· via The Verge
Nvidia and AMD post record quarters as chipmakers shrug off frontier AI pause
Nvidia and AMD posted record AI chip results despite OpenAI's brief training pause and Anthropic's call to slow frontier models, with inference demand and custom silicon now reshaping the market.

A pause that barely registered
Last summer's debate over whether frontier AI development should slow down has had little visible effect on the companies that supply the hardware behind it. According to The Verge, OpenAI paused its largest planned frontier training run after its AI agents accidentally hacked Hugging Face, and early evidence suggested its next model could independently find and exploit security flaws. In September, Anthropic CEO Dario Amodei called for slowing work on the most capable models, warning they could be used for cyberattacks or become harder for their creators to control.
The pause itself proved brief. As The Verge reports, OpenAI had announced a two-week pause and restarted the run ten days later at the end of August, while temporarily holding back some smaller experimental runs. Anthropic, for its part, shipped upgraded versions of its Fable and Mythos frontier models in November, with improvements in coding, reasoning and vision.
Record results from the chipmakers
The companies selling AI silicon are doing better than ever. Nvidia posted data center revenue of $107.6 billion for the quarter ended October 25, up 110 percent from a year earlier. The company began shipping its most advanced AI system, the Vera Rubin NVL72, which interim CEO Colette Kress called the fastest product ramp in the company's history. The Rubin platform generated nearly $20 billion in revenue in its first quarter of production shipments alone, and the racks are already running at CoreWeave, Google, Microsoft and Oracle, with Meta planning expanded deployments.
Nvidia also named Anthropic and OpenAI as partners. Anthropic has expanded access to Vera Rubin capacity through cloud providers and is working with Nvidia on future chip designs tailored to its workloads, while OpenAI is among the companies adopting Nvidia's new Vera CPUs. Kress said major frontier labs are expected to account for roughly 15 percent of Nvidia's data center revenue this fiscal year, and the company raised its revenue growth forecast for next fiscal year from 70 percent to 75 percent.
AMD reported its biggest quarter ever earlier in November, with revenue up 45 percent to $13.4 billion and data center sales nearly doubling to $8.5 billion. OpenAI and Anthropic were among the first customers running workloads on AMD's new Helios systems, which use the latest MI450 GPUs.
Lagging signals and rising inference demand
A training slowdown would take time to appear in chipmakers' earnings regardless. Neil Shah of Counterpoint Research told The Verge in September that existing computing and memory procurement agreements would delay any financial impact by about six months, though weakness could surface sooner in orders placed with manufacturers such as TSMC. TSMC's latest sales figures, however, also show continued strength.
There is also a structural reason a frontier slowdown might not hurt chip vendors: inference. Shah argues that running the models we already have will collectively require more compute infrastructure than training does, producing a net increase in demand for all types of accelerators.
Data analyzed by The Verge from the corporate spending platform Ramp shows frontier model families accounted for about 60 percent of business token usage in early November, a record high roughly three times their share at the start of the year, spanning OpenAI's Astra and Sol models and Anthropic's Opus and Fable. Yet business AI spending is growing only modestly. Ramp's lead economist Ara Kharazian says demand is not the problem; competition among model providers is driving token prices down so quickly that OpenAI and Anthropic's revenue projections are being called into question. On Nvidia's earnings call, Kress noted token prices have fallen nearly 60 percent since May, fueling more usage, while rental prices for Nvidia GPUs across generations have gone up.
The real threat is cheaper silicon
The bigger risk to Nvidia may not be that labs slow down, but that they find cheaper ways to keep going. Counterpoint expects shipments of custom AI server processors to surpass those of data center GPUs by 2028. When UBS analyst Timothy Arcuri asked Kress whether Nvidia would sacrifice some of its hefty margins as frontier labs invest in custom chips and prioritize efficiency, she maintained that the company's systems justify their pricing and reiterated the gross margin outlook. Shah's summary of the dynamic: braking frontier training would not shrink the AI chip market, only shift where the money goes.
Why it matters
The episode exposes a consequential split between AI labs urging caution and the hardware industry shipping record volumes underneath them. Even a genuine pullback in frontier training would not quickly dent chip revenue, because long procurement contracts and rising inference demand cushion the blow. The sharper tension is economic: falling token prices squeeze lab revenues while rewarding compute suppliers, and the labs' pursuit of custom chips may yet erode Nvidia's dominance more effectively than any pause ever could.
- #ai-chips
- #nvidia
- #amd
- #frontier-models
- #data-center