Microsoft is expanding its use of AMD hardware across Azure as it prepares to introduce new cloud systems for artificial intelligence and high performance computing workloads.
The agreement covers more than graphics processors. Microsoft plans to deploy AMD Instinct accelerators, EPYC server processors, Pensando networking technology and the ROCm software platform. This broad commitment gives AMD a larger role in Azure infrastructure and shows that Microsoft is looking beyond individual AI chips when planning future data centres.
The main addition is the AMD Helios rack scale platform. Microsoft intends to use Helios for large AI inference workloads, including frontier models, Azure services and applications created by cloud customers.
Commercial shipments are expected to begin during the second half of 2026. Microsoft will offer the platform through a new Azure ND MI455X v7 virtual machine series.
| Azure platform | Main hardware or capability | Intended workloads |
|---|---|---|
| ND MI455X v7 | AMD Helios with Instinct MI455X accelerators | AI reasoning, search, agents and large scale inference |
| HDv2 | High core count AMD EPYC processors | Data preparation, reinforcement learning and AI pipelines |
| HXv2 | High frequency AMD EPYC processors | Chip design, engineering and scientific simulations |
| Azure networking | AMD Pensando data processing units | Network processing, efficiency and infrastructure acceleration |
Helios combines AMD’s major data centre technologies
Helios integrates Instinct MI455X accelerators with sixth generation EPYC processors, known by the development name Venice. It also includes Pensando networking hardware and AMD’s ROCm software environment.
Microsoft plans to use the system for production scale inference, where trained AI models process requests and generate responses. The ND MI455X v7 configuration is designed for demanding workloads such as reasoning, information retrieval and applications that use autonomous AI agents.
AMD describes Helios as an open rack scale architecture. This approach could appeal to cloud providers that want more control over system design and software while reducing their dependence on a single proprietary hardware platform.
The technical design is only part of AMD’s challenge. The company must manufacture Helios systems in large quantities and continue improving ROCm so developers can deploy applications reliably across large clusters. Software support remains an important factor when cloud providers compare AI infrastructure options.
New EPYC systems will support AI preparation and scientific computing
Microsoft is also introducing Azure HDv2 systems for the work surrounding AI model development. These instances are intended for data preparation, search, reinforcement learning, agent coordination and complex data pipelines.
The largest HDv2 configurations will provide nearly 500 physical processor cores, as much as 4TB of memory and up to 32TB of local NVMe storage. These resources are designed for jobs that need substantial processing capacity before information reaches an accelerator.
Azure HXv2 will serve a different group of customers. It is aimed at semiconductor development, engineering analysis, scientific modelling and other high performance computing tasks. Configurations will include 176 EPYC processor cores operating at frequencies above 5GHz, up to 4TB of memory and 800Gbps InfiniBand networking.
Microsoft will also expand its use of AMD Pensando data processing units. The chips will work with Azure Boost to handle networking and connection related tasks more efficiently, allowing the main processors to spend more time on customer workloads.
Financial terms and deployment volumes were not disclosed, so the immediate revenue benefit for AMD remains unclear. Even so, the scale of the agreement is significant because it covers accelerators, processors, networking and software.
The partnership gives Microsoft additional hardware choices as demand for AI computing continues to grow. For AMD, the larger opportunity depends on delivering Helios on schedule, expanding production and providing a software experience that meets the requirements of one of the world’s largest cloud platforms.
Discussion (0)
Be the first to comment.