Microsoft will provide a trio of new Azure instances powered by AMD Helios racks, it said on July 20.
These span HDv2 VMs for data processing, HXv2 VMs for electronic design automation (EDA) and ND MI455X v7 VMs for AI inference workloads, it said.
With the news today, Microsoft joins Meta and OpenAI as large-scale Helios buyers, as AMD and partners gear up to start shipping the systems at scale in Q3.
HDv2 VMs will support “demanding CPU workloads from AI customers,” Azure said in a canned statement. It cited the likes of agent coordination, data preparation, reinforcement learning and heavy search workloads.
HDv2 VMs will feature 500 6th Gen AMD EPYC CPU cores, 4 TB of RAM, 32 TB of local NVMe storage and 400 Gbs Azure Boost networking, it said.
The HXv2 VMs build on HX instances first released in 2023, primarily used for chip design. The v2 instances boost per core performance, and add "800 Gb InfiniBand, enable large-scale MPI-based simulations and make HXv2 an ideal fit for a wide variety of HPC customers," Azure said today.
ND MI455X v7 meanwhile are “designed for the reasoning, search and agentic workloads behind modern AI services” and represent an attempt by capacity-constrained Azure to offer more firepower to inference buyers.
*The Stack *has asked Microsoft how many Helios racks it is buying or the level of MW/GW capacity it planned to deploy and will update this story if we get a clear answer. Helios starts shipping meaningfully in Q3 said AMD.
Microsoft did not share a date for general availability.
What are AMD Helios racks?
Helios integrates AMD’s Instinct MI400 Series GPUs, Venice EPYC CPUs and Pensando NICs in a double-wide rack built to a Meta-designed rack standard, first introduced at the Open Compute Project conference in October 2025.
The new open rack wide (ORW) form factor for AI workloads was released as an open standard amid what the OCP described as the need for a “fundamental rethinking of how physical infrastructure is built” at scale.
Data center partners typically still often receive “competing design inputs from various hyperscalers, enterprise users, and technology providers” it said last year, adding, “this lack of standardization leads to inefficiencies, slows down deployment, and ultimately hinders innovation…”
In 2026 AMD said it expects the “vast majority” of its MI450 series GPUs to be sold as “rack scale solutions.” HPE and Lenovo are both assembling and selling Helios racks; HPE shipping them with purpose-built HPE Juniper Ethernet switches and software optimised for high-bandwidth networking.
Azure ramps CapEx, AMD, production
Speaking during a May XX earnings call, AMD CEO Dr Lisa Su said: “We will be ramping Helios in the second half of the year…starting with initial volume in Q3 with a significant ramp in Q4 and then continuing to ramp in Q1.”
“There is tightness in the supply chain, there's certainly tightness in data center build-outs, but we are confident in our ability to supply to the levels of growth that we're talking about,” she said, adding, amid grid constraints in many regions, “there is much more power that's coming online in 2027.”
Microsoft posted CapEx of $31.9 billion during its quarter ended May 2026. Some two-thirds of CapEx was for "short-lived assets, primarily GPUs and CPUs," Microsoft CFO Amy Hood told investors.
Sign up for The Stack #
Interviews, insight, intelligence, and exclusive events for digital leaders.
No spam. Unsubscribe anytime.