NVIDIA's revenue doubled, and its receivables grew by $22 billion

NVIDIA reported results for its second fiscal quarter on 26 August, and the top line is the number that will get quoted: $96.2 billion in revenue, up 106% from the same quarter a year ago, with $89.0 billion of it from the data center segment. Guidance for the current quarter is $108.0 billion, plus or minus 2%.

Those figures deserve to be taken at face value. They are reported quarterly results in a securities filing, not a vendor benchmark, and they carry legal liability that marketing claims do not. What is more informative for anyone trying to understand where AI compute is heading is the second layer of the same disclosures: how much NVIDIA has now committed to spend, how much of the quarter's revenue has been booked but not yet collected, and how much outside money is being organized to pay for what comes next.

The top line

• Revenue of $96.22 billion, up 18% from the previous quarter and 106% year over year
• Data Center revenue of $89.0 billion, up 117% year over year
• Edge Computing revenue of $7.2 billion, up 27% year over year
• GAAP and non-GAAP gross margins both 75.0%, against 72.4% and 72.5% a year ago
• GAAP earnings of $2.46 per diluted share; non-GAAP $2.22
• Third-quarter guidance of $108.0 billion plus or minus 2%, with gross margin guided down to 74.0%

One oddity is worth a sentence. GAAP net income of $59.7 billion came in above non-GAAP net income of $54.0 billion, which is the reverse of the usual pattern where the adjusted figure is the flattering one. The difference sits in other income: $7.8 billion for the quarter, of which $7.77 billion was gains on equity securities. Those gains are investment returns rather than chip sales, and NVIDIA reports them separately from operations for that reason.

How the quarter was financed

Revenue grew 18% from the previous quarter. Cash from operations moved the other way: $24.1 billion, down from $50.3 billion in the first quarter. Free cash flow was $21.3 billion, against $48.6 billion three months earlier.

The cash flow statement explains most of it. Accounts receivable consumed $22.3 billion of cash during the quarter, and the balance sheet now carries $63.1 billion of receivables against $38.5 billion at the end of the last fiscal year. Inventories rose to $31.6 billion. In plain terms, a large and growing share of what NVIDIA books as revenue has been shipped and invoiced but not yet paid for.

That is not alarming on its own — receivables grow with sales at any company scaling this fast — but they grew faster than sales did, and the composition of the customer base is what gives it weight. Revenue remains concentrated: one direct customer accounted for 16% of total revenue in the quarter, and the 10-Q notes that a single AI research and deployment company contributed a meaningful share indirectly, by buying cloud services from NVIDIA's customers.

NVIDIA also borrowed, substantially. The quarter brought $24.9 billion of net proceeds from debt issuance, and the 10-Q lists $33.5 billion of senior notes outstanding as of 26 July, across new issues maturing between 2028 and 2046. A company producing this much operating cash raising this much debt is making a deliberate choice about capital structure rather than covering a shortfall — it also repurchased $19.7 billion of stock and paid $6.0 billion in dividends in the same three months.

$279 billion committed, $500 billion being arranged

The 10-Q holds the number that says most about NVIDIA's own conviction. Supply commitments rose from $119 billion last quarter to $279 billion as of 26 July — obligations for data center infrastructure systems, the company says, "primarily memory and manufacturing facilities," covering current and future product architectures. That is a contractual bet on demand several years out, placed with NVIDIA's own balance sheet.

Alongside it, the release announced strategic partnerships with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to establish independent compute financing platforms, intended to mobilize more than $500 billion of third-party capital for AI infrastructure over time. The release adds "subject to definitive agreements," and that qualifier is the part to hold on to: this is a stated intention with named counterparties, not capital that has moved.

Read together, the three facts describe one pattern. NVIDIA is committing its own capital to lock in supply, extending more credit to customers through receivables, and helping assemble outside capital so those customers can keep buying. That is a coherent strategy in a market where the binding constraint is capacity rather than demand. It also means more of the buildout now depends on financing conditions holding, not only on whether the models keep getting better.

On the same day, TechCrunch reported that Amazon has ordered two million additional NVIDIA GPUs for AWS data centers deploying across 2027 and 2028, roughly five months after committing to one million. NVIDIA's own release did not detail that arrangement.

China is out of the forecast

The $108.0 billion guidance explicitly assumes no data center compute revenue from China. The 10-Q shows why that is now closer to a rounding decision than a dramatic one: NVIDIA took a $0.4 billion charge in the first half of the fiscal year on excess H200 inventory and purchase obligations as demand for the part fell away, and shipments made under the US government's H200 licensing program amounted to less than 1% of data center revenue in the most recent quarter. The China business has already contracted to the point where excluding it changes the forecast very little.

What changes for people building on these models

Nothing here moves an API price this week. Over a longer horizon, the supply commitment is the line that matters most to anyone budgeting for inference. $279 billion contracted against memory and manufacturing capacity is the clearest available indication that serving capacity keeps expanding — and expanding serving capacity is precisely what has made long context windows, agent loops that fire many model calls per task, and steadily falling per-token prices economically possible.

The product notes point the same direction. NVIDIA said the Vera Rubin platform is ramping into full production with racks running at CoreWeave, Google Cloud, Microsoft Azure, Oracle Cloud Infrastructure and Nebius; that the Groq 3 LPX inference accelerator Promptea covered on Monday is now in full production; and it introduced NVIDIA Vera, which it describes as the first CPU built for AI agents. The serving stack is visibly being rearranged around inference and agent workloads rather than training alone.

The caution is the other side of the same coin. If financing conditions tighten, the first party to slow down is not NVIDIA — it is the customers whose buildouts depend on capital being arranged on their behalf. Per-token prices have fallen consistently because supply has run ahead of demand at each step. That trend is now contracted for, several years out. It is not guaranteed.

AI has reached its inflection point. It's doing useful work. Its tokens are productive and profitable. Now, compute is revenue.

Huang's framing is a sales pitch and reads like one. But "compute is revenue" is also a fair description of what the filings show: a quarter in which NVIDIA sold $96 billion of it, committed $279 billion to producing more, and began helping its customers find the money to pay for it.