Anthropic is building its own chips for Claude

Renting everyone else’s compute is no longer cutting it for Claude.

Anthropic has officially confirmed that it is building an in-house custom silicon team. After Business Insider first broke the story, TechCrunch confirmed the move with the Claude maker, which is actively hiring chip design engineers. The goal is to co-design hardware and AI models side by side to make Claude run significantly faster and more efficiently.

The great AI silicon gold rush

Anthropic already has computing agreements stacked with AWS, Google, Nvidia, and AMD. But surging demand for Claude has made clear that relying solely on off-the-shelf hardware leaves too much control on the table. Last month, The Information reported that Anthropic was scouting Samsung as a potential foundry partner.

Anthropic is far from alone in this pursuit. In June, OpenAI revealed Jalapeño, its inference chip built with Broadcom. Meta continues to develop its MTIA accelerators, and Google DeepMind has long relied on in-house TPUs.

Why it matters

Frontier AI labs are quickly realizing that software moats eventually hit physical hardware walls.

For everyday users and developers, the Claude bottleneck is always latency and API costs. Tailoring chips specifically for Claude’s architecture is one of the few realistic paths left to slash inference bills and speed up token generation at scale.

We think the move makes strategic sense, but building silicon is notoriously unforgiving. Tape-outs take years and billions of dollars, and working with foundries like Samsung comes with steep execution risks. Anthropic’s models may need compute today, but custom chips are a game of extreme patience.