Anthropic Dropped a Reported $7 Billion MatX Deal and Is Exploring a Partnership

Anthropic appears to be looking beyond simply buying more computing capacity. Reuters reported that the Claude maker discussed acquiring AI-chip startup MatX for roughly $7 billion, then abandoned the proposed purchase and shifted toward exploring a partnership. 1

The reported change does not mean Anthropic has given up on custom silicon. It suggests the company is still weighing how much of its hardware stack it should own, how quickly it can develop specialized chips, and how that effort fits with its major Amazon infrastructure agreement.

What Reuters reported about the MatX talks

According to Reuters, Anthropic held discussions about acquiring MatX for approximately $7 billion to accelerate the development of custom hardware for its growing AI business. The acquisition plan was later dropped, while the discussions evolved into a possible partnership. Reuters also reported that Anthropic was meeting with multiple chip startups and had not selected a definitive approach. 1

MatX was reportedly seeking fresh financing at a valuation of about $4 billion. 1 The difference between that fundraising valuation and the reported acquisition figure does not, by itself, establish the reason the deal was abandoned. The available reporting indicates that the parties moved away from a full purchase, but it does not confirm that a partnership was finalized.

What MatX does

MatX is an AI-chip startup founded by former Google tensor processing unit, or TPU, engineers. Its focus is custom AI accelerators: specialized processors designed to handle the demanding workloads involved in training and operating large AI models.

That expertise could be valuable to Anthropic because model performance depends on more than the model itself. The underlying chips, software, networking, and data-center systems all influence cost, speed, capacity, and reliability. A relationship with a specialist such as MatX could give Anthropic access to chip-design talent without requiring it to absorb the full cost and complexity of an acquisition.

Why Anthropic wants custom hardware

Anthropic’s reported interest in MatX reflects a broader infrastructure problem for AI developers. Advanced models require enormous and continuing amounts of compute, while access to leading third-party chips can be expensive and constrained.

Custom hardware could give Anthropic more control over the systems that train and run Claude. In principle, purpose-built accelerators can be optimized for a company’s particular model architecture and workloads. They may also help diversify supply and reduce reliance on dominant external chip vendors.

Those benefits are not guaranteed. Designing, manufacturing, deploying, and supporting advanced chips requires substantial capital and specialized engineering. A partnership can therefore offer a middle path: Anthropic could pursue hardware tailored to its needs while limiting the integration and execution risks of buying an entire chip company.

The Amazon agreement complicates the picture—but does not rule it out

Anthropic’s custom-chip ambitions would coexist with its deep relationship with Amazon. Reuters reported that Amazon agreed to invest up to $25 billion in Anthropic, while Anthropic committed to spending more than $100 billion on Amazon’s cloud technologies over 10 years. 5

The arrangement also gives Anthropic access to Amazon-designed Trainium chips and large-scale AWS infrastructure. 5 Amazon’s own announcement described access to as much as 5 gigawatts of current and future Trainium capacity for training and deploying Claude. 9

That makes a MatX partnership potentially complementary rather than automatically competitive with Amazon. Anthropic could continue using AWS and Trainium while developing or evaluating other specialized hardware. The reported talks do not show that Anthropic intends to replace Amazon; they show that the company is considering additional control and optionality within its compute strategy.

The financial backdrop is unusually ambitious

The MatX discussions emerged as Anthropic’s business was expanding rapidly. Reuters reported that the company’s annualized revenue run rate topped $65 billion by the end of July. A run rate is an extrapolation of current performance, not the same thing as booked annual revenue, so the figure should be read as an indicator of pace rather than a completed-year result.

Reuters separately reported that Anthropic was projecting roughly $190 billion to $200 billion in revenue for 2028 as investors assessed the company ahead of a potential IPO. Those figures help explain why Anthropic would be evaluating large, long-term infrastructure bets—but they remain reported projections, not guaranteed outcomes.

What the episode says about the AI industry

The reported MatX talks are part of a larger move toward vertical integration in AI. Companies increasingly want influence over several layers of the stack: chips, servers, cloud capacity, models, and applications.

Control over more layers can potentially improve performance, supply assurance, and cost management. It can also create tighter coordination between hardware and software. But vertical integration demands substantial spending and exposes companies to execution risks that would otherwise sit with suppliers.

For Anthropic, the immediate conclusion is narrower than “Anthropic is building its own chip company.” Reuters reported an abandoned acquisition discussion, a possible partnership, and ongoing conversations with other chip startups—not a completed transaction or a final hardware plan. 1 The strongest signal is that custom silicon has become strategically important enough for Anthropic to consider multiple ways to secure it.