Anthropic has announced that it is developing custom chips for Claude and is seeking engineers with experience in "shipping silicon."

      Anthropic has officially announced the formation of an in-house silicon team to create custom chips for Claude, marking the company's first public acknowledgment of this initiative. A representative informed Business Insider that this collaboration will involve co-designing hardware and models to enable Claude to operate more swiftly and efficiently to meet customer demands. The company emphasized its commitment to a "multi-chip approach," continuing to utilize hardware from AWS, Google, Nvidia, and AMD for scaling.

      The company has posted a job listing for this "custom silicon team," seeking engineers with experience in chip design and verification, with salaries ranging from $320,000 to $485,000. Applicants are expected to demonstrate “direct personal contribution” to the completion and delivery of semiconductor designs. The listing specifies that candidates should have experience in shipping silicon, maintain realistic timelines, and be capable of making significant decisions independently.

      This announcement follows months of indications regarding Anthropic's pursuits. In April, Reuters revealed that the company was considering custom chips as the revenue from Claude surpassed $30 billion, although no dedicated team was in place at that time. Last month, The Information reported discussions between Anthropic and Samsung as a possible manufacturing partner. The recruitment of Clive Chan, who previously contributed to OpenAI’s chip development, signaled a shift from exploration to active development, and this confirmation solidifies that transition.

      Anthropic is not the only player in this space. OpenAI introduced Jalapeño in June, a custom inference chip set to be developed with Broadcom, aiming for deployment by late 2026. Meta plans to initiate production of its in-house "Iris" chip in September. Moreover, Mistral's CEO mentioned that the French company is considering developing its own silicon. While designing an AI chip typically costs around $500 million, the economics become more favorable at Anthropic's scale. The company currently operates Claude on Google TPUs, Amazon Trainium, Nvidia GPUs, and AMD hardware, and developing its silicon would provide a fully controllable fifth option.