AIBRIEF
A spokesperson confirmed a 'custom silicon' team after a job listing surfaced; the lab says it will keep buying third party chips alongside its own designs.
Anthropic confirmed Wednesday that it is building an in-house silicon team, joining OpenAI, Google, and Meta as frontier AI labs that now design some of the silicon that runs their own models.
A spokesperson for Anthropic told both Business Insider and TechCrunch that the company plans to hire engineers for a "custom silicon" effort, after a job listing for a silicon engineer and a silicon-focused technical program manager appeared on the company's careers page. The confirmation follows Ars Technica and earlier reporting from The Information that Anthropic was weighing a manufacturing partnership with Samsung, a link the company has not confirmed.
The Anthropic spokesperson was explicit that the move is not a full vertical bet. Anthropic will keep using third-party hardware alongside its own designs, describing a "multi-chip approach" that mirrors how Google runs its TPUs next to merchant GPUs. OpenAI made a similar disclosure last month with its Jalapeño inference chip, developed with Broadcom; Mistral is also reportedly exploring in-house silicon.
The confirmation puts Anthropic in step with the rest of the frontier. The AI compute stack is fragmenting by workload: specialized inference silicon sitting next to general-purpose GPUs rather than consolidating under a single vendor or in-house design.
What remains undisclosed: which foundry would manufacture Anthropic's chips, and when a first design would tape out. The Information's Samsung reporting stays a rumor for now.