Anthropic confirmed this week it is assembling an in-house chip design team to build custom silicon for its Claude AI models, following a software-hardware co-design strategy in which the chips and the models are developed together.
The company called the move "the latest step" in a multi-chip approach that will keep it running on a diversified hardware stack rather than depending solely on Nvidia GPUs. Samsung has been reported as a manufacturing partner for the new inference accelerators. Anthropic is hiring engineers who have "shipped silicon" before, offering salaries between $320,000 and $485,000, and recently brought on Clive Chan, OpenAI's former custom chip program lead, who left to join the effort in June.
The push reflects the scale of Anthropic's business: the company now serves billions of tokens daily on a roughly $30 billion revenue run-rate, and custom-designed chips could sharply cut the cost of running Claude at that volume.