Anthropic is exploring building its own AI chips

The plans are early-stage and Anthropic may still decide to only buy chips rather than design them. The exploration comes days after the company signed a long-term deal with Google and Broadcom for 3.5 gigawatts of TPU compute starting in 2027. A company spokesperson declined to comment.

Anthropic is exploring the possibility of designing its own AI chips, Reuters reported on Thursday, citing three sources familiar with the matter. The effort is at an early stage: the company has not committed to a specific design and has not assembled a dedicated team for the project.

It may still decide to continue purchasing chips from third parties rather than building its own. A spokesperson for the San Francisco-based company declined to comment on the report.

The exploration comes as Anthropic’s revenue has accelerated sharply. The company disclosed earlier this week that its annualised revenue run rate has surpassed $30 billion, up from approximately $9 billion at the end of 2025.

That trajectory has created a scale of compute demand that makes the economics of custom silicon increasingly worth examining. Anthropic currently runs Claude across a mix of chips: tensor processing units designed by Alphabet’s Google, in partnership with Broadcom, alongside Amazon’s custom chips and Nvidia hardware.

The company said it matches workloads to whichever chips are best suited for them.

Just days before the Reuters report, Anthropic signed a long-term deal with Google and Broadcom that will give it access to approximately 3.5 gigawatts of TPU-based compute capacity from 2027, roughly three times the roughly one gigawatt it was consuming earlier in 2026, according to Broadcom’s SEC filing.

The filing flagged that the expanded deployment is contingent on Anthropic’s continued commercial success, an unusual hedge for a regulatory document. The deal builds on Anthropic’s November 2025 commitment to invest $50 billion in US computing infrastructure.

Broadcom is also already a chip design partner for OpenAI, and has a fifth undisclosed XPU customer, placing it at the centre of the custom AI silicon market that is emerging as an alternative to Nvidia’s general-purpose GPUs.

The possibility of Anthropic developing proprietary silicon mirrors moves already underway elsewhere in the industry. Meta has been building its own AI training chips, and OpenAI has been working on custom silicon as well.

Industry sources cited by Reuters put the development cost of an advanced AI chip at roughly $500 million, reflecting the need to hire specialised engineers and validate the manufacturing process.

That figure is not trivial for a company that remains, for now, unprofitable, but it is more manageable against a run-rate revenue base that has more than tripled in four months.