Nvidia and AMD are reporting record demand for AI chips, showing little financial fallout from this summer's brief pause on frontier model training. Nvidia said data center revenue rose 110 percent from a year earlier to $107.6 billion in the quarter ended October 25, and AMD posted its largest quarter ever, with revenue up 45 percent to $13.4 billion and data center sales nearly doubling to $8.5 billion.

Nvidia has begun shipping its Vera Rubin NVL72 system, calling it the fastest product ramp in company history. The Rubin platform brought in nearly $20 billion in its first quarter of production shipments, with racks running at CoreWeave, Google, Microsoft and Oracle, and Meta planning to expand Vera Rubin deployments for training and running AI models and agents. OpenAI is among the companies adopting Nvidia's new Vera CPUs, while Anthropic has expanded access to Vera Rubin capacity through cloud providers and is working with Nvidia on future chip designs for its workloads. Interim CEO Colette Kress said on the earnings call that proprietary models from OpenAI and Anthropic continue to advance the frontier, and that major frontier labs are expected to account for roughly 15 percent of data center revenue this fiscal year. Nvidia raised its revenue growth forecast for next fiscal year from 70 percent to 75 percent.

The pause that unsettled investors began this summer, when OpenAI halted 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 advances in the most powerful models, warning they could be used to launch cyberattacks or become harder for their creators to control. US chip stocks sold off on the news. OpenAI's pause lasted only two weeks, and it restarted the run 10 days later at the end of August while continuing to temporarily hold back some smaller experimental training runs. Anthropic released upgraded versions of its Fable and Mythos frontier models in November.

Even a significant pullback in training would take time to appear in chipmakers' earnings. Neil Shah, vice president of research at Counterpoint Research, told The Verge in September that existing computing and memory procurement agreements would delay financial effects by about six months, though orders with manufacturers like TSMC could reflect a slowdown sooner. TSMC's latest sales figures show continued strength.

Growing demand for inference, the process of running existing models, offers another cushion. Shah said inference workloads will collectively require more compute infrastructure than training, amounting to a net increase in demand for all types of accelerators. According to an analysis of data from corporate spending platform Ramp, frontier model families accounted for about 60 percent of token usage among businesses using Ramp in early November, a record high and roughly triple their share at the start of the year. Kress said token prices had fallen nearly 60 percent since May, spurring more usage, while rental prices for Nvidia GPUs across generations have risen.

More AI news from TechManNews.