AMD and Anthropic plan up to 2 gigawatts of AI GPU capacity in $5 billion investment deal

AMD announced that Anthropic will deploy up to 2 gigawatts of AMD Instinct MI450 Series GPUs as part of a strategic partnership that includes up to $5 billion in AMD equity investment.

AMD and Anthropic set a large AI infrastructure plan

AMD said in a press release that Anthropic plans to use AMD Instinct MI450 Series GPUs in Helios rack-scale systems for training and running Claude models. According to AMD, the first gigawatt of capacity is scheduled to begin deployment in the first half of 2027.

The chipmaker also said it has committed up to $5 billion in strategic equity investment in Anthropic. Reuters reported that the investment may reach that amount depending on deployment milestones, and that AMD expects to sell Anthropic tens of billions of dollars of AI servers under the arrangement.

Anthropic, the company behind Claude, has been expanding its compute supply as demand grows for training and serving large AI models. AMD’s announcement frames the deal as a long-term partnership around AI infrastructure rather than a one-off hardware purchase.

A major customer win for AMD

For AMD, the agreement is another prominent AI infrastructure deal as it works to compete more directly with Nvidia in accelerator hardware. The Decoder described the Anthropic partnership as another major AMD deal following agreements involving Meta and OpenAI, while Reuters emphasized the scale of the planned server sales.

AMD’s Instinct MI450 Series is central to the announcement. The company said Anthropic will deploy the chips through Helios rack-scale systems, language that points to integrated data center systems rather than individual accelerators sold in isolation.

SiliconANGLE also reported that Anthropic will purchase up to 2 gigawatts of AMD GPU capacity and that AMD will invest up to $5 billion in the AI company. The publication added that AMD plans to use Claude for software development, a point also consistent with the companies positioning the deal as broader than hardware supply alone.

The deal raises questions about circular AI financing

The Decoder noted that critics view such agreements as potentially involving circular cash flows, where chipmakers invest in AI companies that then buy large amounts of their hardware. That concern has become more prominent as AI infrastructure deals increasingly combine equity investment, cloud capacity, and hardware commitments.

The available announcements do not show that the AMD-Anthropic deal is improper. They do, however, make clear that AMD is both a supplier and an investor in Anthropic. Reuters reported that AMD’s investment is tied to deployment milestones, which means the final investment amount may depend on the scale and timing of Anthropic’s hardware rollout.

Deployment begins in 2027

The timeline is significant. AMD said the first gigawatt is expected to start in the first half of 2027, meaning the full 2-gigawatt deployment, if completed, would unfold over time. The sources do not state that the entire planned capacity is already installed or immediately available.

The agreement underscores how leading AI developers are seeking large, multi-year supplies of accelerators for model training and inference. It also shows AMD pursuing large anchor customers for its next generation of Instinct GPUs as competition in AI chips remains heavily shaped by access to power, servers, and long-term supply commitments.

Based on the AMD announcement, Reuters reporting, and coverage from The Decoder and SiliconANGLE, the partnership is best understood as a large planned deployment with milestone-based investment, not a completed buildout.