Anthropic weighed a $7B buy of chip startup MatX

Anthropic discussed buying AI chip startup MatX for roughly $7 billion to speed up its own custom hardware work, according to two people briefed on the matter. Those merger talks are no longer active, and a third person said they have since turned into a discussion about a partnership, Reuters reported.

At a glance

• MatX, founded by former Google TPU engineers, has been working on a processor aimed at training large AI models, and is now seeking fresh capital at a valuation of about $4 billion.
• Anthropic's engineers and executives have met a range of AI chip startups in recent weeks to survey current design approaches, and the company has not yet elected to make an acquisition.
• Designing a chip generation costs hundreds of millions of dollars and takes a year or more to yield viable hardware; an acquisition would give Anthropic in-house expertise and could lower costs over time.

The pattern is familiar: labs that have outgrown their compute budgets try to buy their way into silicon design rather than assemble a team from scratch. For Anthropic the pressure appears sharper than for most, because a listing at the valuation being discussed rests on revenue targets that only abundant and cheap training capacity can support. A partnership rather than an acquisition would be a slower version of the same bet.

Anthropic's IPO is expected to chase a $2 trillion valuation

Anthropic is expected to list this year, months after SpaceX went public with a $1 trillion valuation, and the offering will chase a valuation of $2 trillion, Reuters reported earlier this month. That figure hinges on a 2028 revenue number of as much as $200 billion.

The company says it is expanding an in-house silicon team to design custom chips so Claude models run faster and more efficiently, while keeping a multi-chip approach with providers from Nvidia to Google. This week it hired Google chip veteran Amir Salek, and in June it brought in former OpenAI chip engineer Clive Chan, who worked on the OpenAI chip unveiled earlier this year.

Anthropic has committed $36 billion to Google chips and $45 billion to Nscale

Anthropic plans to spend tens of billions of dollars renting computing power from cloud providers and to buy chips directly. It has committed to buying $36 billion worth of Google's AI chips and signed a $45 billion deal to rent AI cloud computing power from Nscale. The company has been seeking more capacity to meet surging demand for its models.

It also agreed to pay SpaceX $1.25 billion per month through May 2029 for computing capacity across its data center clusters, including the Colossus 1 facility, which houses more than 220,000 Nvidia chips. Anthropic was among the first companies to run its models on hardware from several vendors including Nvidia, Google and Amazon.

MatX targets training, while OpenAI's Jalapeno targets inference

MatX has been working on a chip useful for building large AI models through training, which suggests Anthropic may be interested in producing a training chip while other chip startups and OpenAI pursue processors better suited to inference. Anthropic could elect to produce an inference chip as well, the sources said.

At a conference this week OpenAI said its first custom chip, called Jalapeno, outperformed a comparable Nvidia processor, with executives noting it was more energy efficient on inference calculations. Nvidia said on a conference call Wednesday that its processors will be in short supply through 2027. Google has developed TPUs, and Amazon has built its Trainium and Inferentia chips.

Why the MatX talks went inactive

Reuters, which reported the discussions first, could not learn why the talks are no longer active. The terms of a possible partnership have not been described, and no timetable has been given for whether Anthropic will acquire a chip startup at all or when its first in-house silicon might reach production.