two different starting businesses—e-commerce and social ads—arriving at the identical P&L shape. Demand real, margins cratering, cash flow inverted. That's not a coincidence. That's the trade becoming the global default.
Which reframes the only honest question for anyone who owns these names.
For years, the implicit thesis on cloud-and-AI was a software rerating: buy the growth, wait for the operating leverage, collect the fat margins on the way up. Alibaba just told you, in its own guidance, that's not the trade on offer.
The trade on offer is a utility. Verified, price-inelastic demand, funded by permanent heavy capex, converging on a thin-margin, high-volume compute business. You can absolutely make money owning a utility—but you price it like one, and you underwrite it on utilization and scale, not on a coming margin expansion that management is quietly telling you isn't coming.
That's why the market fell 5% and couldn't decide which number to believe. It was pricing a software company and got handed a utility with a $10 billion quarterly infrastructure bill.
The missing revenue isn't missing anymore. Alibaba found it—reportedly around 30% to 35% of external cloud, ARR past RMB 35.8 billion, triple-digit growth for what the CEO described as twelve straight quarters, all of it surviving a 34% price hike.
The harder truth is what that found revenue is worth. At a 20% gross margin, verified demand doesn't rescue the multiple you were hoping to pay. It just tells you what kind of company you actually own.