Microsoft and AMD Partner for Helios AI on Azure

Microsoft to deploy AMD’s Helios rack-scale AI accelerator on Azure in 2026, featuring 72 Instinct MI455X GPUs for frontier-model workloads.

Microsoft and AMD have announced a partnership to bring the Helios rack-scale AI accelerator to Azure infrastructure. Scheduled for release in late 2026, the Helios system integrates 72 Instinct MI455X GPUs, delivering 1.4 exaFLOPS of FP8 and 2.9 exaFLOPS of FP4 compute. The hardware features 31.1TB of HBM4 memory capacity, targeting 260 TB/s of scale-up bandwidth and 43 TB/s of scale-out bandwidth via UALink over Ethernet. Microsoft plans to utilize Helios for internal frontier-model workloads and Azure AI customers. Additionally, Azure will introduce HDv2 and HXv2 VM series powered by sixth-gen AMD Epyc Venice CPUs for agentic AI and semiconductor design. The collaboration also includes integrating AMD Pensando DPUs into Azure Boost, building on AMD's existing partnerships with industry leaders like OpenAI and Meta.