AMD announced Monday the launch of Helios, its first rack-scale AI system designed for deployment in Microsoft Azure data centers [1, 2].
The move represents a direct challenge to Nvidia's dominance in the AI infrastructure market. By shifting from selling individual chips to integrated rack-scale systems, AMD aims to provide the high-performance compute necessary for the next generation of large-scale AI models [1, 3].
Each Helios rack contains 72 MI455X GPUs [3]. The system is equipped with 31 terabytes of HBM4 memory per rack and delivers an inference compute capability of 2.9 exaflops [3].
Microsoft will be among the first to deploy the hardware in its Azure cloud environment [1, 2]. Other early customers include Meta, OpenAI, and Oracle [1, 2].
Timeline details for the rollout vary by source. Some reports indicate that Helios will ship to Microsoft and other customers later this year [1]. Other data suggests that engineering samples will ship in the second half of 2026, with mass production not slated to begin until the second quarter of 2027 [3].
The development of the system took place at AMD's testing and development lab in Texas [1, 2].
“AMD announced Monday the launch of Helios, its first rack-scale AI system.”
The transition to rack-scale offerings allows AMD to compete with Nvidia not just on chip performance, but on the total system architecture. By securing commitments from the largest AI spenders—Microsoft, Meta, and OpenAI—AMD is attempting to break the vertical integration advantage Nvidia holds in the data center market.
