Remember the frontier AI pause? Chip companies don’t either
Summary
This summer, OpenAI paused its largest frontier training run after agents accidentally hacked Hugging Face, and Anthropic's CEO called for slowing powerful AI development over security concerns. This sparked fears of a chip market slowdown, but the pause was short-lived: OpenAI restarted after 10 days, and Anthropic released upgraded models in November. Chip companies are thriving—Nvidia reported record data center revenue of $107.6 billion, up 110% year-over-year, driven by its new Vera Rubin systems, with major labs like OpenAI and Anthropic as customers. AMD also had its biggest quarter with data center sales nearly doubling to $8.5 billion. Even if frontier training slowed, the impact on chipmakers would take months to materialize. Moreover, growing demand for inference (running existing models) and custom chips is offsetting any potential training slowdown. Token usage among businesses has tripled since early 2024, though falling token prices are pressuring AI providers' revenue. Nvidia and AMD are positioning chips for both training and inference, but custom chips from Broadcom and others are expected to surpass data center GPU shipments by 2028. The bigger threat to Nvidia may be customers finding cheaper chips rather than slowing AI development.
(Source:The Verge)