AMD is putting $5 billion into Anthropic and is allocating 2 gigawatts of Helios GPUs to operate Claude.

      **Summary**: AMD plans to invest up to $5 billion in Anthropic and will deploy 2GW of MI450 GPUs. The deployment of the first gigawatt is expected in the first half of 2027. Anthropic is already utilizing MI355X GPUs.

      On Tuesday, AMD and Anthropic revealed a strategic partnership that includes a commitment from AMD to invest up to $5 billion in Anthropic while deploying up to 2 gigawatts of AMD Instinct MI450 Series GPUs within Helios rackscale solutions to support Claude. The rollout of the first gigawatt is set to commence in the first half of 2027, and Anthropic is currently using AMD’s MI355X GPUs, with plans to upgrade to MI455X accelerators in conjunction with EPYC “Venice” CPUs and Pensando networking.

      The collaboration on engineering may prove more significant than the hardware itself. AMD and Anthropic aim to utilize Claude to optimize workloads for AMD Instinct GPUs and enhance ROCm software development. Furthermore, AMD will implement Claude across its engineering and product development teams. ROCm serves as AMD’s counterpart to Nvidia’s CUDA, and the software gap has been a key factor driving AI developers towards Nvidia hardware, even when AMD offers competitive specs. If Claude can significantly enhance the developer experience with ROCm, AMD may finally resolve an issue that has hindered its progress by leveraging its customer's AI to improve its software.

      Anthropic’s computing strategy is now genuinely multi-vendor. The company has signed its largest computing deal with Google and Broadcom, and has separate agreements with Amazon (for Trainium chips, 5GW), CoreWeave (Nvidia GPUs), and SpaceX (Colossus data centers). The addition of 2GW of AMD Helios allows Anthropic to fulfill what Tom Brown, its chief compute officer, describes as the ability to “map the right workloads to the right hardware.” By diversifying its chip vendor options, Anthropic reduces dependency risks at a time when compute resources are a critical constraint in developing advanced models.

      For AMD, this $5 billion equity investment mirrors Nvidia’s strategy, as Nvidia invested $2 billion in Nebius and acquired stakes in various AI infrastructure companies to secure hardware demand. AMD is taking a similar approach by investing in a customer to ensure that its chips, rather than Nvidia’s, power a significant portion of the world’s leading AI models. Lisa Su referred to it as “a major platform for the next generation of AI infrastructure.” The ability of Helios to compete with Nvidia’s NVL72 at a production scale is a question that the deployment of the first gigawatt will address.