AMD invests $5 billion in Anthropic and secures a 2-gigawatt MI450 chip deal

Advanced Micro Devices and Anthropic announced on July 22 that they have struck a landmark deal: AMD will invest up to $5 billion in the Claude maker while Anthropic commits to deploying up to 2 gigawatts of AMD Instinct MI450 Series GPUs starting in the first half of 2027. It is one of the largest chip procurement agreements ever signed in the AI industry.

The partnership covers far more than a hardware purchase. Anthropic is adopting AMD's full-stack Helios rack-scale infrastructure, which combines MI455X GPUs with 6th-generation EPYC "Venice" CPUs, Pensando networking, and AMD's ROCm software platform. The two companies have also signed a separate engineering collaboration agreement focused on optimizing ROCm for Claude model training and inference workloads.

The end of Nvidia's monopoly on frontier AI compute

The scale of this deal sends a clear competitive signal. Two gigawatts of high-density AI compute translates to a hardware commitment in the range of $30 billion or more — making this arguably the most significant challenge to Nvidia's dominance in AI infrastructure since the AI era began.

AMD CEO Dr. Lisa Su framed the partnership as a mutual bet on the future of compute diversity: "This collaboration brings together Anthropic's leadership in frontier AI with AMD's high-performance computing strength." Anthropic co-founder and Chief Compute Officer Tom Brown added that "running across diversified hardware lets us map appropriate workloads to the right infrastructure" — a pointed acknowledgment that over-reliance on a single vendor carries risk at frontier scale.

AMD is already building momentum beyond this announcement. The company signed a 6-gigawatt multi-year agreement with Meta in February 2026, and just a day before the Anthropic deal, it unveiled the MI455X and Helios rack system at its Advancing AI 2026 event, listing Microsoft, Meta, OpenAI, and Oracle as Helios customers. The Anthropic partnership adds the fifth major AI lab to that roster.

What Anthropic gets — and why now

Anthropic currently uses AMD's MI355X GPUs for some workloads, so the MI450 commitment is an acceleration and deepening of an existing hardware relationship. For Anthropic, the deal solves a critical constraint: access to compute at the scale needed to keep Claude competitive with models from OpenAI and Google. As Tom Brown noted, "access to compute is central to keeping Claude at the frontier."

The $5 billion AMD equity investment — contingent on certain milestones — also injects capital into Anthropic at a moment when frontier AI training costs are running into the hundreds of millions of dollars per major model release. Pairing that with guaranteed chip supply from a non-Nvidia source gives Anthropic supply-chain resilience that rivals built on Hopper and Blackwell alone cannot match.

A structural shift in AI infrastructure

The broader implication is a restructuring of the AI chip market that began with Meta's AMD commitment earlier this year. As the largest AI labs diversify away from Nvidia's H100 and B200 GPUs, AMD's ROCm ecosystem gains real-world scale — making it increasingly viable for smaller labs and enterprises to follow suit without accepting the performance penalty that once came with choosing AMD over Nvidia.

The first gigawatt of AMD infrastructure is expected to come online for Anthropic in H1 2027. When it does, it will represent one of the largest non-Nvidia AI compute deployments in history, as first reported by AMD's official investor relations release.