Anthropic Discusses Custom 2nm AI Chips With Samsung Electronics
Anthropic is in preliminary talks with Samsung Electronics to manufacture custom AI inference chips using a 2-nanometer process to reduce costs and reliance on Nvidia.
Anthropic is in early-stage, exploratory discussions with Samsung Electronics to develop and manufacture a custom artificial intelligence chip. The potential collaboration focuses on using Samsung's 2-nanometer foundry process (SF2) and advanced packaging technologies to create inference processors designed specifically for the Claude chatbot. The move aims to lower the high costs and energy consumption associated with serving millions of users while reducing the company's dependence on external suppliers.
To build its hardware capabilities, Anthropic hired a former OpenAI chip engineer in June. The company is currently studying performance targets and server integration, while holding discussions with other design firms including Microsoft and the British startup Fractile. This push for vertical integration follows a similar move by competitor OpenAI, which recently unveiled its Jalapeño inference processor developed with Broadcom.
Samsung, a strategic investor in Anthropic's $65 billion Series H round, views a potential deal as a way to secure a high-profile customer and compete more effectively with TSMC. While the custom silicon would supplement its infrastructure, Anthropic stated that a diversified hardware stack comprising Google TPUs, Amazon Trainium chips, and Nvidia GPUs remains central to its compute strategy. No formal agreement, physical prototypes, or production timelines have been established.