
Earlier this summer, top AI leaders warned that the industry needed to slow down its most powerful models. Those calls for a frontier AI pause faded fast as OpenAI, Anthropic, Nvidia, and AMD kept pushing forward without missing a beat. Huge quarterly profits from major chipmakers confirm that AI labs are spending more on hardware than ever.
In August, OpenAI paused a major training run for only 10 days after its agents accidentally hacked Hugging Face, while Anthropic CEO Dario Amodei urged a broader slowdown in September. By November, however, Anthropic released upgraded Fable and Mythos frontier models, and OpenAI started adopting Nvidia’s new Vera CPUs. As The Verge reports, Nvidia also posted a massive quarter with data center revenue climbing 110% to $107.6 billion, helped by nearly $20 billion in early Vera Rubin NVL72 sales.
Hardware sales keep rising across the industry
Nvidia interim CEO Colette Kress told investors that big frontier labs should make up about 15% of its data center revenue this fiscal year. This prompted the company to raise its next-year growth forecast to 75%. Rival AMD also delivered its largest quarter on record, with overall revenue rising 45% to $13.4 billion as OpenAI and Anthropic ran workloads on its new Helios systems. Chip manufacturer TSMC is reporting strong order volume as well, showing no sign that labs are cutting back on hardware.
Inference workloads drive even more compute demand
Counterpoint Research vice president Neil Shah told The Verge that shifting from training new models to running existing ones—known as inference—actually requires even more computing hardware overall. Business spending data from Ramp shows frontier models hit a record 60% share of corporate token usage in early November. Meanwhile, Ramp lead economist Ara Kharazian and Kress noted that fierce competition has driven token prices down nearly 60% since May, encouraging companies to run even more AI tasks.
For chipmakers like Nvidia and AMD, the bigger long-term challenge is not an AI pause but tech giants designing cheaper custom chips to protect their profit margins. Counterpoint Research expects custom AI server processors to outsell standard data center GPUs by 2028 as companies look for more affordable ways to serve millions of users. Until that shift happens, massive quarterly earnings prove that Silicon Valley’s appetite for bigger models and faster chips remains wide open.
The post OpenAI and Anthropic Warned About Pausing AI, Then Bought Up Nvidia and AMD’s Fastest Chips appeared first on Android Headlines.