Anthropic PBC has agreed to acquire up to 2 gigawatts of AMD's MI450 GPUs, enhancing its AI infrastructure with AMD’s cutting-edge Helios rack systems while AMD invests $5 billion in the AI startup to optimize software development using the Claude AI model.
- Anthropic to deploy AMD’s Helios GPU racks delivering up to 1.4 exaflops performance
- AMD invests $5 billion in Anthropic and adopts Claude AI model to speed development
- Deal complements Anthropic’s diversified GPU supply including Nvidia and Google
What happened
Anthropic PBC has entered a multibillion-dollar agreement with Advanced Micro Devices Inc. to purchase up to 2 gigawatts of GPU capacity. The hardware consists of AMD’s MI450 series graphics cards housed in Helios rack appliances, which combine powerful GPUs with CPUs and data processing units to manage infrastructure tasks efficiently. This new capacity is expected to be deployed starting in the first half of 2027.
Alongside the GPU purchase, AMD is making a strategic investment of up to $5 billion in Anthropic. The partnership includes adopting Anthropic’s Claude AI model to accelerate AMD’s software development, particularly focusing on enhancing ROCm, AMD's open software platform designed to optimize AI workloads and streamline development for customers using the Helios system.
Why it matters
The deal signals continued growth and competition in AI infrastructure, with Anthropic reinforcing its hardware base by partnering with AMD. Helios systems offer significant advantages over competing products like Nvidia’s Vera Rubin by providing 50% more high-bandwidth memory and superior scale-out bandwidth, critical for large AI clusters processing massive datasets.
AMD's investment and integration of the Claude AI model into its development pipeline underline the increasing importance of AI-driven software efficiency in hardware deployment. This collaboration could speed up innovation in both GPU architecture and supporting ecosystems, potentially influencing market dynamics as other major players including Meta and Microsoft also adopt AMD solutions.
What to watch next
Stakeholders should monitor the deployment progress of Helios GPU racks in Anthropic’s data centers starting in 2027. The performance and scalability of these systems will be crucial in assessing AMD’s competitiveness against industry leaders like Nvidia and the overall impact on AI infrastructure capabilities.
Further developments in the multi-year software collaboration focusing on ROCm and AI-powered development workflows, enabled by Claude, will be important. These enhancements could reduce time-to-market for new AI technologies and improve the efficiency of large-scale AI model training, influencing future agreements between AI developers and chip manufacturers.