Anthropic walked away from 7B USD MatX chip deal

Anthropic discussed buying AI-chip startup MatX for roughly 7B USD, then abandoned the acquisition talks, Reuters reports, citing people briefed on the matter. A possible partnership is still being discussed, according to the report, but neither company has confirmed a deal.

MatX is interesting because it was founded by former Google TPU engineers and is building processors aimed at training large AI models. For Anthropic, buying that team would have been a shortcut into custom silicon at a time when Claude’s growth depends on expensive and scarce accelerator capacity.

The company is already building part of that capability internally. Reuters says Anthropic is expanding an in-house silicon team and recently hired chip veterans from Google and OpenAI. The work is still slow by nature: designing a chip can take a year or more and cost hundreds of millions of dollars for one generation.

The useful signal is not that Anthropic suddenly owns a chip company. It does not. The signal is that frontier model labs are treating hardware as strategic product infrastructure. Custom chips, supplier diversity and possible startup deals are becoming part of the same race to make large models cheaper and less dependent on any single vendor.