Anthropic has begun early discussions with Samsung Electronics about manufacturing a custom AI server chip, according to reports citing The Information. The talks are described as preliminary: Anthropic has not finalized a design, started manufacturing, or committed to a deployment schedule.

The reported plan would add another hardware option to Anthropic's compute stack. Recent coverage says Samsung's 2-nanometer process and advanced packaging are under consideration, but the most important point is the uncertainty. For now, this is a potential foundry relationship, not a signed chip program.

Even so, the talks fit a clear pattern among frontier AI labs. In April, Anthropic announced a separate agreement with Google and Broadcom for multiple gigawatts of next-generation TPU capacity starting in 2027. A Samsung-made chip would not replace that deal, but it would give Anthropic another route to lower inference costs and reduce dependence on any single supplier.

Samsung also has an obvious reason to pursue the work. Its foundry business is trying to win more advanced AI customers as TSMC remains the dominant manufacturer for high-end accelerators. Landing a Claude-focused chip program would give Samsung a visible foothold in one of the most compute-hungry parts of the AI market.

The cautious read is that Anthropic is keeping optionality open. The company already buys or reserves capacity from several chip ecosystems, and early chip talks do not guarantee production. But the direction is notable: model labs are increasingly treating silicon access as part of product strategy, not just infrastructure procurement.