<>
Microsoft is expanding its Azure cloud computing infrastructure by integrating advanced artificial intelligence and high-performance computing solutions from AMD.
The initiative addresses surging compute demand and the scaling challenges of modern AI workloads, which require greater hardware specialization across the technology stack.
The collaboration brings AMD’s latest Helios AI platform and next-generation EPYC datacenter processors into the Azure ecosystem. These components will directly power three new virtual machine offerings: HDv2 instances for data processing, HXv2 instances for electronic design automation, and ND MI455X v7 instances optimized for AI inference tasks. Industry collaboration across hardware and software providers remains central to meeting enterprise demands for energy efficiency and lower operational costs. By supporting a heterogeneous platform, cloud providers aim to give organizations greater flexibility as they deploy complex machine learning models and data pipelines at scale.
Powering AI Data Systems With Azure HDv2
CPU infrastructure forms the foundation for data-intensive AI operations, handling tasks like data preparation, search operations, and reinforcement learning. According to technical specifications, Azure HDv2 virtual machines are co-designed with AMD to eliminate processing bottlenecks and support agentic workload adoption.
Each HDv2 instance features nearly 500 physical 6th Generation AMD EPYC CPU cores, 4 terabytes of RAM, 32 terabytes of local NVMe storage, and 400 Gb Azure Boost networking. This high-density architecture ensures that AI accelerators receive a continuous supply of processed data, preventing idle compute cycles during large-scale training and execution pipelines.
Optimizing Silicon Design via Azure HXv2
The demand for specialized semiconductor design has grown rapidly as firms build custom silicon to power modern AI systems.
Building on earlier HX offerings launched in 2023, the new Azure HXv2 virtual machines leverage AMD’s 3D V-cache technology to improve single-threaded performance and memory handling for electronic design automation workflows. HXv2 instances feature 176 AMD 6th Generation EPYC CPU cores operating at clock frequencies exceeding 5 GHz, alongside 50 percent more addressable cache per core and VM configurations offering either 2 or 4 terabytes of RAM. Additionally, the inclusion of 800 Gb InfiniBand networking enables large-scale MPI-based simulations for scientific engineering and technical computing.
“Engineering teams are pushing the limits of simulation, chip design and scientific computing,” said Mark Papermaster, executive vice president and chief technology officer at AMD. “Azure HX is an important platform for scaling complex EDA workloads, and we’re excited about Azure HXv2, which is designed to deliver even greater performance and scalability.”
Accelerating Electronic Design Automation
Semiconductor design tools have also been optimized to take advantage of these cloud environments.
Shankar Krishnamoorthy, chief product development officer at Synopsys, noted that the collaboration allows engineering teams to extend electronic design automation workloads beyond traditional infrastructure constraints.
“Our collaboration with Microsoft on the Azure HX-series demonstrates a shared vision for enabling customers to deliver next-generation AI systems with precision and scale in accelerated design cycles,” Krishnamoorthy stated.
Scaling Production-Scale AI Inference With ND MI455X v7
To handle the computational demands of reasoning, search, and agentic services, Microsoft is introducing the ND MI455X v7 virtual machine series.
Powered by the AMD Helios rackscale solution, these instances are engineered to provide efficient, production-scale inference performance for enterprise customers deploying generative AI applications. Scott Guthrie, responsible for Microsoft’s hyperscale cloud computing solutions, data platforms, and cybersecurity services, oversees the development of these infrastructure offerings. The expansion provides enterprises with a wider range of configuration options tailored to specific AI pipelines, ranging from initial data processing to semiconductor design and final model inference.
Related reading