Anthropic explored acquiring MatX, an AI chip startup founded by former Google engineers, in a deal valued at roughly $7 billion — before changing its mind and walking away. The two companies are now reportedly in discussions over a design partnership instead, a shift that speaks to how the largest AI labs are approaching hardware strategy in 2026.

The abandoned bid stands out for its sheer size. A $7 billion offer would have represented a significant premium over MatX's current valuation, which landed in the multi-billion dollar range after a $500 million Series B round in February 2026. That round was led by Jane Street and Situational Awareness LP, and it transformed a startup barely three years old into one of the most closely watched chipmakers in the AI space.

Why Anthropic wanted MatX, and why it walked

The logic behind the proposed acquisition was straightforward. Anthropic, which has built its business around the Claude family of models, needs specialized hardware to run those models efficiently at scale. Buying MatX would have given the AI lab instant access to a team of chip designers with deep Google TPU pedigree, as well as a product — the MatX One — specifically engineered for large language model workloads.

MatX was founded in 2023 by engineers who cut their teeth building Google's Tensor Processing Units, the custom chips that power much of Google's AI infrastructure. The MatX One chip is designed to deliver higher throughput and lower latency for LLM inference and training. Shipments are expected to begin in 2027.

So why did Anthropic walk away from the deal? Neither company has commented publicly on the discussions, which leaves the reasons behind the abandoned acquisition open to interpretation. What is clear is that the move away from a full acquisition toward a possible partnership fits a broader pattern across the AI industry: access to custom silicon matters, but so does keeping options open while chip supply, model demand, and infrastructure plans continue to evolve.

Anthropic's broader silicon strategy

The MatX talks did not happen in a vacuum. Anthropic has confirmed that it is assembling its own internal custom silicon team — a move that signals a long-term commitment to owning more of its hardware stack. The company is also investing up to $5 billion in compute partnerships, including deals with AMD, as part of an effort to diversify its chip supply beyond Nvidia.

For Anthropic, which raised $8 billion from Amazon in late 2024 and early 2025, the financial resources to pursue custom silicon are clearly available.

A design partnership could take several forms. MatX might license chip architecture elements to Anthropic. The two companies could co-develop a processor optimized specifically for Claude model architectures. Alternatively, Anthropic could act as a design customer, funding MatX's R&D in exchange for early access and customization rights to the MatX One or future chips. Any of those arrangements would let Anthropic gain more control over performance and supply planning without taking on the integration burden of a full acquisition.

What this means for the AI chip market

MatX is positioning itself as a direct competitor to Nvidia in the AI accelerator market. The founding team's TPU background gives the startup credibility that most Nvidia challengers lack, and its $500 million war chest provides the runway needed to actually execute on a chip tape-out.

The fact that Anthropic was willing to pay $7 billion — a figure reportedly well above MatX's last private valuation — illustrates how much AI labs value custom silicon capabilities at this moment. It also underscores how hardware has become a core strategic layer for frontier model developers, not just a procurement line item.

For MatX, the outcome may ultimately prove better than being acquired. Remaining independent preserves optionality: the company can partner with Anthropic while continuing to sell chips to other AI labs, cloud providers, and enterprises. An acquisition, by contrast, would have locked MatX's technology inside a single customer's ecosystem.