AMD's $5 Billion AI Power Play: Investing Big in Anthropic

AMD has committed to a strategic investment of up to $5 billion in AI developer Anthropic, forming a substantial infrastructure agreement valued at tens of billions of dollars for AI systems. Under this agreement, Anthropic plans to deploy up to two gigawatts of capacity utilizing AMD's Instinct MI450-series accelerators, with the initial gigawatt deployment slated to commence in the first half of 2027. AMD's investment is contingent upon Anthropic reaching specified deployment milestones, though the precise conditions have not been publicly disclosed by either company. The transaction integrates an equity commitment from AMD with a significant hardware order from Anthropic, representing distinct yet interconnected facets of the overall agreement.

This deal underscores AMD's strategy of intertwining financial incentives with large infrastructure orders. Similar arrangements include AMD's October agreement to supply OpenAI with systems supporting up to six gigawatts of capacity, accompanied by warrants allowing OpenAI to acquire approximately 10% of AMD. These OpenAI warrants vest in stages based on deployment, commercial success, and share-price conditions. AMD also secured a deal with Meta in February for up to six gigawatts of GPU capacity, which included performance-based warrants tied to shipment and purchase targets. While the structures vary – a direct investment in Anthropic versus rights to acquire AMD shares for OpenAI and Meta – they all reflect chip suppliers pairing investments or equity incentives with commercial agreements involving major AI developers. Notably, Reuters also reported on Nvidia's discussions to potentially invest up to $30 billion in OpenAI.

The Anthropic deployment will leverage AMD Helios systems, which incorporate MI455X GPUs from the Instinct MI450 series, EPYC “Venice” processors, Pensando networking solutions, and the ROCm software platform. This builds upon Anthropic’s existing use of AMD’s earlier MI355X accelerators. The configuration provided is an integrated rack-scale platform, encompassing accelerators, CPUs, networking components, and software, rather than merely a standalone GPU purchase. AMD chair and chief executive Lisa Su expressed enthusiasm for deepening the partnership, highlighting the collaboration as a union of Anthropic's leadership in frontier AI with AMD's robust high-performance computing capabilities. Su further indicated the companies' intention to establish Helios as a foundational platform for future AI infrastructure deployments, reflecting AMD's broader ambition to expand the adoption of its rack-scale systems among companies developing and operating large-scale AI models.

With this commitment, Anthropic joins OpenAI and Meta on AMD’s roster of customers planning to utilize its next-generation accelerator platform at gigawatt scale, with OpenAI and Meta each committing to up to six gigawatts, and Anthropic to two gigawatts. The two-gigawatt commitment for Anthropic translates to 2,000 megawatts of power capacity. AMD executives have previously stated that establishing one gigawatt of AI computing infrastructure can entail costs running into tens of billions of dollars, depending on the specific equipment and facilities involved. This AMD deployment significantly expands Anthropic’s compute capabilities; for context, Anthropic has independently secured over 300 megawatts through SpaceX’s Colossus 1 facility in Memphis, which houses more than 220,000 Nvidia GPUs. The full planned AMD deployment would represent more than six times the power capacity obtained through Colossus 1, though direct computational performance comparisons are complex due to differing hardware and deployment models.

Anthropic's strategy for deploying the new AMD systems involves a combination of its own data centers, alongside facilities hosted by cloud providers and specialist AI infrastructure companies, according to The Wall Street Journal. Both AMD and Anthropic are actively seeking suitable operators to host these planned systems. Lisa Su emphasized that achieving gigawatt-scale capacity necessitates a significant planning horizon of 12 to 24 months before actual deployment. The Wall Street Journal also reported that AMD had engaged in discussions regarding providing financial guarantees for Anthropic’s future data-center leases, although neither company has confirmed these proposed terms or the extent of any potential obligations.

The AMD systems will integrate into Anthropic’s diverse compute network, which already incorporates infrastructure based on Nvidia GPUs, Amazon Trainium processors, and Google tensor processing units. Amazon remains Anthropic’s principal cloud and training partner, from whom Anthropic has secured up to five gigawatts of capacity, deploying over one million Trainium2 processors and planning Trainium3 deployments for 2026. Anthropic has also secured multiple gigawatts of next-generation TPU capacity from Google and Broadcom, with deployments anticipated to begin in 2027. Anthropic's AI model, Claude, is accessible via Amazon Web Services, Google Cloud, and Microsoft’s Azure-based Foundry platform. Tom Brown, Anthropic co-founder and chief compute officer, highlighted the advantage of running across a diversified range of hardware: it enables mapping the appropriate workloads to the optimal hardware. He underscored that the AMD partnership would provide crucial additional capacity while facilitating the optimization of systems for training and serving Claude, ensuring access to compute—a central element for keeping Claude at the frontier of AI and meeting customer demand. Anthropic has also committed to using computing capacity at SpaceX’s Colossus facilities and has explored leasing infrastructure from Meta, with Reuters reporting a potential Meta agreement valuing up to $10 billion over two years.

Further enriching the partnership, the AMD agreement includes a multi-year engineering program. Under this initiative, the companies will utilize Claude to optimize workloads specifically for Instinct accelerators and to support the ongoing development of AMD’s ROCm software platform. AMD intends to deploy Claude across its engineering and product-development teams. This collaborative effort positions Anthropic not only as a customer of AMD’s infrastructure but also as an active participant in the development of the software environment that runs workloads on that infrastructure, playing a pivotal role in adapting Claude workloads to AMD's hardware and software ecosystem. Following the announcement of this agreement, AMD shares saw a 2.4% increase, demonstrating market confidence and outperforming the broader Philadelphia Semiconductor Index.