Anthropic in Talks With Samsung to Build Custom AI Chip on 2nm Process
Summary: Anthropic is exploring a custom AI chip with Samsung Foundry, targeting the 2nm process node, in what would be the company's first deliberate step toward owning a piece of its hardware stack.
Key Facts
- Stage: early-stage talks only; Anthropic says use case, power specs, and server integration are still undecided — the effort could be abandoned
- Process target: Samsung's 2nm node and advanced packaging — the same generation Samsung is ramping for mobile SoCs and HPC clients
- Key hire: Clive Chan, an early member of OpenAI's custom chip team, has joined to build Anthropic's chip engineering organisation
- Existing stack: Google TPU, Amazon Trainium, and Nvidia GPU remain central to Anthropic's compute strategy; a custom chip would be additive, not a replacement
- Timing: report first published by The Information on July 2, 2026
Why It Matters
OpenAI shipped its Jalapeño inference chip with Broadcom in June; Google and Amazon have run custom silicon programmes for years. Anthropic moving toward its own chip would complete the big-lab hardware sweep and give the company a path to reducing per-token cloud costs — critical as inference volumes scale with Claude Sonnet 5's $2/M input pricing and agentic workloads. Samsung wins either way: landing Anthropic would diversify its advanced-node customer base and add to TSMC-challenging momentum.
Read More
- TechCrunch — TechCrunch
- Bloomberg — Bloomberg
- The Next Web — The Next Web