The processors would be designed alongside the models themselves, the company says, so that Claude runs faster and more efficiently at the scale its customers demand — while hardware from AWS, Google, Nvidia and AMD stays central to how Anthropic expands.

Anthropic is assembling a group of its own engineers to build processors made specifically for its Claude models, the company said earlier this month, the first time it has publicly confirmed it intends to design its own silicon. A company spokesperson told Business Insider that the chips and the models would be developed together, so that Claude would gain speed and efficiency at the volumes its customers demand.

That is not a step away from the company's suppliers. The spokesperson said Anthropic has pursued, and means to keep pursuing, a strategy drawing on several kinds of chips, with equipment from AWS, Google, Nvidia and AMD remaining at the heart of how it expands. Its AI compute at present depends on hardware built by other firms — AWS Trainium, Google TPUs and Nvidia GPUs — and it has committed to multiple gigawatts of next-generation Google TPU capacity starting from 2027.

A job advertisement for the custom silicon group sets out the work: the company is recruiting across front-end design, verification ahead of silicon, physical implementation, analog and mixed-signal engineering and packaging, at pay of between $320,000 and $485,000. The posting asks applicants to show they contributed personally and directly to semiconductor designs that were finalized and shipped, and to be at ease making weighty calls with no large organization behind them. Once a part comes back, whoever is hired is asked to help bring up and debug the first silicon.

Creating an AI chip internally is expensive and hard. It can, though, lessen a company's dependence on Nvidia, which holds the leading position in the market for AI accelerator chips, and it lets AI labs tailor their computing to what their models specifically require. The work comes as appetite for Claude grows and rival AI firms fight hard over contracts for the infrastructure that runs AI.

Anthropic is not the only lab with chip work of its own. OpenAI presented a chip named Jalapeño in June, made with Broadcom and designed specifically for inference, and the chief executive of Mistral has said the French firm is weighing whether to build silicon of its own.

As of its confirmation, Anthropic had named neither a manufacturing partner nor a release date for the custom chips.