Anthropic is telling investors that the 15-person team designing its custom chips already uses Claude, according to The Information, which cited unnamed sources. The company's engineers found that Claude cuts design time compared with their estimates of how designers at Nvidia and AMD work.

No figure has been published, and the comparison rests on internal estimates, not a head-to-head test. It looks like part of the pitch ahead of Anthropic's Nasdaq IPO.

A real saving would matter. Anthropic's amended filing shows about $895bn in compute and infrastructure commitments. Industry estimates put the cost of designing an advanced AI chip near $500m.

The claim fits what the company has said in public. It confirmed on Aug 5 that it is building an in-house silicon team, with no timeline, and says the effort sits alongside chips from AWS, Google, Nvidia and AMD. Its job listings describe Claude-assisted RTL review and coverage checks. A separate July posting seeks an engineer to train Claude on RTL generation, verification and physical design.

Outside evidence is mixed. Samsung reportedly used Claude Code to cut a month-long chip verification job to about two days, though it also reported serious mistakes.

As of Oct 14, Nvidia and Chipagents were separately building formal evaluation tools for AI-assisted chip design. That suggests the industry is testing such claims rather than taking vendor estimates at face value.

© Raydn · raydn.us