AMD announced on July 20 that Microsoft will deploy its Helios rack scale AI system at scale on Azure, the anchor of an expanded long term strategic partnership between the two companies. Azure will use Helios to power frontier model inference for Microsoft itself, its AI customers and Azure AI services.
Helios is AMD's first full rack AI system and its most direct shot at Nvidia yet. Each rack integrates Instinct MI455X GPUs, EPYC Venice CPUs, Pensando networking and the ROCm software stack in an open rack design, built for both large scale training and inference. AMD positions it as the first real rival to Nvidia's rack scale systems, which have so far had that market to themselves. Shipments to customers, including Microsoft, begin in the second half of 2026.
The partnership goes beyond the racks. Microsoft is introducing two new Azure virtual machine series on sixth generation EPYC Venice processors, HDv2 aimed at agentic AI workloads and data pipelines, and HXv2 designed for semiconductor work. AMD's Pensando data processing units will also be deployed more broadly across Azure networking infrastructure.
Microsoft is not the first name on the Helios list, and that is the point. Meta, OpenAI, Oracle and Tata Consultancy Services have already made deployments or commitments to the system. Hyperscalers have spent two years looking for a credible second source at the rack level, not just the chip level, because rack scale is where Nvidia's integration advantage has been hardest to challenge and where supply pressure hurts most.
Whether this cracks Nvidia's dominance depends on delivery. Helios has to ship on time, at volume, and run frontier inference at competitive cost, and AMD's software stack has to hold up under hyperscaler load. But a top three cloud committing to deploy at scale is the strongest signal yet that the AI compute market is becoming a two supplier conversation, and procurement decisions for 2027 will be made with that assumption on the table.