Anthropic establishes custom silicon team to address compute demand

Anthropic is currently recruiting engineers for a new custom silicon team, marking its entry into the hardware design space. The company intends to co-design hardware and models to improve the speed and efficiency of its Claude family of AI assistants. This decision follows a broader industry trend where major model developers seek to reduce reliance on third-party hardware providers and optimize for specific inference workloads.

While Anthropic maintains existing infrastructure partnerships with AWS, Google, and Nvidia, the move toward proprietary silicon suggests a long-term need for specialized performance that off-the-shelf accelerators may not provide. Reports indicate the firm has explored potential manufacturing partnerships with Samsung to bring these designs to fruition.

The strategy mirrors efforts at Google DeepMind and Meta, both of which have developed internal accelerators to manage the rising costs and technical requirements of large-scale AI deployment. By controlling the stack from the architecture of the model down to the silicon, Anthropic aims to secure a more sustainable scaling path as global demand for its services increases.