Anthropic has confirmed it is building an in-house "custom silicon team" to design chips for running its Claude models, following the discovery of job listings and reports first surfaced by Business Insider and TechCrunch. The move mirrors similar steps by rivals such as OpenAI, Google and Meta, and reflects growing pressure among AI firms to reduce reliance on Nvidia hardware amid fierce competition for scarce compute capacity, as well as the potential performance gains from designing chips and models together.

Anthropic says it will pursue a "multi-chip approach," continuing to use hardware from other suppliers alongside its own custom designs as it scales up, and it has previously been reported to be exploring a manufacturing partnership with Samsung. This follows OpenAI's recent unveiling of its own inference chip, Jalapeño, developed with Broadcom, while Google and Meta already run models on proprietary hardware. Anthropic says its hardware and model teams will now co-design side by side, though since hiring for the team is only just beginning, any practical benefits are likely to be some way off.