[Digital Today reporter Suk Dae-geon] Competition to build computing infrastructure is spreading to gigawatt-scale capacity as AI demand grows. AMD and Anthropic said on July 22 (local time) they signed a strategic partnership to deploy an 'AMD Helios' rack-scale solution using AMD Instinct MI450 series GPUs with capacity of up to 2 gigawatts.

Anthropic is expanding its computing infrastructure to meet growing demand for Claude. The first 1-gigawatt build is scheduled to start in the first half of 2027, expanding its existing environment based on AMD Instinct MI355X GPUs.

The 'AMD Helios' that Anthropic will deploy consists of AMD Instinct MI455X GPUs, AMD EPYC 'Venice' CPUs, AMD Pensando networking and ROCm software. The configuration is optimised for AI training and inference, and has performance, power efficiency and scalability needed for large workloads. AMD sees the agreement as an opportunity to strengthen its position in the global AI infrastructure market.

The two companies will also pursue multi-year engineering cooperation in addition to hardware supply. They plan to optimise AMD Instinct GPU workloads using Claude and accelerate development of ROCm software. AMD decided to adopt Claude across its engineering and product development organisations. AMD also plans to pursue a strategic equity investment in Anthropic of up to $5 billion.

AMD CEO Lisa Su (리사 수) said the cooperation combines Anthropic's technological capabilities in leading cutting-edge AI and AMD's high-performance computing capabilities. She said she expects large-scale AI adoption to accelerate and Helios to establish itself as a core platform for next-generation AI infrastructure.

Anthropic co-founder Tom Brown (톰 브라운), chief commercial officer, said working with AMD across the full stack will allow the company to secure the computing capacity it needs while optimising Claude training and service delivery. He said using diverse hardware will enable flexible use of the hardware best suited to each workload.