AMD unveils its first rack-mounted AI system Helios, with Microsoft joining its procurement ranks to challenge NVIDIA’s AI computing power hegemony.

AMD has officially launched its first artificial intelligence-focused rack-scale system, Helios, and announced that Microsoft will deploy the system in its Azure data centers, making Microsoft another key customer following the likes of Meta, OpenAI, and Oracle. Helios is regarded as AMD’s key product in recent years to challenge NVIDIA’s dominance in AI infrastructure, and it is also the first AI rack solution directly comparable to NVIDIA’s Grace Blackwell and Vera Rubin systems. It is expected to begin shipping later this year. Microsoft said it will leverage AMD Helios to support frontier model inference, AI application development, and Azure AI services, while launching two types of compute instances based on AMD’s latest "Venice" CPU for use cases such as agent AI and semiconductor design. AMD noted that Helios integrates its self-developed GPUs, CPUs, network chips, and software ecosystem, enabling lower per-token cost. The system uses the MI400 series GPUs, with each compute tray equipped with 4 Instinct GPUs and 1 EPYC CPU. Currently, Meta plans to deploy a Helios system with a maximum scale of 1GW, while OpenAI, Oracle, and other firms have also announced adoption of related solutions. AMD added that 8 of the world’s top 10 AI companies are already running its Instinct GPU workloads.