Nvidia Reports $96.2 Billion in Quarterly Revenue, Forecasts $108 Billion for the Current Quarter

$96.2 Billion In, $108 Billion Promised, and a Number That Changed Three Times in One Night

Nvidia's stock did something strange on Tuesday, August 26. It fell. Then it rose. By Thursday, three different news outlets were reporting three different numbers for the same move — 7.63%, 6%, and 4.2% — and all three were technically right^[3][4][12]. Bloomberg's first headline, written in the minutes right after the earnings release, said the forecast "fails to wow investors"^[5]. An hour later, that headline was already wrong.

What actually happened is simple enough. Nvidia reported $96.2 billion in revenue for the quarter that ended July 26, 2026. That's up 106% from $46.7 billion a year earlier — the company's sales more than doubled in twelve months^[1]. Data-center revenue, almost all of it AI chips, hit $89 billion, up 117%^[4]. Net income came in around $59.7 billion^[1]. For the next quarter, Nvidia guided to $108 billion, above the roughly $104.2 billion Wall Street expected^[1][6].

Then, on the analyst call, chief financial officer Colette Kress did something Nvidia has never done before. She gave a forecast for the year after next. Nvidia expects about 70% revenue growth in fiscal 2028, she said — against the roughly 44% analysts had modeled^[3][15]. That's the number that moved the stock. The initial earnings dip happened before anyone heard it.

Two Sentences That Are Both True and Point in Different Directions

Here's the tension at the center of this story. Nvidia sold $96.2 billion of real chips in three months, and customers paid real money for them. At the same time, Nvidia itself has become one of the biggest sources of financing for the companies buying those chips.

Both facts came from Nvidia's own disclosures. The company has roughly $750 billion in announced AI deals, a $30 billion equity stake in OpenAI, and a guarantee of up to $105 billion tied to an OpenAI data-center campus in Pike County, Ohio^[9][10][16]. That guarantee was reportedly cut down from about $250 billion earlier this year, after some investors raised concerns about the size of the exposure^[10][16].

This is the mechanism behind the term "circular financing," and it's worth walking through slowly because the whole debate hangs on it. Nvidia invests in, lends to, or backs the debt of a company. That company uses some of that money to buy Nvidia chips. Nvidia then books the purchase as revenue. On the surface, it looks like outside demand paid for those chips. In practice, some of the money may have made a round trip through Nvidia's own balance sheet before coming back as a sale.

That doesn't mean the chips are fake or the money isn't real. It means a dollar of Nvidia's reported revenue is harder to trace back to an end user actually making money from AI, versus a dollar that came from capital raised on the bet that AI will be profitable later — some of which Nvidia itself supplied. CNBC's Jim Cramer put it directly to CEO Jensen Huang, asking whether the OpenAI investment was buying loyalty or financing a future rival^[10].

The Case That Nvidia Doesn't Need to Explain Itself

Huang's answer, delivered both in the earnings release and in interviews around it, is that the causation runs the other way. He argues Nvidia invests in AI companies because it sees real opportunity, not to manufacture its own customers. He called the OpenAI-scale investments a "once-in-a-generation opportunity" and said he wished he'd invested more, sooner^[6][10].

The earnings release itself opens with a bigger claim than a sales number. "AI has reached its inflection point," Huang said. "Its tokens are productive and profitable. Now, compute is revenue^[1]." A token is the basic unit of text or data an AI model processes. His argument is that customers are now charging money for what those tokens produce, which turns a chip purchase into a revenue decision rather than a research bet.

Bullish analysts add a structural argument on top of that. Nvidia faces no serious competitor at its layer of the AI stack — the chips themselves and the CUDA software that runs on them — even though the software and app layers built on top of those chips are crowded and competitive^[6]. The comparison analysts reach for is the gold rush: whoever sells the shovels gets paid whether or not any individual miner strikes gold^[13]. Under that view, a company facing weak demand doesn't hand out a forecast 26 percentage points above what Wall Street expected^[3][15]. Nvidia also disclosed a margin squeeze on itself in the same release, which bulls read as evidence it isn't hiding bad news.

The Cost the Company Volunteered

That margin squeeze is real, and Nvidia reported it without being asked. Gross margin — the share of each sales dollar left after the direct cost of making the product — was 75.0% this quarter. Nvidia guided that down to about 74% next quarter, and as low as 71% to 72% by the fiscal fourth quarter^[1][11]. The reason is that memory chip prices, a key input in every AI accelerator, are rising faster than Nvidia expected^[1][11].

That means booming demand and shrinking profit per chip can both be true in the same quarter. Revenue is climbing, but each dollar of it now carries a slightly thinner slice of profit than it did a year ago. It's a detail that cuts against a simple everything-is-great narrative, and it came from Nvidia's own guidance, not from a critic.

The Market Nvidia Is Already Writing Off

There's a second constraint that both Washington and Beijing are shaping, for opposite reasons. Nvidia shipped its first H200 chips to China this quarter under a U.S. licensing arrangement that lets approved Chinese buyers purchase the chips, with 25% of that revenue going to the U.S. Treasury and shipments routed through U.S. territory for screening^[7].

Those China sales came to less than 1% of Nvidia's $89 billion in data-center revenue. Nvidia's forward guidance assumes zero China data-center compute revenue going forward — not a shrinking number, but a line item the company has essentially written to zero^[7]. Washington wants the chips restricted, not banned outright, to slow China's AI progress without killing a major U.S. exporter's business. Beijing, meanwhile, has reportedly limited how many chips it lets in, so that domestic suppliers, mainly Huawei, can grow their own market share without being outcompeted^[7]. One estimate from Bernstein puts Nvidia's share of China's AI chip market falling to about 8% in 2026, down from nearly 40% a year earlier, with Huawei's share climbing above 50%^[7].

Where the Skeptics Have Been Wrong Before, and Where They Haven't

Reporting from NPR, CNN, ABC News and others has kept the bubble question open even after a record quarter, pointing to Nvidia's roughly $5.1 trillion market value as a concentration of risk sitting in a single stock^[6][9][14]. Their concern isn't whether the chips shipped. It's whether the profits AI is generating for end users have caught up to the hundreds of billions being spent building the infrastructure underneath it^[14][17].

That argument has a real track record problem: skeptics have called an AI bubble before, and the stock kept climbing anyway, including this week^[3]. But the specific mechanism they're pointing to — a chipmaker financing its own customers — is one Nvidia hasn't fully rebutted so much as reframed. Huang's position is that his company sees genuine opportunity where critics see risk; "from our vantage point," he said, "we see something very different^[6]." Coverage split largely along those lines too. Fortune called the year-ahead forecast a "bombshell meant to silence AI bubble critics"^[8], while NPR led its own coverage with the $750 billion figure and the word "bubble" in the headline^[9]. Reuters captured both sides of the stock's move in one line — "Nvidia stock jumps after early dip" — while still calling the broader mood "AI fever," a diagnosis dressed as a fact^[4].

None of that resolves the underlying question, and nothing in Tuesday's numbers was built to resolve it. Nvidia will report again next quarter, against a forecast it set for itself at $108 billion. Whether that number holds, and whether it looks more like demand or like financing in hindsight, is still an open account.