Nvidia Booked $96.2 Billion in a Single Quarter. Weeks Earlier, It Quietly Shelved the Program Built to Bankroll Smaller AI Clouds.

Nvidia reported $96.2 billion in revenue for the quarter that ended July 26, a 106 percent increase from the same period a year earlier and roughly $4 billion more than Wall Street had modeled. The data center segment, which sells the accelerators that nearly every large AI system in the world runs on, accounted for $89 billion of it. Adjusted earnings came in at $2.22 a share against expectations closer to $2.09.

Chief executive Jensen Huang framed the quarter as the moment the AI build-out stopped being speculative. "AI has reached its inflection point," he said on the earnings call. "It's doing useful work. Its tokens are productive and profitable. Now, compute is revenue. And demand is accelerating." The company took the unusual step of guiding beyond the next quarter, telling investors to expect roughly 70 percent revenue growth in fiscal 2028 — well above the 44 percent analysts had penciled in. Shares, which had slipped in the first minutes after the release, turned and gained more than 5 percent during the call.

Inside the data center number is a split that matters for the rest of the industry. Roughly $48.7 billion came from hyperscalers — the handful of companies large enough to buy chips by the hundred thousand. The remaining $40.3 billion came from what Nvidia groups as AI clouds, industrial and enterprise customers, a category that includes the smaller specialist providers that rent out GPU capacity and that have to borrow heavily to buy the hardware in the first place.

That financing problem is what Nvidia had proposed to solve in July with something it called the AI Compute Partnership: the company would provide credit support to smaller cloud providers buying its chips, and take a share of the rental revenue those chips later earned. It gave Nvidia two bites of the same apple — margin on the silicon, plus a cut of the income stream the silicon generated.

The Wall Street Journal reported this week that Nvidia has paused deals under the program, fewer than two months after announcing it. According to that reporting, some Nvidia employees warned current and prospective customers that the structure could attract antitrust scrutiny, and several would-be partners were irritated by how much control the company insisted on — including a requirement that providers run their own customers through a Nvidia approval process before capacity could be sold.

The pause matters more to the borrowers than to Nvidia. The largest buyers do not need help financing chips. The smaller providers that emerged over the past two years to rent GPU capacity built their business plans around cheap access to hardware, and a credit backstop from the manufacturer was the cheapest access available. Removing it does not slow the hyperscale build-out at all; it thins out the layer underneath.

Nvidia also told investors it is planning without any revenue from China, where export controls have effectively closed the market for its most advanced parts. Management flagged that gross margin will compress toward 71 to 72 percent by the fourth quarter of the fiscal year, down from about 75 percent, driven largely by the rising cost of high-bandwidth memory — the one component in an AI accelerator that Nvidia does not make itself and cannot simply price around.