Nvidia reported results for its second fiscal quarter, ended 26 July 2026, on 26 August 2026. Revenue was 96.2 billion dollars, up 18 percent on the previous quarter and 106 percent on the same quarter a year earlier. Data centre revenue was 89.0 billion dollars, up 18 percent sequentially and 117 percent year on year. The company’s remaining segment contributed 7.2 billion dollars.

GAAP and non-GAAP gross margins were both 75.0 percent. GAAP operating income was 63.7 billion dollars and GAAP net income 59.7 billion dollars. Jensen Huang’s line in the release was: “AI has reached its inflection point. It’s doing useful work. Its tokens are productive and profitable. Now, compute is revenue.”

Those are the figures that will be quoted for the next week. Three others in the same set of documents are more informative, and none of them is a record.

Why it matters

Nvidia’s quarterly reporting is the closest thing this industry has to a disclosed view of how much compute is actually being bought rather than announced. For anyone pricing an AI product, the useful signal is not the revenue line but the commitments line, which records what the company has contracted to pay suppliers before it has sold anything to anyone. That number moved a great deal further this quarter than revenue did.

The adjusted figure is lower than the statutory one

Nvidia reported GAAP diluted earnings per share of 2.46 dollars and non-GAAP diluted earnings per share of 2.22 dollars. That ordering is the reverse of the usual arrangement. Adjusted figures are normally the higher of the two, because the adjustments customarily strip out costs — stock-based compensation, acquisition-related charges — rather than income.

The explanation is in the release itself. Gains from equity securities, net, came to 7.771 billion dollars for the quarter. GAAP net income of 59.7 billion dollars includes them; non-GAAP net income of 53.954 billion dollars does not. Roughly thirteen percent of the statutory bottom line therefore came from the appreciation of Nvidia’s holdings in other companies rather than from selling hardware to anyone.

This is not an irregularity and Nvidia discloses it plainly. It is, however, the number to keep hold of when the headline profit is quoted back at you, because gains on equity holdings do not repeat the way chip sales do, and a number of those holdings are in companies whose own valuations depend on the continued expansion of the market Nvidia sells into. On this occasion the non-GAAP figure is the more conservative read of the operating business, which is an unusual sentence to be able to write about an adjusted number.

The commitments line grew faster than the revenue line

Nvidia’s supply and capacity commitments — money it has contracted to spend with suppliers — rose from 119 billion dollars at the end of the previous quarter to 279 billion dollars, per the company’s quarterly filing as reported by Fortune, which attributes the increase principally to rising memory costs. That is an increase of about 134 percent in three months, against revenue growth of 18 percent over the same period. The increase alone is roughly 160 billion dollars.

It is worth being exact about what this figure is and is not. It is a contracted commitment to purchase, recorded in the commitments note of a quarterly filing. It is not revenue, it is not an order book, and it is not a measure of customer demand. It is the company’s own future outgoings, which fall due whether or not the demand that justified them materialises. The filing breaks these commitments down further by period and lists additional categories of obligation; the document itself returned an access error to our tooling, so we are not reproducing figures we could not read ourselves.

Read against the revenue line, it says something quite specific. Nvidia is contracting for supply considerably faster than it is growing sales, and the input it is contracting against is memory rather than logic. That is consistent with the direction we have been tracking elsewhere — that the model race has become a systems race, in which the binding constraint keeps relocating.

The outlook assumes nothing at all from China

Guidance for the third fiscal quarter is revenue of 108.0 billion dollars, plus or minus 2 percent, with GAAP and non-GAAP gross margins of 74.0 percent, plus or minus 50 basis points, and operating expenses of approximately 9.2 billion and 9.0 billion dollars respectively. This is guidance. It is a company forecast rather than a result, and it should not be recirculated as one, which is a distinction that tends to dissolve within about a day of any earnings release.

The clause worth reading twice is Nvidia’s own: “NVIDIA is not assuming any Data Center compute revenue from China in its outlook.” A 108 billion dollar quarter is being guided with one of the largest markets for this hardware modelled at zero. That is either conservatism with substantial upside attached to it, or an accurate reflection of policy conditions the company does not expect to change. Nvidia does not say which, and we are not going to guess on its behalf.

Separately, Fortune reports the company projecting roughly 70 percent growth for fiscal 2028, implying revenue somewhere in the range of 690 to 700 billion dollars. That is a projection, about a year that has not started, reaching us at one remove. We mention it only because it is the figure most likely to circulate this week stripped of both qualifications.

What this means for buyers

Do not budget for compute prices falling because supply loosens. A supplier that has just contracted 279 billion dollars against rising memory costs is not a supplier preparing to pass savings downstream. If your AI cost model assumed hardware deflation would eventually arrive and rescue your unit economics, that assumption now has a specific number arguing against it.

The price war in model APIs has been running on margin compression and on efficiency gains at the model layer, not on cheaper silicon. Nothing in this quarter changes that, and the memory-cost pressure points the other way. Expect further price movement to come from architectural efficiency and from competition between labs, rather than from Nvidia’s own cost base easing.

For anyone assessing alternative accelerators — whether specialised inference silicon or the domestic Chinese chips now reported to be serving frontier models at scale — the economic case turns on the same memory constraint that Nvidia has just added roughly 160 billion dollars of commitments to address. That is the variable worth tracking, rather than headline accelerator performance.

What would change our reading

The commitments figure over the next two quarters. A commitment line that keeps climbing while revenue growth decelerates is the shape that precedes a write-down. One that flattens while revenue keeps growing is the shape of a company that read demand correctly. A single quarter establishes the direction and tells us nothing whatever about the destination, and we will not pretend otherwise until there is a second data point.

The equity-gains line. If gains from equity securities keep contributing a tenth or more of statutory net income while Nvidia continues investing in customers who buy its chips, the question its own finance chief pre-empted becomes harder to set aside. Colette Kress addressed it directly, per Fortune: “We recognize the scale of this support, and we know some will call this circular financing. We see it differently.” We will keep quoting both the GAAP and the non-GAAP figures rather than picking whichever suits a narrative.

And the China assumption. If data centre compute revenue from China reappears in a future outlook, this quarter’s guidance becomes a floor rather than an estimate, and the beat that follows will be a policy event rather than a demand signal. If it stays at zero across several quarters, it stops being conservatism and becomes a structural fact about the addressable market. Either way, the assumption is doing more work in that 108 billion dollar figure than most coverage of it will acknowledge.

Sources

  • NVIDIA, official announcement of second-quarter fiscal 2027 financial results, 26 August 2026 — nvidianews.nvidia.com
  • StockTitan, verbatim reproduction of the same NVIDIA press release, used to confirm net income, the equity-securities gains line and the exact outlook wording — stocktitan.net
  • NVIDIA investor relations, CFO commentary on second-quarter fiscal 2027 results (PDF, not machine-readable to our tooling) — s201.q4cdn.com
  • NVIDIA, Form 10-Q for the quarter ended 26 July 2026, which carries the commitments note (EDGAR returned an access error to our tooling) — sec.gov
  • Fortune, named reporting on the commitments increase, the fiscal 2028 growth projection and the finance chief’s remarks on circular financing — fortune.com