Microsoft to deploy AMD’s Helios AI racks on Azure

Microsoft ↗ 0.52% announced Monday it will deploy AMD's Advanced Micro Devices, Inc. ↘ 1.47% Helios rack-scale AI system across its data centers and Azure cloud infrastructure, marking the most direct challenge yet to Nvidia's ↘ 1.47% dominance in large-scale AI compute. The partnership also includes two new Azure virtual machine series built on AMD's upcoming sixth-generation Epyc "Venice" processors.amd+1

A Rack-Scale Rivalry

The AMD Helios system packs 72 Instinct MI455X GPUs into a double-width rack, delivering up to 1.4 exaFLOPS of FP8 compute for AI training and 2.9 exaFLOPS of FP4 performance for inference, alongside 31 TB of HBM4 memory and 1.4 PB/s of aggregate memory bandwidth. Microsoft will use the platform to power frontier model AI inference for its own services, its AI customers, and Azure AI workloads.amd+3

The announcement comes as AMD prepares to formally launch its Epyc Venice server CPUs at the company's "Advancing AI 2026" event in San Francisco on July 22–23. AMD said it will begin shipping Helios systems in the second half of 2026.finance.yahoo+3

Competing With Nvidia at Scale

Helios positions AMD as a direct rack-level competitor to Nvidia's Vera Rubin NVL72, which delivers 3.6 exaFLOPS of NVFP4 inference with 20.7 TB of HBM4 memory across 72 Rubin GPUs and 36 Vera CPUs. Both systems target the same second-half 2026 availability window and trillion-parameter workloads.insiderfinance+1

Microsoft joins Meta, OpenAI, and Oracle as early Helios customers, according to CNBC. Meta signed a multi-year agreement in February to deploy up to six gigawatts of AMD Instinct GPUs built on the Helios architecture.amd+1

Open Standards Strategy

AMD has built Helios on open industry standards including the Open Compute Project's Open Rack Wide form factor, Ultra Accelerator Link for GPU interconnect, and Ultra Ethernet Consortium specifications for scale-out networking. Each MI455X GPU features 432 GB of HBM4 memory and 19.6 TB/s of bandwidth, built on AMD's next-generation CDNA 5 architecture using TSMC's 2nm process.wccftech+1

The new Azure VM series built on Venice CPUs will broaden Azure's AMD Epyc portfolio across AI, data, and engineering workloads, according to AMD's press release. AMD shares rose following the announcement.seekingalpha+1