AMD to invest up to $5 billion in Anthropic Helios deal

AMD will invest up to $5 billion in Anthropic under an infrastructure agreement linked to the deployment of up to two gigawatts of AMD Helios AI systems. The first gigawatt is scheduled to begin rolling out in the first half of 2027. AMD’s equity commitment is tied to deployment milestones; the companies have not disclosed the specific conditions, payment schedule or cancellation terms.

Anthropic will install AMD Helios rack-scale systems built around Instinct MI455X GPUs from the MI450 series, EPYC “Venice” processors, Pensando networking components and AMD’s ROCm software. The configuration is described as an integrated platform spanning accelerators, CPUs, networking and software. Anthropic has previously used AMD’s MI355X accelerators and plans to operate AMD systems alongside hardware from other suppliers.

Some systems will be placed in Anthropic’s own data centres, while others will be hosted by cloud providers and specialist infrastructure operators. Both companies are seeking operators capable of hosting gigawatt-scale deployments. AMD executives say planning for that level of capacity typically requires 12 to 24 months before installation begins. Building one gigawatt of AI computing infrastructure can cost tens of billions of dollars, depending on equipment and facilities.

The agreement adds AMD to Anthropic’s multi-supplier compute strategy. Anthropic runs Nvidia GPUs, Amazon Trainium processors and Google TPUs, and has secured more than 300 megawatts through leased facilities. The company has arrangements for multiple gigawatts of Trainium and TPU capacity, with additional deployments planned through 2026 and 2027. Anthropic also maintains a large cloud partnership with Amazon.

The pact includes a multi-year engineering programme in which Anthropic will use its Claude model to optimise workloads for AMD’s Instinct accelerators and to support development of the ROCm software stack. AMD will deploy Claude across its engineering and product teams as part of the work, giving Anthropic a role in adapting models and software to AMD hardware.

Lisa Su, AMD’s chair and chief executive, described the plan as an effort to establish Helios as a platform for future AI infrastructure deployments. Tom Brown, Anthropic’s co-founder and chief compute officer, said running across a diversified set of hardware allows the company to map workloads to the appropriate systems and secure the compute capacity needed for Claude.

The structure of the agreement pairs an equity commitment with a large hardware order, similar to other recent industry deals that link financial incentives to chip deployments. The companies have not disclosed Anthropic’s purchase obligations, the valuation implied by AMD’s investment or how delayed deployments will be treated. AMD’s shares rose modestly after the announcement.