Anthropic has confirmed it is building a team to design custom silicon optimized for AI workloads. Business Insider first reported the development, and Anthropic subsequently confirmed the news to TechCrunch.
The maker of the Claude model said it plans to co-design hardware and models to help its technology run faster and more efficiently. The company intends the co-design approach to better align software and hardware for improved performance and scalability.
According to reporting by The Information from last month, Anthropic has explored Samsung as a potential partner for building such chips. At the same time, Anthropic already has agreements to access AI compute hardware with Amazon Web Services (AWS), Google, Nvidia, and AMD.
The move comes as demand for Claude increases and AI companies are rapidly securing infrastructure deals. Anthropic has indicated that relying exclusively on external providers will not be sufficient to scale to the level of demand it expects.
Anthropic is not the first AI company to pursue proprietary hardware: in June, OpenAI unveiled the Broadcom-built Jalapeño chip, designed specifically for inference workloads. Google DeepMind has long used Alphabet’s TPU chips to power its models, and Meta is developing its own MTIA accelerators for AI tasks.
A job listing shows Anthropic is seeking engineers with experience in chip design for its “custom silicon team,” indicating active hiring in the hardware design area.
The article has been updated to include Anthropic’s confirmation of the plan.
