Anthropic Confirms It’s Building Custom AI Chips to Power Claude
Anthropic has confirmed it’s assembling an in-house silicon team to design custom chips for Claude, joining OpenAI and Meta in a costly bet to cut reliance on Nvidia.

Anthropic has confirmed it is building an internal team to design custom AI chips specifically for Claude — the first time the company has publicly acknowledged the effort, after months of reports that it was exploring the move.
Business Insider first reported the news, which Anthropic confirmed to TechCrunch and Reuters. A company spokesperson told Reuters that Anthropic is assembling a custom silicon team that will work on chips built specifically for Claude. The job listing for the effort pays between $320,000 and $485,000, and seeks engineers who have “shipped silicon” before. Change language to +2
The push comes down to compute. AI accelerators are the bottleneck for every frontier lab, and Anthropic’s demand for compute keeps climbing — the company recently signed a deal with SpaceX to double Claude’s usage limits. Anthropic’s annualized revenue has crossed $30 billion in 2026, up sharply from around $9 billion at the end of 2025. Industry sources estimate that developing custom silicon from scratch can cost roughly half a billion dollars — a steep bet, even for a company growing this fast. Change language to +2
Anthropic isn’t walking away from its existing partners, though. The company is continuing to run Claude across AWS Trainium chips, Google TPUs, and Nvidia GPUs, describing its approach as a “multi-chip” strategy rather than a full pivot to in-house hardware. Anthropic has not disclosed a timeline for when the custom chips might be ready, or whether it plans to manufacture them itself — or through a third-party foundry. aiweeklyfreepressjournal
The move puts Anthropic alongside a growing list of AI labs pursuing their own silicon. OpenAI has recently launched its own AI chip project, while Meta is developing next-generation AI processors of its own, and Mistral AI is reportedly exploring similar plans. The logic is the same across the industry: chips designed specifically around a company’s own model architecture can, in theory, run those exact workloads faster and more cheaply than general-purpose hardware — a meaningful edge as compute costs keep climbing industry-wide, and as the race for AI infrastructure increasingly becomes a race for raw electricity and silicon supply.


