Anthropic and Samsung discuss custom AI chip development

Anthropic, the US‑based AI startup behind the Claude models, is in early‑stage talks with Samsung Electronics to explore manufacturing a custom artificial‑intelligence chip. Reports indicate the discussions are preliminary: no chip design, performance specifications, or server integration has been finalized, and the project could be abandoned. Anthropic’s move reflects a broader industry push to lessen dependence on Nvidia’s GPUs, which currently power most of its Claude inference workloads. The company, which reported a revenue run‑rate above $30 billion and a $4 billion commitment to Amazon Web Services, seeks greater control over compute costs and a more vertically integrated business model.

Samsung, a major foundry partner for Nvidia’s AI chips and co‑developer of an AI‑chip fab in South Korea, can offer 4‑nm and upcoming 2‑nm process technologies that could suit Anthropic’s inference‑focused needs. The talks follow a similar strategy by rivals such as OpenAI, which recently unveiled a custom inference processor built with Broadcom, and Google, which runs its own TPUs. Anthropic has also hired former OpenAI chip lead Clive Chan, signaling a shift from exploratory talks to active development. No formal agreement or timeline has been announced, and the companies have declined to comment further.