NVIDIA's $96.2 Billion Quarter Reframes AI Around Supply, Not Demand

$96.2 billion in three months. That is the number that matters from NVIDIA's August quarter, a sum large enough to swallow most S&P 500 companies' full-year sales and still leave room for dessert. For the fiscal second quarter ended July 26, 2026, NVIDIA reported revenue of $96.2 billion, up 18% sequentially and 106% from a year ago. ³¹¹ Data Center alone was $89.0 billion, up 117% from the year-ago period. ³¹¹ GAAP and non-GAAP gross margins were both 75.0%. ³¹¹

The report landed after the close on Wednesday, Aug. 26, 2026, released at 4:20 PM ET. ³ The results "more than doubled the year-ago period," and management "felt comfortable enough to provide a financial outlook further into the future than ever before." ² That second clause is almost as important as the first. A company selling at this velocity does not normally volunteer a view eighteen months out unless it wants Wall Street to stop guessing about the ceiling and start arguing about the constraints.

That longer view was startling. Mid-earnings-report, NVIDIA forecast 70% growth for fiscal 2028, the period ending Feb. 1, 2028, "significantly higher than Wall Street analysts were expecting." ⁴ Chief Executive Jensen Huang's quarter was framed in the press as demand accelerating, with Fortune's headline reading "Nvidia doubles Q2 revenue to $96 billion, as CEO Jensen Huang says demand is accelerating." ⁶ CNBC's verdict on the beat was simpler: he "did it again." ²

But the 70% is not a demand forecast. It is a supply forecast. Chief Financial Officer Colette Kress projected the 70% FY2028 growth as supply-constrained, then warned memory costs are rising beyond expectations and "headed even higher." ⁸ Read that pairing carefully. NVIDIA is telling investors it could sell more if it could build more, and at the same time telling them the parts it needs to build are getting more expensive. That is a very different kind of guidance than the old beat-and-raise routine, where upside fell straight to margin. Here upside has a cost.

Supply-constrained has become the most abused phrase in semiconductors, but in this case the company put a number behind it. Supply and capacity commitments jumped to $279 billion, alongside $29 billion in cloud agreements, long-term data center leases and equity investments. ⁷ In plain language, NVIDIA is writing enormous forward checks for memory, packaging and capacity stretching years into the future in order to secure the right to deliver that 70% growth. It is buying visibility at a time when visibility is the scarcest commodity in AI infrastructure.

That also explains why margins are now a debate rather than a coronation. At 75.0% on both a GAAP and non-GAAP basis, gross margin remains a figure most industrial companies would trade a division for. ³¹¹ Yet Kress paired the growth outlook with a warning on memory costs, a signal that the next leg of scaling will be fought in procurement as much as in architecture. ⁸ When your biggest input is headed even higher, every extra point of revenue growth requires more negotiation, more prepayment and more engineering to keep the margin line from bending.

The other half of the constraint is financing — not NVIDIA's financing, but its customers'. NVIDIA "also committed credit support," a disclosure consistent with the idea that it is helping backstop customer purchases of its own systems. ⁷ Portfolio manager Richard Clode at Janus Henderson said NVIDIA's update provided reassurance on exactly that front, among others. The market's question is obvious: if demand is so strong, why does the seller need to support the buyer? The company's answer, implicit in the quarter, is that the projects have grown so large — gigawatts, multi-year leases, power procurement — that even hyperscalers and sovereign-scale builders want risk-sharing to move faster.

Clode's summary captures why the stock debate shifted after the print. Demand remaining strong despite increasing scale was the takeaway, in his reading, with greater clarity on the three things skeptics had seized on: how fast NVIDIA can grow, what it will earn while growing, and what it has promised to make that growth possible. ⁹ That is a fair description of what a $96.2 billion quarter does. It does not settle the AI buildout argument. It narrows it to execution: can memory arrive, can power be connected, can customers pay, and can NVIDIA keep 75% margins while saying yes to all three?

There is reason to take the demand side seriously, even without adding new claims. A Data Center business doing $89.0 billion in a single quarter, up 117% year over year, does not happen because one customer over-ordered. ³¹¹ It happens because compute has become revenue infrastructure — tokens produced and sold, clouds rented, enterprises re-platformed — rather than an experiment budgeted out of research. Huang's characterization of demand as accelerating fits that math better than any story about pull-forward or channel fill. ⁶ You do not grow 18% sequentially at this scale, off a base that was already a record, without broad-based consumption. ³¹¹

The honest limit here is what the filed excerpts do not show. They confirm the headline numbers, the 70% forecast, the $279 billion in commitments, and the supply-constraint and margin-pressure themes. ³⁴⁷⁸¹¹ They do not provide verbatim executive quotes suitable for quotation, nor do they verify in these pages the detailed income-statement profit total, the precise Street-beat arithmetic, the customer mix, or the platform ramp schedule. That absence matters because NVIDIA at this size is really two stories: the audited quarter just printed, and the forward liabilities and supports that make the next four quarters possible.

On the first story, the evidence is clean. Revenue more than doubled, Data Center grew even faster, margins held at 75%, and the company chose to guide further out than ever. ²³¹¹ On the second, the direction is clear but the price is not. Committing hundreds of billions in future supply while also committing credit support is how a toll collector becomes a co-builder. ⁷ It secures growth and shares risk in the same stroke. If memory arrives on time and tokens keep selling, the 70% will look conservative in hindsight. If memory costs keep climbing or power and construction lag, NVIDIA will have paid to discover the ceiling it just promised it could break through.