Anthropic, the artificial intelligence company behind the Claude family of large language models, has begun preliminary discussions with Samsung Electronics to develop and manufacture a custom AI chip, according to a report from The Information published Thursday. The two companies are discussing the use of Samsung Foundry's leading-edge 2-nanometer process technology and advanced chip packaging capabilities, which would allow Anthropic to position processors in closer proximity to memory chips, boosting data throughput and reducing latency bottlenecks in AI inference workloads.
The development is early-stage: Anthropic has not finalized chip specifications, performance targets, or an architectural design, and no manufacturing agreement has been signed. The company told TechCrunch it had nothing further to add on the Samsung discussions, while noting that a diversified hardware stack drawing on chips from Google, Amazon, and Nvidia remains central to its compute strategy. Separately, Anthropic said that demand for Claude continues to grow and that the relationships it has cultivated with semiconductor partners will help scale its computing infrastructure at the pace its customers require.
Anthropic's hardware ambitions appear to have accelerated meaningfully in recent months. The company hired Clive Chan last month, a technical staff member with prior experience on application-specific integrated circuit development at both Tesla and OpenAI. In May, Samsung Electronics participated in Anthropic's $65 billion Series H funding round as a strategic infrastructure partner, alongside two other of the world's largest memory chip manufacturers, a relationship that industry observers in South Korea viewed as a natural precursor to a foundry engagement.
The timing adds context to a competitive dynamic that has been shifting rapidly across the AI chip market. Last week, rival AI lab OpenAI announced a custom inference processor, dubbed "Jalapeño," developed in partnership with Broadcom, citing better performance per watt compared with competing silicon. Anthropic's Samsung discussions appear partly responsive to that development, as the company evaluates whether to pursue a similar path toward hardware self-sufficiency rather than relying indefinitely on third-party GPU suppliers.
For Samsung Foundry, the potential Anthropic engagement would add another major AI client to a roster that already includes Tesla, whose AI chip order Samsung won last year, and Meta, which is reportedly in discussions with Samsung for manufacturing of its third-generation MTIA AI accelerator. Industry analysts in Seoul expect Samsung Foundry's medium- to long-term order backlog to approach 50 trillion won as these AI infrastructure orders accumulate, with foundry profitability potentially turning positive as early as the fourth quarter of 2026.
The broader read-through for publicly traded names is significant. Any diversification of AI compute away from Nvidia's GPU architecture — whether through Broadcom-designed ASICs for OpenAI or Samsung Foundry-manufactured custom silicon for Anthropic — incrementally dilutes the total addressable market that Nvidia would otherwise capture exclusively. Investors in Nvidia, which has retained more than 80% of the AI accelerator market on a revenue basis, are watching these ASIC deals closely as the clearest leading indicator of how much of the AI infrastructure buildout will ultimately flow through GPU architectures versus purpose-built alternatives. Samsung Electronics' dual position as both a foundry manufacturer and the world's second-largest HBM memory producer gives it an unusual advantage in attracting AI clients seeking integrated hardware solutions across compute and memory stacking.