Microsoft Deepens AI Partnership with AMD, Azure Introduces Helios Rack-Level AI System

AMD and Microsoft are expanding their long-term cooperation, adding another key advantage for AMD as it challenges Nvidia's dominance in the AI chip market.

On July 20, it was reported that Microsoft will deploy AMD’s Helios rack-level solution on its Azure cloud platform to power cutting-edge model AI inference workloads. AMD plans to begin shipping to customers including Microsoft in the second half of 2026.

The cooperation also covers two new Azure virtual machine series based on the sixth-generation AMD EPYC Venice processors, as well as the deployment of AMD Pensando Data Processing Units (DPU) in AMD's AI backend network infrastructure and some Azure services.

Microsoft CEO Satya Nadella said, “We are expanding Azure’s infrastructure portfolio with AMD Helios, providing customers with the performance, scale, and choice needed to build and run next-generation AI applications.” This collaboration further enriches Microsoft’s AI computing power sources and holds major strategic significance for Microsoft as it accelerates in-house model development and continues to scale up investment in compute.

Helios: AMD’s First Rack-Level AI System

Helios is AMD's first rack-level AI system and the most direct competitor so far to Nvidia’s Grace Blackwell and Vera Rubin systems. The system integrates AMD’s own GPUs, CPUs, network chips, and software. Each compute tray is equipped with four Instinct GPUs, driven by a single EPYC central processor, and each tray also carries up to 12 network chips based on technology acquired when AMD bought Pensando in 2022.

AMD Data Center business head Forrest Norrod stated that Helios’ core competitiveness lies in "offering the best total cost of ownership and lowest per-token cost", and said customer feedback has validated this goal. AMD CEO Lisa Su previously noted that Helios has a "significant advantage" over Nvidia’s rack-level systems in inference performance, memory bandwidth, and memory capabilities.

In terms of cost, market research firm Futurum Group estimates that Helios will be priced between $5 million and $5.5 million, higher than their $3.5 million to $4 million estimate for Nvidia’s second-generation rack-level system Vera Rubin.

Expanding Customer Base—AMD Accelerates Market Share Acquisition

Microsoft’s entry into the Helios customer lineup further expands AMD’s customer portfolio.

Previously, Meta announced it would over time use up to 6 GW of AMD GPUs and would be the first to deploy 1 GW on the Helios racks; OpenAI, Oracle, and India's largest IT company Tata Consultancy Services have also made significant deployment commitments. AMD states that eight of the world’s top ten AI companies are running workloads on its Instinct GPUs.

Currently, Nvidia holds over 95% share of the data center GPU market, while AMD accounts for about 4.5%. Futurum Group analyst Daniel Newman believes AMD is likely to increase its market share to between 20% and 25%, “which means hundreds of billions of dollars in revenue”.

AMD stated that in the first quarter of 2026, data center business revenue grew 57% year-on-year, and it plans to achieve tens of billions of dollars in data center AI revenue starting from 2027, with most of it coming from Helios.

Deepening Long-Term Cooperation—Software Ecosystem Remains the Key Variable

This cooperation is a continuation of AMD and Microsoft’s long-term partnership. AMD chips have long powered Microsoft Surface PCs and Xbox game consoles; in 2023, Microsoft was the first to use AMD MI300X GPUs for AI workloads and has also deployed its self-developed Maia chips in data centers.

Reports citing analysts note that AMD has essentially caught up with Nvidia on the hardware level, but the software ecosystem remains the crucial gap. Counterpoint Research analyst Neil Shah said, “AMD’s Helios chips are performance-competitive with Nvidia’s GPUs and CPUs, but the key is in the software and optimization,” highlighting that Nvidia’s CUDA provides a much more mature and extensive ecosystem.

Newman raised another core issue: “Is AMD winning due to true technological edge, or simply because AI compute supply is so tight that as long as they can ship, there will be buyers?” The actual performance of Helios in its early deployments will largely determine the answer to this question.