For the 16th consecutive quarter, Nvidia reported revenue that topped Wall Street estimates.
Sales in the three months to July 26 were $96.22 billion, up 106% YoY. So, revenue more than doubled versus the same period in 2025. It was the first triple-digit gain in two years.
The guide was impressive on any standard definition of the term. Nvidia sees current-quarter sales of between $106 billion and $110 billion. Taking the midpoint, that’d constitute sales growth of almost 90% in Q3.
Although the forecast was better than expected (analysts were collectively looking for $105 billion), the loftiest estimates from Nvidia bulls approached $140 billion.
Data center revenue rose 117% to $89.02 billion, about $3 billion ahead of consensus. The hyper-scalers accounted for nearly 55% of that, down from 59% in the same period a year ago.
Recall that Nvidia now splits its numbers into two “market platforms.” The data center platform comprises hyper-scaler revenue, which rose 102% YoY, and sales associated with AI clouds, industrial and enterprise applications, which jumped 138%.
The other platform’s “edge computing.” It includes revenue from data processing devices for agentic and physical AI including PCs, game consoles, robotics and automotive. Revenue there was $7.2 billion last quarter, up 27%.
Vera Rubin — which is shipping now — is in “full production,” Jensen Huang said, employing his trademark hyperbole and self-aggrandizing to talk up not just his company’s results, but the AI revolution itself, which has “reached its inflection point.”
“Now, compute is revenue,” Huang said, on the way to declaring “a golden age of new AI labs and startups.” Here’s hoping this “golden age” goes better than the one Donald Trump declared upon retaking the Oval Office 19 months ago.
Nvidia’s facing renewed skepticism around what some critics say are hopelessly circular deals reminiscent of dot-com-style vendor financing. Bloomberg this month reported on three-quarters of a trillion in such arrangements. Subsequently, Huang took the unusual step of announcing $500 billion in partnerships with a hodgepodge of Wall Street and private equity firms before any deals were actually signed in a bid to reassure the company’s clients that their financing needs will be met.
In the CFO commentary accompanying Wednesday’s report, Nvidia said some AI clouds and model makers “are growing faster than their balance sheets and long-term credit profiles can support.” Nvidia’s addressing the issue with “arrangements that help select customers secure the land, power and data center capacity needed to support their growth.”
As to agreements involving the purchase of Nvidia’s own hardware, the company said its various tie-ups will accrue to Nvidia “on the upfront sale of our infrastructure” and also via a share of the revenue earned by AI cloud companies from third-party customers when “certain criteria are met.”
Nvidia put its maximum exposure from all such agreements at $3.5 billion. “We have land, power and shell guarantees for certain AI cloud partners’ data center lease obligations in the event of their default,” the company said. Cash and equivalents were $56.6 billion at quarter-end, up YoY and QoQ.
On the call, CFO Colette Kress said sales will probably rise by around 70% during the company’s next fiscal year despite supply bottlenecks. Suffice to say Colette buried the lede. Analysts were expecting top-line growth of less than 50% next year. That’s the news.
Kress was diplomatic in addressing the circular financing concerns mentioned above. “We see it differently,” she said. “We believe these investments measured against the strength of their demand, the business they create for us, the ecosystem they build on Nvidia’s platform, and the equity returns on our invested capital will be excellent and our risk is limited.”
Nvidia came into Wednesday’s report trading at just 19 times forward earnings, the cheapest in almost eight years.
Jenson Huang obviously doesn’t have to worry about his legacy at this point, but Nvidia’s 2027 forecast strikes me as one of those calls we end up looking back on and wonder what kind of glue he was sniffing.
Then again, I’m a chump who forecasts revenue for a company with 0.1% of Nvidia’s revenue, so what do I know?
Make that 0.01%. My brain can’t process numbers that large.
If you take a look at the product roadmap for Nvidia’s customers, he’s not sniffing anything. There are lots of plans to employ next generation GPU/NPU hardware sets with unimaginable quantities unified memory offerings. Localized LLMs and SLMs are the next phase of the AI project.
It’s certainly possible, but that money has to come from somewhere. Hyperscalers are already wrecking their cashflows. Can their revenue grow enough to justify a 70% increase in spend or is there enough market share for Nvidia to take from other semis?
That math doesn’t math to me, but as I said, these numbers are too large for my feeble brain.
Excepting that I saw a presentation today with a company saying local LM’s costing $0.04 per task vs. $1.20 for LLM without using Nividia products. The industry is evolving so fast that any prediction is suspect.
I was hoping that it would drop, post earnings…like it usually does. We see if it drops into my range, not expecting much.
Did Huang a once the banking charter for these projects?
Icarus flying towards the Sun.
Perhaps. But maybe this is just a case of CEO FOMO. Well, the term “just” is too tame given the numbers involved. These folks are not gods. It’s no different than when waves of FOMO sweep over the retail ranks.
Their motivations are the same. In fact, in many ways it’s less of a risk for the titans of industry. If they’ve been pouring shareholder money down the toilet what’s the worst that can happen to them? If it proves to be a bit “ill-advised” it’s excusable because everyone else has been doing it. At worst, it’ll be announced that you are stepping down early to spend more time with your family. With a very nice exit package to reward you for your “outstanding leadership”.
Sadly, your hapless retail speculator or investor has no such backstop. You’re losing your own rather than shareholders money.
See also class action settlements with no criminal liabilities.
Now that is how you bury a lede.
Problem is that local LM’s do not need a data center to run. I saw a presentation today of a company selling 3,000 LM machines and all to do apply LM’s to applications without a data center. Seems to me the industry is evolving so fast that no one has a good read on what the trends are.
Yes. On premises LLM hardware rigs seem to be gaining favor from real end-users. Much reduced token charges and hopes for better data security are driving this.
Smaller models that focus on your business needs. As that guy at Uber put it a couple or three months ago, something like “I don’t need a LLM that can write poetry. I need one that can help us forecast next quarter sales.”
Or challenge insurance claims or answer the chat bot better than before.
That’s not scalable and it’s not reliable for a firm that needs to have 4 9’s of uptime. Cloud offerings will still have the business demand, the consumer market will benefit most from localized LM’s, that and as a low cost development platform for AI products.
Not according to the vendor.