Microsoft expands Azure AI infrastructure with three new AMD-powered VM families

Microsoft announced three new Azure virtual machine families built on AMD's latest silicon and rackscale platform, expanding Azure's AI and high-performance computing infrastructure to cover data preparation, chip design, and large-scale inference workloads. The announcement was published July 20, 2026 on the Microsoft blog by Scott Guthrie, EVP of Cloud + AI.

What's new

Three distinct VM families target three different points in the AI pipeline:

• Azure HDv2 (AI data systems) — built for CPU-intensive AI workloads such as data preparation, search, reinforcement learning, and agent coordination. Specs include "nearly 500 physical 6th Gen AMD EPYC CPU cores, 4 terabytes of RAM, 32 terabytes of local NVMe storage," plus 400 Gb Azure Boost networking.
• Azure HXv2 (chip design and technical computing) — the successor to the HX platform Microsoft launched in 2023, aimed at RTL simulation and other EDA workloads. It packs "176 AMD 6th Gen EPYC CPU cores with a clock frequency of more than 5 GHz," 50% more addressable cache per core via AMD's 3D V-Cache, up to roughly 4 terabytes of RAM, and 800 Gb InfiniBand.
• ND MI455X v7 (AI inference) — powered by AMD's Helios rackscale solution, aimed at reasoning, search, and agentic inference workloads at production scale.

AMD CTO Mark Papermaster called Azure HX "an important platform for scaling complex EDA workloads," adding AMD is "excited about Azure HXv2." Synopsys's Shankar Krishnamoorthy said the collaboration with Microsoft "demonstrates a shared vision for enabling customers to deliver next-generation AI systems."

Context

The HXv2 launch extends a chip-design-focused Azure/AMD partnership that dates to 2023's original HX platform. The broader announcement reflects Microsoft's stated view that "AI workloads are scaling faster than any single infrastructure approach can support," pushing Azure toward more specialized, workload-matched silicon rather than a one-size-fits-all compute tier.

Why it matters

This is another data point in hyperscalers diversifying their AI silicon supply beyond GPU-centric buildouts — Microsoft is pairing AMD CPUs and Helios-based inference systems alongside its existing Nvidia and in-house silicon commitments. For enterprise customers, three purpose-built VM tiers (data prep, chip design, inference) should mean better cost-to-performance matching for workloads that don't need a single monolithic GPU SKU. The chip-design angle also matters beyond AI narrowly: EDA/RTL simulation is itself a bottleneck in the semiconductor supply chain feeding the AI buildout, so faster Azure-hosted chip design cycles can compound into faster silicon availability industry-wide.