Money and business in the Middle East.

AI

Nvidia's Vera Rubin NVL72 Systems Go Live in Azure Data Centers

Nvidia and Microsoft are widening Azure access to virtual machines running Vera Rubin NVL72 systems.

· Last verified: 20 Nov 2026 (Nvidia company blog; Signal65)

Summary

  • Nvidia and Microsoft are widening Azure access to virtual machines running Vera Rubin NVL72 systems.
  • Microsoft Foundry is adding Nvidia's open Nemotron 3.5 models with enterprise customization and deployment tools.
  • EY cut costs 15x by adapting Nemotron models instead of using proprietary frontier systems, Nvidia said.

The latest

Nvidia's next-generation Vera Rubin NVL72 systems are now running inside Azure data centers, with Microsoft expanding customer access to virtual machines built on them, Nvidia said in a blog post timed to Microsoft Ignite. The two companies also opened broader access to Nvidia's open Nemotron models through Microsoft Foundry, alongside new options for running AI workloads on local and edge hardware.

Details

  • The infrastructure: Nvidia said Vera Rubin NVL72 systems are operating in Azure data centers, aimed at reasoning workloads, agentic applications and large-scale model training and inference. The Azure offering is described as a preview; Nvidia did not give a general availability date or pricing for the virtual machines.
  • The benchmarks: In MLPerf Training v6.1, Azure scaled to 10,240 Nvidia GPUs using GB200 NVL72 systems on the DeepSeek-V3 model, the largest scale in that round, according to Nvidia. In MLPerf Inference v6.1, Azure's GB300 NVL72 systems led cloud submissions on DeepSeek-R1 interactive throughput by over 10%, per Signal65.
  • The EY result: EY used Microsoft's Programmed Training service in Foundry to adapt Nemotron open models for classifying corporate purchasing records, Nvidia said. Accuracy on EY's evaluation dataset rose 55% and cost fell 15 times compared with proprietary frontier models.
  • The model lineup: Nemotron 3.5 Super targets long-running agents across coding, research and business workflows, while Lightning is built for fast responses and tool use. Companion models cover retrieval, document parsing and content safety. Nvidia said Bolt, for high-throughput low-latency reasoning, and Transcribe, for speech recognition, are coming soon.
  • The security work: The companies said they are extending confidential computing to bare-metal and rack-scale AI infrastructure, protecting models and data during processing. Nvidia said attestation lets customers verify the computing environment before releasing sensitive models, data or keys — a pitch aimed at regulated industries.
  • On-premises push: Azure Local will support Nvidia RTX PRO 6000 Blackwell GPUs from hardware partners and is expanding to the Jetson Thor and Jetson Orin edge platforms for machine vision, robotics and autonomous operations. Nemotron 3.5 support in Foundry Local extends reasoning and coding agents to those environments.
  • Windows and Cloud PCs: Nvidia said Direct Stream capabilities are coming to GPU-accelerated Windows 365 Cloud PCs, supporting up to two 4K streams at 60 frames per second with low latency. A limited preview is open for sign-ups, with public preview planned for early 2027.
  • Microsoft as customer: Microsoft is using Nvidia technology internally, deploying Dynamo inference software for reinforcement learning rollouts during post-training and RTX PRO 6000 Blackwell GPUs for Bing search models. Nvidia said the companies are also working on power efficiency in data centers using MaxLPS.
  • Scientific research: Nvidia's BioNeMo Agent Toolkit and ALCHEMI Toolkit are expanding into the Microsoft Discovery catalog for life sciences, drug discovery and materials science. An initial integration uses the RFD4-Proteina model, developed with the University of Washington's Institute for Protein Design, to generate candidate protein structures.

Between the lines

The EY figures are the commercial argument in this announcement: a 15x cost reduction against proprietary frontier models positions customizable open weights as the cheaper enterprise path. Pairing that with Foundry Managed Compute and on-premises Nemotron support ties the cost pitch to Microsoft's infrastructure rather than to open models generally.

What's next

Watch for general availability dates and pricing on the Vera Rubin NVL72 Azure virtual machines, the releases of Nemotron 3.5 Bolt and Transcribe, and the Windows 365 Direct Stream public preview slated for early 2027.

Source: Nvidia company blog; Signal65

Nvidia's Vera Rubin NVL72 Systems Go Live in Azure Data Centers · INXEN