Anthropic builds in-house chip team for Claude amid global shortage

Anthropic has confirmed plans to design custom AI chips, standing up a new silicon team and actively recruiting hardware engineers. The company frames the decision as a response to a persistent global chip shortage and the escalating compute demands of frontier models like Opus 5 and the Mythos line, rather than a wholesale break from existing suppliers.

Crucially, Anthropic stresses it will maintain a diversified hardware stack — continuing to lean on AWS Trainium, Google TPUs, Nvidia GPUs, and AMD accelerators — while custom silicon supplements rather than replaces them. Earlier reporting has floated a potential collaboration with Samsung for chip development, though Anthropic has not formally confirmed a fabrication partner.

The move slots Anthropic into a clear theme-of-the-week: the in-house silicon race. OpenAI unveiled its Broadcom-designed 'Jalapeño' semiconductor in June, Meta's next-generation AI chip is reportedly heading to production in September, and even Mistral has explored custom hardware. AMD's acquisition of Taalas — a startup etching models directly into silicon for faster inference — was a top Hacker News story this week (476 points), signaling broad market interest in escaping Nvidia's margins and supply constraints.

The strategic logic is straightforward: at Anthropic's compute scale (it recently signed a reported $71B compute deal and is leasing 300MW from xAI at $1.25B/month per SpaceX's S-1), even modest per-chip efficiency gains compound into billions. The skeptical read is that custom silicon takes years to reach production and Anthropic is starting well behind hyperscalers who have shipped multiple TPU/Trainium generations. Watch for named hires and any confirmed foundry partner as the first concrete signals this is more than a recruiting push.