AMD and Anthropic announced a multibillion-dollar partnership on July 22, 2026, in which the AI company will deploy up to 2 gigawatts of AMD’s next-generation Instinct MI450 chips. AMD committed to invest as much as $5 billion in Anthropic as part of the deal.

The arrangement marks a significant shift in how chipmakers back AI companies. Unlike previous deals between AMD and OpenAI or AMD and Meta—which included equity warrants—this Anthropic agreement contains no warrant provisions, signaling a new dynamic in the semiconductor-AI relationship.

The Computing Deal

Anthropic will purchase tens of billions of dollars’ worth of AMD’s AI server chips over time. The first gigawatt of deployment begins in the first half of 2027. The chips will be housed in AMD Helios rack-scale solutions featuring the Instinct MI455X GPUs alongside AMD EPYC “Venice” CPUs, Pensando networking, and ROCm software.

The 2-gigawatt commitment represents substantial hardware infrastructure. For perspective, major data centers typically consume between 100 and 300 megawatts. Two gigawatts would power the equivalent of 6 to 20 large data center clusters.

Strategic Positioning

The deal signals AMD’s ambition to compete with Nvidia in the high-performance AI chip market. Nvidia dominates this segment with its H100 and newer Blackwell GPUs, but AMD’s MI series offers price advantages and flexibility for specific workloads.

For Anthropic, the agreement ensures reliable access to cutting-edge processing power as the company scales its Claude models. Long-term partnerships in semiconductor supply are increasingly critical as AI companies face chip shortages.

Investment Structure

AMD’s $5 billion investment is contingent on Anthropic meeting certain milestones. This performance-based approach differs from traditional venture capital rounds, where money transfers upfront.

The deal demonstrates how AI companies and chipmakers are converging—sharing risk, aligning incentives, and building the infrastructure that powers the next wave of language models.