Microsoft and AMD announced an expansion of their long-term strategic partnership spanning AMD GPUs, CPUs, networking, and software on Azure. Microsoft will deploy AMD’s Helios rack-scale platform to power frontier model inference for its own use, AI customers, and Azure AI services. They are adding two new AMD EPYC CPU-powered VM series and widening the use of AMD Pensando DPUs in Azure networking infrastructure.
Microsoft said the collaboration is meant to give customers more performance, scale and choice for building and running AI workloads, while AMD framed the partnership as an extension of work the two companies have built over the years.
According to AMD, Microsoft will ramp Helios at scale on Azure for frontier model inference, while AMD begins shipping Helios to customers, including Microsoft, in the second half of 2026. Helios combines AMD Instinct MI455X GPUs, 6th Gen AMD EPYC “Venice” CPUs, Pensando networking and ROCm software in an integrated rackscale platform.
On the compute side, Azure will add Azure HDv2 virtual machines for agentic AI and data pipelines, and Azure HXv2 virtual machines for semiconductor design. Both new VM series will use 6th Gen AMD EPYC “Venice” processors. Microsoft is integrating Azure Boost with AMD technologies to improve networking performance and connection processing at cloud scale.
What this means for Azure
Microsoft is broadening the hardware base behind Azure AI. Azure is blending AMD GPUs, AMD CPUs, AMD networking and Microsoft’s own cloud stack to serve different classes of AI work. Customers are looking for infrastructure optimized for training, inference, data preparation, search and reinforcement learning.
Azure is being positioned less as a general cloud with AI bolted on, and more as a portfolio of specialized compute layers for different AI tasks.
Microsoft would offer cloud customers AMD’s AI chip platform on Azure as an alternative to Nvidia-based systems. Microsoft was planning to offer AMD AI chips to Azure customers and preview new Cobalt 100 processors.
Cloud vendors are no longer just selling access to large AI accelerators; they are assembling purpose-built systems for inference, data pipelines and engineering workloads. Microsoft’s use of AMD’s new HDv2 and HXv2 virtual machines shows that the company is now matching hardware more closely to workload type.
[ALSO READ: AMD Launches Ryzen 9 9950X3D2 Dual Edition with First Dual 3D V-Cache Design ]
