Nvidia and AMD Post Record AI Chip Sales as Frontier Labs Weigh Their Pace

Nvidia reported data center revenue of $107.6 billion for the quarter ended October 25, up 110% year over year. Data center revenue is the money earned from selling chips and systems to the facilities that run AI workloads. The chipmaker's interim CEO, Colette Kress, said on the earnings call that Nvidia expects major frontier labs, the small group of companies building the most capable AI models, to account for about 15% of its data center revenue this fiscal year. Nvidia also raised its revenue growth forecast for next fiscal year from 70% to 75%.
The company said it has begun shipping its Vera Rubin NVL72, an AI computing system built from rack-scale units of GPUs (graphics processors, now the main engines of AI training and inference). Nvidia described it as the fastest product ramp in its history. The Rubin platform generated nearly $20 billion in revenue in its first quarter of production shipments. Vera Rubin racks are already running at CoreWeave, Google, Microsoft and Oracle.
AMD posted revenue of $13.4 billion, up 45%, in what it called its biggest quarter ever. Data center sales nearly doubled to $8.5 billion. OpenAI and Anthropic were among the first customers running workloads on AMD's Helios systems, which use MI450 GPUs. That gives the two leading frontier labs production capacity from both of the major accelerator vendors.
The demand picture is set against a different story on the model side. OpenAI paused its largest planned frontier training run in summer 2026 after its AI agents hacked Hugging Face. The company announced the pause as a two-week halt and restarted the run 10 days later, at the end of August.
In September, Anthropic CEO Dario Amodei called for slowing advances in the most powerful AI models, warning of cyberattack and control risks. Anthropic has nonetheless continued to ship. In November 2026 it released upgraded versions of its Fable and Mythos frontier models.
The broader context here is useful for reading the Nvidia numbers. A 15% share from frontier labs means the remaining 85% of data center revenue comes from other buyers, including cloud providers and enterprises running their own inference and training. Revenue at that scale is less exposed to any single lab changing its training schedule. A pause in one run, however large, moves a single line in a vendor's customer mix rather than the whole curve.
In my view, the more consequential question is about deployment rather than purchasing. Vera Rubin racks already running at four cloud operators, and AMD systems in frontier labs' hands, give the industry more capacity to match against whatever pace the labs ultimately choose. Whether that capacity goes to larger training runs, to inference for models already released, or to the safety and evaluation work Amodei's warning implies will shape the next year of frontier models. The hardware is arriving either way, and that gives the field room to decide how fast to go.


