Data center operators finally have a serious second option for AI hardware. AMD’s Helios rack-scale system, unveiled July 23 at the company’s Advancing AI event, brings together Instinct GPUs and EPYC processors in a single cabinet aimed directly at Nvidia’s data center stronghold.
Chief executive Lisa Su positioned Helios as the highest-performance AI rack on the market and backed it with hard numbers. The company projects the AI accelerator market will reach $1.4 trillion by 2030, a bet that demand for agentic AI workloads will multiply compute requirements many times over.
The customer list reads like a who’s who of AI infrastructure buyers. OpenAI, Anthropic, Meta, Oracle, and Microsoft have all committed to Helios deployments. Microsoft plans to add the racks to its Azure fleet, while AMD and Anthropic signed a separate agreement to build up to two gigawatts of GPU infrastructure together.
AMD claims Helios beats Nvidia’s Vera Rubin on performance metrics and cost simultaneously. Analysts say the real test is whether AMD can deliver at scale. The company currently accounts for roughly 4.5% of the data center GPU market, leaving enormous room to grow if Helios performs as advertised.
The launch signals that hyperscalers may finally get the supply diversity they have long sought. With Nvidia’s systems frequently oversubscribed, a credible competitor at the rack level could reshape how the largest AI labs plan their infrastructure builds.
