The extraordinary pace of AI adoption is driving rapid growth in demand for training and inference capabilities. Anthropic is expanding its computing infrastructure to meet the strong and growing demand for Claude, and its adoption of gigawatt-scale AMD Helios represents a significant expansion of AMD's position at the heart of global AI infrastructure development.

Anthropic will deploy AMD Helios rack-scale solutions, which incorporate AMD Instinct™ MI455X GPUs—part of the AMD Instinct MI450 series—along with AMD EPYC™ “Venice” CPUs, AMD Pensando™ networking technology, and ROCm™ software. Designed specifically for next-generation AI training and inference, Helios delivers the performance, efficiency, and scalability required for the most demanding AI workloads. This deployment complements Anthropic’s existing use of AMD Instinct MI355X GPUs.

In addition, AMD and Anthropic are launching a multi-year engineering collaboration to use Claude to accelerate AMD software development. Specifically, the teams will use Claude to optimize workloads for AMD Instinct GPUs and accelerate the development of ROCm software. AMD will also widely adopt Claude across its engineering and product development teams.

AMD has committed to a strategic capital investment of up to $5 billion in Anthropic in the future.
“We are thrilled to deepen our collaboration with Anthropic and deploy AMD Helios at gigawatt scale,” said Dr. Lisa Su, AMD Chairman and CEO. “This collaboration combines Anthropic’s leadership in cutting-edge AI with the full power of AMD’s high-performance computing. Together, we will accelerate the adoption of AI at scale and solidify Helios as a key platform for the next generation of AI infrastructure.”

“Access to computing power is critical to keeping Claude at the forefront and meeting our customers’ demands,” said Tom Brown, co-founder and chief computing officer of Anthropic. “By partnering with AMD across the entire technology stack, we secure the power we need and optimize it for training and running Claude. Running the system on a diverse range of hardware allows us to assign the right workloads to the right hardware.”