Anthropic has opened preliminary discussions with Samsung Electronics about manufacturing a custom AI accelerator chip, according to a report from The Information published July 2, 2026. The talks are still at an early stage, with workloads, performance targets, and how the chip would fit into server racks all undecided.
The conversations center on Samsung's advanced 2-nanometer SF2P process, which uses gate-all-around transistors to cut power leakage and is expected to enter production later in 2026. Anthropic reinforced its chip ambitions by hiring engineer Clive Chan away from OpenAI's chip team in early June. The move mirrors OpenAI's own push into custom silicon: its Broadcom-built inference chip, Jalapeño, unveiled June 24, reportedly cut inference costs by roughly 50% compared with standard GPUs in early testing.
Anthropic said its existing hardware stack — chips from Amazon, Google, and Nvidia — "will continue to be pivotal" to its compute strategy, and the company has separately committed to a $50 billion U.S. data center buildout with Fluidstack and more than $100 billion in AWS infrastructure spending over the next decade. A custom inference chip could still cut costs by stripping away general-purpose overhead that GPUs carry for AI-specific workloads.