While most of Silicon Valley’s giants claim that their big ideas came while dwelling in basements or tinkering in garages, a diner on Berryessa Road in San Jose birthed a tech titan over a full stack of flapjacks.
Jensen Huang founded Nvidia in a Denny’s booth in 1993, believing specialized 3D graphics accelerators would be the future of gaming. He was right, and the company became synonymous with the future of artificial intelligence. Huang still runs Nvidia, and for the past year, a growing chorus has argued that the boom he sits at the center of is inflated.
On Wednesday, after market close, Huang addressed that concern with a number: $96.2 billion in sales in three months, up 106 percent from a year ago, beating Wall Street estimates. Then came the guidance. Huang told investors to expect $108 billion in the current quarter, against Wall Street’s $104.2 billion. For the full fiscal year ahead, Nvidia projects roughly 70 percent revenue growth, against an analyst consensus of 44 percent. The stock rose about 8 percent overnight.
‘Compute Is Revenue’
Nvidia’s quarterly report has become a checkpoint for the entire A.I. economy, because nearly every dollar spent building A.I. passes through Huang’s graphics processing unit (GPU) company along the way. Every Nvidia chip that runs earns money for whoever is running it.
“A.I. has reached its inflection point. It’s doing useful work. Its tokens are productive and profitable,” Huang explained on Wednesday’s call. “Now, compute is revenue.”
The quarter’s numbers are evidence. Data-center sales hit $89 billion, up 117 percent—roughly 93 percent of everything Nvidia sold—and Nvidia kept 75 cents of every sales dollar as gross profit. Total profit was $59.7 billion, but $7.8 billion of that (about 13 percent) was paper gain on Nvidia’s equity stakes in other companies, not money earned selling chips.
‘The Unconstrained Would Be a Lot Higher’
Seventy percent growth by this time next year means some $700 billion in a single year. By CNBC’s math, that is enough to pass Apple and Alphabet in revenue, trailing only Amazon. And that it’s the conservative version. “Even though our demand is much greater than 70 percent, our supply allows us to confidently deliver 70 percent,” Huang said. “The unconstrained would be a lot higher.”
The cap is the supply chain. “Memory scarcity today is being driven in large part by the A.I. build-out itself,” said Colette Kress, Nvidia’s chief financial officer. Margins will dip to roughly 71 or 72 cents on the dollar this winter before they recover. Huang can promise the rest because the order book is already full; Kress counted a “cloud industry backlog now greater than $2 trillion.”
Vera Rubin, the next generation of Nvidia chips, is in full production and slated to account for a fifth of next quarter’s data-center sales. The five biggest cloud companies plan to spend nearly $800 billion this year and $1.3 trillion next. Amazon alone signed up for 2 million more chips. China sits at zero in the forecast, where export controls have left it; anything from there is a bonus.
‘We See it Differently’
Those orders are exactly what Nvidia’s skeptics point to, citing “circular financing.” Nvidia puts money into customers like OpenAI, the customers buy Nvidia chips, and the loop, critics argue, makes demand look bigger than it is. The $7.8 billion in paper gains on those stakes is what that loop looks like on the income statement. “At some point, the money runs out,” Boston College’s Aleksandar Tomic told Al Jazeera in July.
A second objection is about accounting rather than financing. Michael Burry of The Big Short fame posted on X in November 2025 that the cloud giants are “understating depreciation by $176 billion” through 2028 by stretching how long chips are supposed to last, which inflates their profits. If Burry is correct, it means Nvidia’s biggest customers are buying on economics that don’t hold up. A third is about the technology itself. MIT’s Daron Acemoglu, a Nobel laureate, is dubious of the whole decade; his numbers put A.I.’s contribution to total factor productivity at less than 1 percent over ten years.
Kress sidesteps the equity critique and instead reframes the circular financing issue around Nvidia cloud partnerships. “We know some will call this circular financing. We see it differently,” she said Wednesday, pointing to agreements where Nvidia provides hardware in exchange for a recurring cut of cloud compute rentals rather than extending credit. “We’re not making loans. In this model, we get paid twice—once on the hardware sale, and again through the share of rental revenue.”
As for the bigger doubt—the A.I. bubble of it all—Huang’s answer is payback speed. A.I. data centers, he said during the call, now recoup their costs in under a year. Huang did not detail the utilization or pricing assumptions behind that figure.
Thirty-three years ago, Huang and two other engineers (Chris Malachowsky, who is still fully involved with the company, and Curtis Priem) talked chips over coffee in East San Jose. The Denny’s booth now wears a plaque that says “the booth that launched a trillion-dollar company.” NVIDIA crossed the five-trillion-dollar market-value threshold last October, becoming the first company in the world to do so. As of Wednesday night, the plaque undersells the company by four trillion.
