In short
Nvidia and AMD posted blockbuster earnings even after fears that frontier AI labs might slow training. The results suggest AI chip demand is broadening into inference and custom silicon, not fading.
- Nvidia lifted its next fiscal year revenue growth outlook to 75% from 70%.
- AMD said data center sales nearly doubled and helped produce its biggest quarter ever.
- OpenAI briefly paused its largest frontier training run this summer, but resumed it 10 days later.
- Analysts say any slowdown would likely shift chip spending, not eliminate it.
Nvidia and Advanced Micro Devices both delivered blockbuster results, signaling that the AI hardware boom was still running hot through late 2026 even after worries that frontier-model developers might slow their biggest training projects. The numbers suggest the immediate risk to chipmakers is not a collapse in AI spending, but a shift in where that spending goes.
Why it matters
For months, investors have treated the pace of frontier AI development as a direct proxy for chip demand. If OpenAI, Anthropic and other major labs cut back on giant training runs, the assumption went, Nvidia and AMD would feel it quickly.
This week’s earnings picture complicates that narrative. Nvidia said its data center business grew 110% year over year to $107.6 billion in the quarter ended Oct. 25, while AMD said data center sales nearly doubled and helped drive its biggest quarter ever. Together, the results suggest AI infrastructure demand remains broad enough to absorb a temporary pause in some frontier work.
That does not mean the market is risk-free. The more immediate challenge may be customers shifting toward cheaper custom AI chips rather than using fewer chips overall. In other words, the threat is not necessarily less AI; it may be a different mix of silicon.
What Nvidia and AMD said
Nvidia used its earnings call to argue that frontier labs still matter to the business, even as some model developers have publicly questioned the pace of progress. Colette Kress said, “Proprietary models from OpenAI and Anthropic continue to advance the frontier” while the company pointed to about 15% of data center revenue coming from frontier labs this fiscal year.
The company also raised its next fiscal year revenue growth outlook from 70% to 75%, another sign that management sees demand remaining resilient. Nvidia’s results came after a quarter in which its data center revenue reached $107.6 billion, a 110% increase from a year earlier.
AMD offered a similar message through the strength of its own results. The company reported $13.4 billion in quarterly revenue, up 45%, with data center sales of $8.5 billion nearly doubling from the prior year. That was enough to make it AMD’s biggest quarter ever.
| Company | Recent result | Key figure | What it signals |
|---|---|---|---|
| Nvidia | Quarter ended Oct. 25 | $107.6 billion data center revenue, up 110% | AI infrastructure demand remained intense |
| Nvidia | Next fiscal year outlook | Revenue growth guide raised from 70% to 75% | Management sees continued momentum |
| AMD | Biggest quarter ever | $13.4 billion revenue, up 45% | Data center demand kept accelerating |
| AMD | Quarter result | $8.5 billion data center sales | AI spending extended beyond Nvidia |
How the slowdown scare started
The anxiety began months earlier, when signs emerged that some of the biggest AI builders were becoming more cautious about the most expensive kind of development work: frontier training. OpenAI paused its largest frontier training run this summer after internal safety and security concerns, according to The Verge. Around the same period, Anthropic chief executive Dario Amodei urged slower progress on the most powerful AI systems in September.
Those developments fed a wave of concern across U.S. chip stocks. Investors worried that if frontier labs reduced or delayed the largest training projects, they would also buy fewer high-end GPUs and related systems. Broadcom, which has not yet reported earnings, remains one of the open questions in that debate.
But the picture did not stay static for long. OpenAI resumed the paused training run 10 days later, at the end of August, though it still held back some smaller experimental training runs after restarting. In November, Anthropic released upgraded Fable and Mythos frontier models, undercutting the idea that leading labs were stepping away from the race entirely.
Why chipmakers may be less exposed than the market feared
The latest earnings suggest that even if one part of the AI cycle slows, another may be filling the gap. Training remains important, but inference — the everyday running of deployed models — is rising quickly and can keep hardware demand strong.
That matters because training is only one slice of the market. Frontier models are the largest and most capable AI systems, and they require heavy training compute. But once those models are in use, the ongoing cost of serving queries, generating outputs and handling business workloads can become just as important for chip suppliers.
Ramp’s data points in that direction. The company said frontier model families accounted for about 60% of token usage among businesses in early November, roughly three times their share at the start of the year. Ramp economist Ara Kharazian summarized the price-and-usage dynamic bluntly: “It’s not really a demand issue though”
Nvidia added its own evidence for that trend, saying token prices have fallen nearly 60% since May. Lower prices can stimulate more use, even if they make revenue per token less lucrative. That helps explain why demand for AI chips can remain strong even when model costs come down.
Training versus inference: what is changing?
The market is moving from a narrow question — how many giant training runs will happen — to a wider one: how much total AI compute will be needed across training, inference and custom deployments?
- Training demand can pause when labs slow major experiments.
- Inference demand can rise even if training cools.
- Custom chips can shift spend away from general-purpose GPUs without reducing overall AI use.
Background: why the market got so nervous
Frontier AI development has been the main engine of chip spending for much of the last two years, with Nvidia and AMD standing among the biggest beneficiaries. A handful of leading labs bought massive accelerator clusters to train ever-larger models, and investors came to treat those purchases as a direct read on the health of the broader AI economy.
