NEWS →  ·  Clouds · NVIDIA + Microsoft

Vera Rubin NVL72 is running in Azure data centres — and VM access is expanding, as a preview

Nov 19, 2026

At Microsoft Ignite today, Nov 19, 2026, NVIDIA said Vera Rubin NVL72 systems are running in Azure data centres, and that the two companies are expanding access to Azure virtual machines powered by them. NVIDIA's own post calls it Azure's Vera Rubin preview — and preview is the operative word: the announcement publishes no price, no region list, no VM series name and no general-availability date. The picture is NVIDIA's own announcement art for the post.

Vera Rubin NVL72 is running in Azure data centres — and VM access is expanding, as a preview

Image: NVIDIA via blogs.nvidia.com

What NVIDIA's post (Nov 19, 2026, 20:00 UTC, read first-hand by this desk) actually says: the Rubin systems are in place in Azure data centres, providing infrastructure for reasoning, agentic applications and large-scale training and inference; confidential computing is being extended to bare-metal and rack-scale Rubin infrastructure, with NVIDIA Confidential Computing and attestation; and the two companies are working on power efficiency with NVIDIA MaxLPS. The performance colour is Blackwell, not Rubin numbers: in MLPerf Training v6.1, Azure scaled to 10,240 GPUs on GB200 NVL72 systems training DeepSeek-V3, the largest scale in the round, and in MLPerf Inference v6.1 Azure's GB300 NVL72 systems led cloud submissions on DeepSeek-R1 interactive throughput by over 10%, per Signal65, as cited by NVIDIA.

The same post widens Microsoft Foundry's open-model shelf: NVIDIA Nemotron 3.5 Super (long-running agents) and Lightning (fast responses and tool use) are available on Foundry Managed Compute, with Bolt and Transcribe listed as coming soon, alongside Nemotron Embed, Rerank, Parse and Content Safety models; Foundry Managed Compute now supports deploying custom model weights, and the serverless Programmed Training route handled EY's adaptation of Nemotron models for purchasing-record classification — a 55% accuracy improvement on EY's evaluation dataset at 15 times lower cost than proprietary frontier models, both figures NVIDIA's, from its customer example. Beyond Azure, Azure Local will also support RTX PRO 6000 Blackwell GPUs from hardware partners, and GPU-accelerated Windows 365 Cloud PCs gain Direct Stream in a limited preview, with public preview planned for early 2027.

The honest framing for the scoreboard: Azure is not first with Rubin. Oracle announced general availability of Vera Rubin NVL72 on OCI on Oct 27, 2026, and CoreWeave and Google Cloud have earlier Rubin deployments. Microsoft's step is a preview expansion by the largest cloud of the three — a supply signal, not a bookable product. No Azure Rubin rate exists to check, so nothing about this announcement changes a single price on this board today.

What it means if you rent GPUs

For renters and builders, this is the next generation announcing its queue position. Rubin-class racks running inside Azure data centres means the capacity the whole market is waiting on is physically arriving at hyperscale; what would make it a booking fact — a VM series, regions and a price — is exactly what the announcement does not contain. This desk will treat the preview as real and the timeline as unstated, and will check Azure's own pages for the first priced Rubin VM the way it checks every other provider: from the provider's own figures, dated.

Go deeper

← All news