Anthropic Explores Chip Startup Acquisition and Partnerships to Accelerate In-House Hardware Development

Artificial intelligence lab Anthropic discussed acquiring AI chip startup MatX for roughly $7 billion as part of its strategy to accelerate the development of custom hardware. According to sources familiar with the matter, the merger talks eventually evolved into discussions regarding a potential partnership.

MatX, which was founded by former Google tensor processing unit engineers, is currently seeking to raise new capital at a valuation of approximately $4 billion. Anthropic declined to comment on the deal talks, and MatX did not respond to requests for comment.

As Anthropic scales its Claude family of AI models, the company is aiming to produce custom hardware capable of handling high data-crunching demands while reducing its reliance on Nvidia chips. Nvidia recently noted that its processors are expected to remain in short supply through 2027. Anthropic is expanding its in-house silicon team to design chips that enable Claude models to run faster and more efficiently, while maintaining a multi-chip approach alongside providers like Google and Amazon.

Designing custom silicon is capital-intensive and time-consuming. Producing a viable piece of hardware can take over a year, with design costs for a single generation running into hundreds of millions of dollars. Acquiring a startup like MatX would provide Anthropic with established in-house design expertise and potential long-term cost savings. MatX has focused on building chips specifically for training large AI models, whereas other startups and rivals like OpenAI have concentrated on inference processors designed for generating chatbot responses.

In addition to exploring acquisitions, Anthropic has held meetings with several other AI chip startups to understand various design approaches. The company has also brought on industry talent, hiring former Google chip veteran Amir Salek and former OpenAI chip engineer Clive Chan.

Anthropic's infrastructure investments extend beyond hardware design. The company plans to spend tens of billions of dollars on computing power, including agreements to purchase $36 billion worth of Google's AI chips and a $45 billion deal to rent cloud computing power from Nscale. It also agreed to pay SpaceX $1.25 billion per month through May 2029 for computing capacity across data center clusters, including the Colossus 1 facility housing over 220,000 Nvidia chips. Meanwhile, Anthropic is reportedly preparing for an initial public offering, eyeing a valuation of up to $2 trillion by 2028.

"The pursuit of custom silicon by major AI labs like Anthropic highlights a critical shift in the technology sector toward infrastructure independence. Developing in-house chips is an expensive and lengthy process, but heavy reliance on external suppliers creates supply chain and cost vulnerabilities as demand for computing power surges. Strategic discussions and talent acquisitions in the hardware space indicate that long-term scalability in AI will depend heavily on owning the underlying compute stack." — Dr. Shishir Gupta, Founder & CEO, StartupLanes