That is why any sign of restraint from OpenAI or Anthropic hits the market so hard. If the biggest buyers are even temporarily less aggressive, the first instinct on Wall Street is to discount future chip orders. The selloff in U.S. chip stocks reflected that logic.
Yet the supply chain has not shown the kind of break that would confirm a meaningful demand collapse. TSMC sales figures still showed continued strength, a sign that semiconductor demand remained robust across the chain. And Counterpoint Research warned that even if a slowdown were underway, it would take roughly six months to show up in chipmakers’ earnings.
That lag matters. It means a quarter or two of strong results does not prove the AI cycle is immune to change, but it does suggest that any real slowdown would take time to show through. For now, the earnings reports from Nvidia and AMD look more like a rebuttal to panic than a final verdict on the cycle.
By the numbers
| Metric | Figure | Context |
|---|---|---|
| Nvidia data center revenue growth | 110% year over year | Quarter ended Oct. 25 |
| Nvidia data center revenue | $107.6 billion | Quarter ended Oct. 25 |
| Nvidia next fiscal year growth outlook | 75% | Raised from 70% |
| AMD quarterly revenue | $13.4 billion | Up 45% |
| AMD data center sales | $8.5 billion | Nearly doubled |
| OpenAI pause to restart | 10 days | Largest frontier training run |
| Frontier labs’ share of Nvidia data center revenue | 15% | Projected for this fiscal year |
| Token price decline | Nearly 60% | Since May |
What the market may be missing
The strongest signal in the earnings reports may be that AI demand is broadening. The market once fixated on a simple equation: more frontier training equals more chip sales. That is still true, but incomplete.
Inference demand is rising. Business usage is rising. Model access is getting cheaper. And, crucially, lower pricing can expand usage fast enough to offset some of the spending anxiety tied to slower training.
Counterpoint Research argues that the real pressure point for Nvidia may not be a drop in AI activity, but a change in what kind of hardware customers buy. The firm expects custom AI server processor shipments to exceed data center GPU shipments by 2028. That would not mean the AI market is shrinking. It would mean more of the value chain is moving toward cheaper, purpose-built silicon.
Neil Shah of Counterpoint captured that distinction in a way investors are likely to remember: “Pumping the brakes on frontier training won’t shrink the AI chip market but it will simply reallocate the spend.”
For Nvidia, that is both comforting and cautionary. It suggests demand is alive, but it also points to a future in which pricing power and product mix matter as much as raw compute appetite.
What happens next
The next test is not whether AI demand exists — this week’s results show that it does — but whether any slowdown in frontier training begins to show up in chip orders over the next two quarters.
Investors will also be watching three things closely:
- Whether OpenAI or Anthropic change the pace of large training runs again.
- Whether Broadcom’s upcoming results confirm or weaken the no-slowdown signal.
- Whether custom AI chips keep taking share faster than expected from general-purpose GPUs.
For now, Nvidia and AMD have bought the AI trade more time. The burden of proof has shifted to anyone arguing that the spending cycle is breaking. The more immediate question is whether the industry’s next phase is defined by less AI spending, or by different AI spending.
On the evidence in these earnings, the second scenario looks more likely.
Timeline
| Date | Event | Why it mattered |
|---|---|---|
| This summer | OpenAI paused its largest planned frontier training run | Triggered fears of softer chip demand |
| September 2026 | Dario Amodei urged slower progress on the most powerful AI models | Added to caution around frontier spending |
| End of August 2026 | OpenAI restarted the training run | Showed the pause was temporary |
| November 2026 | Anthropic released upgraded Fable and Mythos models | Suggested frontier work continued |
| Tuesday, Nov. 2026 | Nvidia reported a blockbuster quarter | Reinforced the strength of AI infrastructure demand |
| Earlier this month | AMD reported its biggest quarter ever | Confirmed broad chip demand beyond one supplier |
Bottom line
The market spent much of the fall worrying that AI model developers were pulling back just as chipmakers were pricing in endless growth. Nvidia and AMD’s latest results show that fear was, at minimum, premature.
AI spending may be changing shape, but it is not disappearing. For chipmakers, that is the key distinction.
Frequently asked questions
Why did Nvidia and AMD stocks hold up despite AI slowdown fears?
They held up because both companies reported exceptionally strong results, showing that AI infrastructure spending remained robust even after concerns that some frontier model developers might slow their biggest training runs. The latest numbers suggest demand is broadening into inference and custom silicon rather than disappearing.
Did OpenAI actually stop its biggest training run?
Yes. According to The Verge, OpenAI paused its largest frontier training run this summer because of internal safety and security concerns, then resumed it 10 days later at the end of August 2026. It also kept back some smaller experimental training runs after restarting.
What does Nvidia expect from frontier labs now?
Nvidia says frontier labs still matter to its business and projected that they will account for about 15% of its data center revenue this fiscal year. The company also raised its next fiscal year revenue growth outlook from 70% to 75%.
Is a slowdown in AI training bad for chipmakers?
Not necessarily. Counterpoint Research says a slowdown in frontier training would likely reallocate spending rather than shrink the AI chip market. The bigger risk for Nvidia may be customers moving to cheaper custom chips, not using less AI overall.
When would a real slowdown show up in earnings?
Counterpoint Research says it would take about six months for the financial effects of a slowdown to appear in chipmakers’ earnings. That means even a recent pause in training activity would not immediately show up in quarterly results.









