Microsoft Expands AMD Partnership With New Processors and AI Accelerators for Azure

Microsoft and AMD announced on July 20, 2026 an expansion of their multi-year collaboration on Azure cloud infrastructure. The companies confirmed that Azure will adopt AMD's full AI technology stack, from graphics accelerators to processors to data center networking, marking a clear shift for Microsoft away from its near-exclusive reliance on Nvidia chips for AI computing power.

What the deal covers

The expanded partnership covers four elements of AMD's infrastructure. The first is the Helios rackscale platform, an integrated system combining Instinct MI455X accelerators into a deployment-ready compute unit built to handle demanding AI model inference workloads. The second is sixth-generation EPYC Venice processors, which Microsoft will use as the basis for two new virtual machine series on Azure.

The third element is Pensando network controllers, whose deployment across Azure's network infrastructure is set to expand, with integration into the Azure Boost system meant to boost network performance and connection-processing efficiency. The fourth is the ROCm development environment, AMD's open software stack that competes with Nvidia's CUDA, intended to make it easier for Azure customers to move AI workloads between different types of accelerators.

Why it matters for the cloud market

For several years, nearly all growth in compute power for training and running large language models has relied on Nvidia chips, giving the company a dominant pricing position over cloud providers. The expanded partnership with AMD gives Microsoft a real purchasing alternative and negotiating leverage, while giving Azure customers a wider choice of hardware for specific tasks, from model training to data preparation and inference.

For AMD, it confirms the company is building a genuine second business line alongside the processor segment, where it has long competed with Intel. The Instinct accelerators and Helios platform are meant to answer hyperscalers' growing demand for an alternative to Nvidia chips, particularly in inference, where energy cost and memory bandwidth matter more than raw training power.

Market context

The announcement comes as the data center industry grapples with DRAM memory shortages and rising costs of building AI infrastructure, with major cloud providers looking for ways to diversify their hardware supply chains. AMD has also made acquisitions in recent months to strengthen its data center offering, including memory-optimization startup MEXT, pointing to a broader company strategy that extends beyond processors and graphics cards alone.

For Polish companies using Microsoft Azure services, the expanded hardware lineup means potentially wider choice of compute instances and possible downward pressure on prices for renting AI power in the cloud, though the new virtual machines based on EPYC Venice and the Helios platform will roll out to Azure's global offering gradually, starting in the second half of 2026.

Microsoft did not disclose financial terms or the scale of its hardware order from AMD. The companies also did not specify which Azure regions will be first to receive the new virtual machines based on EPYC Venice processors or the Helios platform. Both companies said only that the rollout would proceed in stages as hardware becomes available.