Anthropic has begun developing its own proprietary AI chip and held exploratory talks with Samsung Electronics as a potential manufacturing partner, according to three people with direct knowledge of the discussions.
The project remains at an early conceptual stage. Anthropic is still determining what the chip should do, how powerful it should be, how it would fit into a server, and what process node to target. No design work or manufacturing commitments have been made, and the company may decide not to proceed.
The clearest signal of intent is a personnel move: Anthropic hired Clive Chan, an early member of OpenAI’s custom-chip team behind the Broadcom-designed “Jalapeño” inference chip that OpenAI unveiled in June. Anthropic has also spoken with multiple chip-design firms beyond Samsung.
The timing follows a broader capital alignment between the two companies. Samsung participated in Anthropic’s US$65 billion (approximately £52 billion) Series H funding round in May, alongside SK Hynix and Micron, at a valuation of US$965 billion (approximately £780 billion). Anthropic’s press release at the time described the three as “memory, storage and logic-chip partners,” phrasing that fueled speculation in Seoul that Samsung’s role could extend beyond memory supply to fabricating the custom silicon that powers Claude.
Among the three memory partners, only Samsung operates a contract foundry business. Samsung Foundry has run at a loss for several years but has recently won orders including Tesla’s next-generation AI5 and AI6 chips and Nvidia’s Grok3 inference processor. It ranks second in the global foundry market with a 7.2 percent share, trailing TSMC by roughly 63 percentage points.
Anthropic’s official position, communicated to The Information, is that AWS Trainium, Google TPUs, and Nvidia GPUs “will remain central” to how it scales compute, a statement compatible with a long-horizon internal chip program running in parallel.
The move mirrors a broader industry trend. OpenAI unveiled its first custom inference chip built by Broadcom in June, and every frontier AI lab has been moving toward vertical silicon integration as inference demand drives up compute costs. For Anthropic, owning the chip design would let it tune power, memory bandwidth, and interconnect to its own model architectures while reducing dependence on Nvidia’s pricing leverage for inference workloads.
Sources: Anthropic in talks with Samsung to manufacture custom AI chip (The Information, July 2, 2026); Samsung, SK hynix invest in Anthropic (The Investor, May 29, 2026); OpenAI unveils its first custom chip (TechCrunch, June 24, 2026)