AMD is preparing to launch a comprehensive new AI infrastructure lineup, including the Helios server rack, Instinct MI450 GPUs, and EPYC Venice processors, directly challenging NVIDIA’s dominance in data center AI. The company also announced a major partnership with Anthropic, the developer of the Claude AI model, to deploy up to two gigawatts of AMD chips starting in 2027.
Helios is AMD’s first full server rack designed for AI workloads, combining MI455X graphics processors, EPYC Venice CPUs, and Pensando networking technology. It targets both AI training and inference, competing with NVIDIA’s integrated systems like the upcoming Rubin-Vera platform. AMD has already demonstrated Helios at the Moscone West convention center, with cloud providers Vultr and TensorWave showing systems using AMD hardware.
Under the Anthropic agreement, the first gigawatt of AMD chips is scheduled to begin operating in the first half of 2027, using Helios systems with MI455X processors. The companies will also collaborate on software development, with Claude optimizing workloads on AMD graphics processors and aiding development of the ROCm software platform. AMD has committed to an equity investment of up to $5 billion in Anthropic.
AMD is expanding its position in the AI chip market as demand for AI computing grows. While NVIDIA still controls a large share of the data center accelerator market, AI developers are seeking more supplier options. AMD previously signed a multiyear agreement with OpenAI that could generate tens of billions of dollars in yearly revenue, and OpenAI received an option to acquire a roughly 10% stake in AMD. The Anthropic partnership follows a similar strategy, combining hardware supply with closer access to a major AI model developer.
Thursday’s launch will provide more details on Helios and the Venice processor, showcasing AMD’s full-system approach to compete with NVIDIA across processors, accelerators, networking, and software.
Leave a Reply