Anthropic explored acquiring AI chip startup MatX for approximately $7 billion before the negotiations ended, as the Claude developer looks to accelerate development of its own artificial intelligence hardware, according to Reuters. The discussions have since evolved toward a potential partnership between the companies.
Two people briefed on the matter told Reuters that Anthropic discussed the approximately $7 billion acquisition as part of its effort to bring more chip design talent and expertise in-house.
Reuters could not determine why the acquisition talks were abandoned.
MatX is now seeking new financing at a valuation of approximately $4 billion, according to one of the sources. Anthropic declined to comment on its discussions with MatX, while MatX did not respond to Reuters’ request for comment.
MatX was founded by former Google tensor processing unit engineers and is developing chips designed for training large artificial intelligence models.
Acquiring a company with specialized chip design capabilities could have given Anthropic an established engineering organization and potentially shortened the timeline for developing proprietary hardware.
Anthropic is expanding an internal silicon team as the company attempts to make its Claude models faster and more efficient while reducing its dependence on outside semiconductor providers.
The company nevertheless plans to maintain a multi-chip strategy and continue working with suppliers including Nvidia and Google rather than completely replacing third-party hardware.
Building custom AI chips is an expensive and lengthy process. A single chip generation can require hundreds of millions of dollars in design costs, and producing a viable processor can take a year or longer.
Anthropic’s infrastructure requirements are increasing rapidly as usage of its Claude models grows.
The company plans to spend tens of billions of dollars on computing infrastructure and has made several major capacity commitments.
Those include plans to purchase $36 billion of Google’s AI chips, a $45 billion agreement to rent AI computing capacity from Nscale and an agreement to pay SpaceX $1.25 billion per month through May 2029 for computing capacity across its data center clusters.
Anthropic’s interest is also broader than MatX.
The company has recently met with multiple AI semiconductor startups to evaluate different approaches to chip design, but has not selected an acquisition strategy.
Those discussions indicate Anthropic is evaluating both training processors, which are used to build AI models, and potentially inference chips designed to efficiently generate responses after models have been trained.
Anthropic has been hiring experienced semiconductor executives and engineers as part of the effort.
The company recently recruited Google chip veteran Amir Salek and in June hired former OpenAI chip engineer Clive Chan.
The moves come as the leading AI developers increasingly view custom silicon as a strategic advantage.
OpenAI has developed its own Jalapeno processor, while Google has spent years developing TPUs and Amazon has built Trainium and Inferentia chips.
Anthropic has historically taken a diversified approach to infrastructure by operating its models across hardware supplied by Nvidia, Google and Amazon.
Developing its own chips could eventually give Anthropic more control over hardware specifically optimized for Claude while potentially reducing long-term computing costs.
It could also provide another source of capacity at a time when Nvidia has warned that supplies of its processors could remain constrained through 2027.
The MatX discussions therefore highlight how strategic semiconductor technology has become for frontier AI companies. Even though the proposed $7 billion acquisition did not proceed, Anthropic’s continuing talks with MatX and other chip startups suggest proprietary hardware is becoming an increasingly important part of its long-term infrastructure strategy.
