Anthropic is getting serious about the chips that power artificial intelligence. The company behind Claude explored buying AI chip startup MatX for roughly $7 billion before those acquisition talks fell apart, according to Reuters.
The companies later discussed a possible partnership. The discussions still reveal something important about Anthropic’s plans. The AI company wants more control over the hardware needed to train and run increasingly powerful Claude models.
Building custom AI chips could help Anthropic reduce its dependence on outside suppliers. Nvidia remains a central force in AI computing, but booming demand has made access to advanced processors expensive and fiercely competitive. Anthropic is already spending heavily to secure computing capacity from several providers. Developing its own silicon could eventually give the company another source of computing power while allowing engineers to optimize hardware around Claude’s specific needs.
Anthropic Looked at a $7 Billion MatX Acquisition
The proposed acquisition could have given Anthropic experienced chip engineers without requiring the company to build an entire hardware team from scratch. Those talks were abandoned, and the reason for ending the proposed deal remains unclear.
MatX brings valuable experience to the table because its founders previously worked on Google’s tensor processing units, commonly called TPUs. Reiner Pope worked on software for Google’s TPU program, while co-founder Mike Gunter worked on the hardware side.
That background matters because Google’s TPUs have become a major alternative to Nvidia GPUs for artificial intelligence workloads. Engineers who helped develop such systems carry knowledge that AI companies increasingly want inside their own organizations. MatX is developing processors designed specifically for demanding large language model workloads.
The company also has plenty of financial backing. MatX raised more than $500 million in a Series B financing round in February 2026, with Jane Street and Situational Awareness leading the investment. MatX said at the time that its valuation had reached several billion dollars, although it did not publicly disclose an exact figure. The funding is expected to support development and manufacturing as the startup works toward shipping its chips.
MatX plans to produce its processors with Taiwan Semiconductor Manufacturing Company, better known as TSMC. The company has targeted 2027 for the first shipments, putting its hardware into an increasingly crowded race for AI computing customers.
Custom Chips Could Give Claude More Room to Grow
The hires include Amir Salek, a veteran of Google’s chip operations, and Clive Chan, who previously worked on chips at OpenAI. Bringing experienced engineers in-house could help Anthropic develop processors tailored specifically to its own AI workloads.
The economics behind that decision are hard to ignore. Training advanced AI models requires enormous clusters of specialized processors, while serving millions of users creates another constant demand for computing power.
Every Claude request has a computing cost attached to it. Those costs become especially significant for demanding AI products that handle coding, research, data analysis, and longer reasoning tasks.
Custom silicon could eventually help Anthropic improve performance while reducing the cost of each workload. Engineers could design chips around the operations Claude uses most instead of relying entirely on processors built for a broad range of customers.
The move would also give Anthropic another way to manage hardware shortages. Reuters reported that tight supplies of Nvidia processors are expected to remain an issue through 2027, making access to computing capacity a strategic concern for leading AI companies.