AMD to Supply 'Helios' AI Accelerators to Anthropic in $5 Billion Deal

Strategic equity investment and 2 GW AI infrastructure challenge Nvidia's market dominance

AMD has entered into a strategic partnership with Anthropic and is pursuing an equity investment of up to 5 billion dollars (approximately 6.9 trillion Korean won). As AMD secures Anthropic as a major client following OpenAI, Meta, Oracle, and Microsoft, cracks are beginning to appear in the AI accelerator market, which had solidified into a one-strongman system led by Nvidia.

◇ AMD to Supply 'Helios' AI Accelerators on a Large Scale to Anthropic

AMD announced the partnership on the first day of its 'AMD Advancing AI 2026' event, held at the Moscone West in San Francisco on the 22nd (local time). The core of the agreement is the supply of the 'Helios' AI accelerator rack system, which has a maximum capacity of 2 gigawatts (GW), to Anthropic.

Under this contract, AMD will begin constructing the first 1 GW facility in the first half of 2027. A 1 GW capacity is equivalent to the power generated by one nuclear power plant, indicating that Anthropic is moving to secure large-scale AI infrastructure. The remaining 1 GW is also scheduled to be added sequentially, bringing the total contract size to a power capacity equivalent to two nuclear power plants.

Anthropic plans to build a data center by integrating AMD's latest AI semiconductors, including the graphics processing unit (GPU) 'MI455X,' the server central processing unit (CPU) EPYC 'Venice,' Pensando networking equipment for data transmission, and driving software. This holds significant meaning as Anthropic has expanded its collaboration with AMD on a large scale, demonstrating trust in AMD's AI accelerators.

Alongside this, AMD plans to pursue a strategic equity investment of up to 5 billion dollars in Anthropic. Given that it is rare for semiconductor companies to invest in equity while supplying large-scale GPUs to clients, this move is seen as creating a close-knit structure intertwined through capital, beyond a simple 'supplier-client' relationship.

◇ Collaboration Extends to Semiconductor Development with Claude... Strengthening Software Ties

Separately from the hardware supply agreement, the two companies announced a collaboration plan to directly utilize Anthropic's AI model Claude in AMD's semiconductor development. AMD aims to enhance the completeness of its software platform 'ROCm,' which maximizes GPU performance in actual workloads, by using Claude as a development assistant tool for improving its semiconductors and software.

AMD plans to further expand the use of Claude across its engineering and product development organizations. In the semiconductor industry, this approach of strengthening ties with clients from the software development stage, rather than merely selling hardware, is analyzed as a response to Nvidia's developer lock-in strategy built through the CUDA ecosystem.

This announcement gains greater significance when combined with the previously revealed partnership with Microsoft for Azure. AMD has now secured all major AI companies -- OpenAI, Meta, Oracle, Microsoft, and Anthropic -- as clients. During the briefing, AMD stated that eight out of the world's top 10 AI companies are already operating workloads on its Instinct GPUs.

According to a source from AMD, "The AI accelerator market has effectively been dominated by a single player, Nvidia. However, as AMD continues to secure major AI companies as clients, the trend of establishing itself as a practical alternative supplier is becoming clear."