Anthropic is in deeper talks with Samsung Electronics on a custom AI chip project, according to TechCrunch, which cited a report from The Information. The discussions are focused on Samsung’s 2-nanometer process and advanced packaging capacity, with the design aimed at handling Claude inference and training workloads. If the effort moves forward, it would become Anthropic’s first in-house custom chip.
Chain News had previously reported on July 3 that the two sides were discussing this direction. The latest update adds detail on the makeup of the design team and confirms the process node now under consideration.
Clive Chan, hired from OpenAI, is leading the chip team
Anthropic hired former OpenAI hardware engineer Clive Chan in June. According to the report, Chan was the second person to join OpenAI’s internal custom chip project, which was being developed with Broadcom, and he had been involved in the architecture design from the early stages.
On June 7, Chan said on social media that he had left OpenAI and joined Anthropic. His arrival gives Anthropic a key engineering hire as the company turns its custom chip plans into a more structured effort.
Samsung was already a strategic infrastructure investor in Anthropic
Samsung participated in Anthropic’s $65 billion Series H fundraising round in May 2026. It was listed alongside SK Hynix and Micron as a strategic infrastructure investor.
Among those three companies, SK Hynix and Micron are memory-focused businesses, while Samsung is the only one that also owns a foundry operation through Samsung Foundry and can manufacture custom ICs for other companies. That structure made Samsung a logical production partner for Anthropic’s chip ambitions even at the financing stage.
2-nanometer manufacturing is the current focus, but the project is still early
On the technology side, Samsung’s most advanced mass-production node is 2 nanometers, and that is the node Anthropic is now discussing for the custom chip.
The project remains in an early stage. Anthropic has not yet finalized the chip’s exact functions, its compute targets, or how it will be integrated with server architecture.
Custom silicon is meant to add to, not replace, Anthropic’s hardware mix
Anthropic has said its in-house chip will not replace its current multi-vendor hardware strategy. The company plans to keep using Google TPUs, AWS Trainium and Nvidia GPUs.
Instead, the custom chip is being positioned as an added layer in Anthropic’s overall compute stack, tuned specifically for Claude so the company has more flexibility on inference cost and training efficiency.
More broadly, after OpenAI’s work with Broadcom on custom silicon, Anthropic is also moving toward a mixed supply-chain model that combines in-house chip design with general-purpose GPUs.