Anthropic begins building custom chips to power Claude

Anthropic has confirmed it is building a custom silicon team to design its own chips for running Claude, marking a major step toward reducing reliance on Nvidia and improving performance through vertical integration. Job listings for semiconductor engineers revealed the plan, later verified by company spokespeople, who emphasized that Anthropic will still use a multi‑chip strategy combining its own designs with external hardware. The move mirrors similar efforts across the AI industry: OpenAI recently announced its Jalapeño inference chip, Google has long used custom TPUs, Meta has deployed its own silicon, and Mistral is exploring the same path. Custom chips allow AI companies to optimize hardware specifically for their models, potentially boosting efficiency and lowering costs as demand for compute continues to outpace supply. Anthropic plans to co‑design future models and hardware together to gain a competitive edge, especially as smaller and cheaper models increasingly run on user‑owned devices. However, since the company is still hiring key engineers, it will take time before any benefits materialize.

Read the full story on Ars Technica →