Anthropic's planned custom AI chips could be manufactured on Samsung's 2nm process, according to a report by SammyGuru that links the AI developer to the Korean foundry's most advanced production node.

The report does not state that a deal has been signed. It indicates that Samsung's 2nm process is a candidate manufacturing route for silicon Anthropic is designing for its own AI workloads. If the arrangement moves forward, Anthropic would join the ranks of AI companies building proprietary accelerators rather than relying solely on merchant suppliers.

What does the report actually claim?

The core claim is narrow: Anthropic's custom AI chips — chips the company is developing for its own use — could be produced using Samsung's 2nm-class fabrication technology. The report attributes the possibility to industry information but provides no contract terms, no volume figures, and no production timeline.

Readers should treat the story as a supply-chain indication rather than a confirmed foundry engagement. Foundry assignments at leading-edge nodes are typically finalized only after lengthy design, qualification and yield-validation cycles, none of which the report describes.

Why does the 2nm detail matter?

Samsung has positioned its 2nm process family as its most advanced logic platform, intended for high-performance and power-efficient compute customers. An AI chip developer choosing that node would signal a design targeting top-tier performance per watt — the metric that dominates training and inference economics.

For Samsung's foundry business, an Anthropic engagement would add a high-profile AI customer at a node where the company is competing directly with TSMC for leading-edge orders. SammyGuru's report does not say whether other foundries are also under consideration.

For Anthropic, custom silicon would reduce dependence on third-party accelerator suppliers, whose capacity has been the binding constraint across the AI sector. The report does not specify which workloads the chips would serve — training, inference, or both.

What remains unconfirmed?

Several questions stay open:

• Whether Anthropic and Samsung have a formal agreement or are in earlier discussions
• Which 2nm variant Samsung would use for the chips
• When tape-out or volume production could begin
• How the silicon would compare with the GPUs Anthropic currently uses
• Whether the design work is internal or involves a partner

The report offers no answers on these points. It establishes only that Samsung's 2nm process is a plausible manufacturing path for Anthropic's custom AI chips.

The wider read

For the semiconductor market, the signal is directional: AI developers continue to push toward in-house silicon, and leading-edge foundry capacity is the prize. Any confirmed Anthropic order would land in a segment Samsung has targeted explicitly — AI and high-performance computing customers for its 2nm-class technologies.

Until Anthropic or Samsung confirms the engagement, the report stands as an unverified but market-relevant indication of where Anthropic intends to have its custom accelerators built.

Die Signal will update this story as manufacturers or the company respond.