Technology News • • 5-8 minutes

Nvidia posts a $115.8B quarter and guides $130B, with no China compute in the math

Diego Cortés
Diego Cortés
Full Stack Developer & SEO Specialist
Share:
Nvidia posts a $115.8B quarter and guides $130B, with no China compute in the math
Image generated with AI

Nvidia reported $115.8 billion in revenue, up 103% year over year, and guided $130 billion for the closing quarter. The headline is the record; the fine print is a gross margin dropping to 71-72% and $390 billion already committed to purchases.

What Nvidia reported in the quarter ended October 25

The November 17 release, the first without Jensen Huang at the helm, puts revenue at $115.8 billion: up 20% from the previous quarter and up 103% from the $57.0 billion of a year earlier. Analysts had been looking for roughly $109.1 billion, so the number came in above consensus.

Data Center was close to the whole business again: $107.6 billion, up 21% quarter over quarter and 110% year over year. Everything else, gaming included, is a small and fairly stable slice of the total.

The numbers against consensus

Net income was $65.3 billion, up 9% from the prior quarter and 105% from a year ago. Operating income reached $77.5 billion (up 22% quarterly, 115% year over year) while GAAP operating expenses rose 57% year over year to $9.2 billion. Diluted earnings per share came in at $2.70 GAAP and $2.68 non-GAAP, against the $2.47 consensus.

One detail is easy to miss: net income grows 9% quarter over quarter while revenue grows 20%. The company is selling faster than it converts that volume into profit, and that is exactly what the margin guidance reflects.

Vera Rubin: the fastest ramp in company history

Vera Rubin NVL72 is in production and shipping. The release calls it the fastest product ramp in the company's history, with racks running at partners including CoreWeave, Google Cloud, Microsoft Azure, Oracle Cloud Infrastructure and Nebius. On the analyst call, management put the platform's contribution to the quarter at almost $20 billion.

The fourth-quarter outlook: $130 billion and a lower margin

For the current quarter Nvidia expects $130.0 billion, plus or minus 3%, which is a range of $126.1 billion to $133.9 billion. Consensus sat closer to $125 billion. Operating expenses are guided at about $11.0 billion GAAP and $10.7 billion non-GAAP, with a full-year tax rate of 16-18%.

The figure that changes the reading is gross margin: 71-72% for the quarter, against the 74.8% just reported. That is almost three points on a business booking more than a hundred billion dollars a quarter.

The extra week and the China gap

Nvidia's fiscal fourth quarter has 14 weeks, not 13. The guidance says so in its first line, and without that detail the 12% sequential jump is easy to misread: part of the step is calendar, not new demand.

The second condition is also spelled out: the outlook assumes no Data Center compute revenue from China. In the quarter just closed, Data Center Hopper shipments to China were $2.4 billion, and Colette Kress said on the call that the company no longer ships licensed Data Center Hopper to that market.

Why gross margin drops from 74.8% to 71-72%

The release does not provide a full breakdown of the decline, and management did not detail it on the call either. The sensible reading sits with the rest of the quarter: the launch of a new platform usually weighs on cost, and Nvidia is reserving capacity and memory years ahead, which is also paid in advance. What the company said supports the drop; the exact cause is not published.

The commitments: $390 billion in purchases, $102 billion in leases

This is the number that never makes a headline and probably matters most to anyone buying compute. According to the call, purchase commitments reached $390 billion, up from $279 billion the previous quarter, mostly for multi-year memory supply.

Add roughly $102 billion in data center leases and cloud service commitments, up from $74 billion, plus $32 billion in commitments to neoclouds. Management also described a $2.5 trillion cloud industry revenue backlog, up from $2 trillion three months earlier.

It helps to be precise about what these are: future payment obligations, not revenue. Nvidia is buying its key input in advance, and that ties up cash across several fiscal years.

What Kress said about the price of compute

The most useful part of the call for anyone renting GPUs or paying for an API was not the balance sheet but the pricing. The interim CEO described two moves pulling in opposite directions.

Tokens 60% cheaper, Hopper re-contracted at up to double

On one side, token prices fell about 60% since May, which pushed usage up and compute demand with it. On the other, Hopper rentals are being re-contracted at up to double their original price on multi-year deals, and B300 rentals are up roughly 60% since launch.

In plain terms: what you charge per token falls because models are more efficient, while older hardware gains value because there is not enough of it to go around. A provider that renewed a contract this quarter paid more for the same machine.

80% of tokens are now agentic work

Per the company, agentic workloads account for about 80% of the tokens processed on its platform, and frontier labs contributed around 15% of Data Center revenue this year. The roadmap mentioned on the call puts Vera Rubin Ultra next year and Feynman in 2028.

Governance: no Huang, a new export committee and a reported SEC probe

The board unanimously accepted Jensen Huang's resignation and, through chairman Harvey Jones, said it has no intention of reinstating him as CEO; the search for a permanent chief executive continues. The board also created a National Security and Export Controls Committee. Separately, the Wall Street Journal reported on November 17 that the SEC is investigating the company, something Nvidia did not confirm in its release. The case and the handover are already covered in our report on the CEO change.

What it means for anyone buying compute

Three practical consequences. First, the price per token you pay will keep falling, but the price of the machine behind it is rising: new Hopper and B300 contracts are signed higher. Second, with $390 billion committed in advance, much of 2027 capacity is sold before it exists, which narrows the room for cheap spot instances. Third, a gross margin three points lower next quarter is the kind of signal that usually precedes list-price increases on the new generation.

For a 3D studio or a team training small models, the takeaway is the usual one: if compute is part of your unit cost, lock in longer contracts now rather than waiting for the market to soften, because the commitments just published point the other way.

Conclusion

Nvidia grew 103%, guided above consensus and still cut its expected margin: both things are true at once and both are in the same release. The question the quarter leaves behind is not how fast AI is growing, but how much of that growth is already paid for in advance and at what price the contract renews next year.

If you want to follow the hardware thread and what it does to the real cost of AI, the blog has more earnings and infrastructure analysis.

Categories