Overview
- Multiple outlets report that Anthropic began preliminary discussions with Samsung this week to explore making a custom AI processor using Samsung’s 2‑nanometer node and advanced packaging, but no design, testing, or contract exists yet.
- Anthropic says it will continue to rely on Nvidia GPUs, Google TPUs and AWS Trainium for core compute while it evaluates whether any custom chip would target training, inference, or other server roles.
- The company’s $65 billion Series H in May and rapid revenue growth give Anthropic the funds to pursue costly chip development, and the June hire of Clive Chan signals a deliberate buildout of in‑house hardware engineering.
- A formal Samsung deal would strengthen Samsung’s foundry profile and pressure TSMC, yet analysts note Samsung’s past yield challenges on leading nodes could complicate production for high‑end AI accelerators.
- The move follows a broader industry trend of AI labs designing bespoke silicon to cut per‑query costs and diversify supply, and it could shift data center economics and supplier leverage if Anthropic proceeds.