Anthropic has held early-stage talks with Samsung Electronics about manufacturing a custom AI chip, The Information reported. The discussions center on Samsung's 2-nanometer manufacturing process and its advanced packaging facilities, but no detailed chip design or manufacturing work has begun, and Anthropic may not proceed with the project at all.
The talks follow a hire that signaled Anthropic was building out its own chip engineering effort. Clive Chan, who joined OpenAI in January 2024 from Tesla's Dojo supercomputer team, became the second hardware engineer on OpenAI's custom-chip program, working on its collaboration with Broadcom to build in-house AI accelerators. Chan left OpenAI for Anthropic in early June.
Samsung is already financially tied to Anthropic. Samsung Electronics, SK Hynix, and Micron all took strategic stakes in Anthropic's $65 billion Series H round in May, which valued the company at $965 billion post-money, ahead of OpenAI's $852 billion valuation. Samsung was notable among the three as the only investor that also operates a foundry business capable of manufacturing logic chips, rather than just memory.
The timing lines up with a capacity gap on Samsung's side. Separate reporting from Digitimes and Korean trade press says Samsung's own custom-chip talks with OpenAI, a system-on-chip effort distinct from OpenAI's Jalapeño inference chip, which Broadcom designed and TSMC manufactures, have stalled, and that Meta separately paused its own custom AI chip project with Samsung. Samsung Foundry has reportedly been looking for large logic-chip customers to fill that gap, and Anthropic's talks come as one candidate.
Anthropic has not signaled the custom chip would replace its existing compute arrangements. The company has said AWS Trainium, Google's TPUs, and Nvidia's GPUs remain central to how it scales, consistent with a multi-vendor approach that it has been building out under a $50 billion infrastructure push. A Samsung-built chip, if it materializes, would be an addition to that mix rather than a replacement, following the same logic that has pushed OpenAI, Google, Amazon, and Meta to each develop proprietary silicon alongside their existing GPU and cloud-chip supply.
Sources: The Information, Bloomberg, Digitimes
–
By the Control Plane Editorial Team