Subscribe
Hyperscalers

Nvidia's Vera Rubin systems are now running in Azure — in preview, with no price yet

Friday, November 20, 2026 · 4 min read

Microsoft says it was the first cloud to validate the NVL72 system in March and run production systems in August; customers can now try it, but there is no region list, pricing or launch date.

Bullish$MSFT$NVDA

📊 Scorecard: our Bullish call on $MSFT is scored from $498.30, the last close before it ran. Its first result comes after the next close.

Nvidia's Vera Rubin systems are now running in Azure — in preview, with no price yet

The News

Nvidia's Vera Rubin NVL72 systems are now running in Microsoft's Azure datacenters, the two companies announced at Microsoft Ignite on November 19. Vera Rubin is Nvidia's newest AI computing platform — the hardware that trains and runs the biggest AI models. Access to Azure virtual machines powered by the systems is expanding in preview, which means selected customers can try them before a full public launch.

The announcement comes with notable gaps. Neither company gave a virtual machine series name, a list of regions where the systems are available, pricing, eligibility criteria or a general-availability date. Microsoft says it became the first cloud to bring up a Vera Rubin NVL72 system for validation in March, and delivered the first production Vera Rubin systems to its datacenters in August. In plain terms, Microsoft has been testing the hardware since the spring and running it in production since the summer; this is the first time customers can use it.

The announcement also covers confidential computing on Vera Rubin — a technology preview starting with confidential bare-metal instances — Nvidia's Nemotron 3.5 open models on Microsoft Foundry with managed compute, Azure Local support for Nvidia RTX PRO 6000 Blackwell graphics processors, and Microsoft's own use of Nvidia hardware for Bing search models. In MLPerf Training version 6.1, an industry benchmark, Azure scaled to 10,240 Nvidia graphics processors on DeepSeek-V3. Separately, on November 18, Intel and Microsoft announced that Microsoft is the first company to commit Intel Foundry's EMIB packaging to Azure's own first-party silicon, alongside multi-year Xeon supply including next-generation Diamond Rapids parts. (Intel/Microsoft announcement of November 18, 2026 — see the Intel newsroom release; this is a second source beyond the AIStockWire article linked for this story.)

First to validate, first to run in production — but Oracle got Vera Rubin to customers first, with general availability on October 27.

Why It Matters

The news matters because Vera Rubin is the next rung on the AI hardware ladder. Cloud providers compete on who has the newest, fastest chips first, because the companies training the biggest AI models will go where the hardware is. Microsoft being first to validate the system in March and first to run production systems in August — and now first to open it to customers in preview — is a claim to leadership in that race, even without a price or a launch date.

The missing details matter too. A preview with no price, no regions and no general-availability date is a soft launch: customers can try the systems, but they cannot plan a budget or a rollout around them yet. That is normal for early hardware — the systems are scarce, and Microsoft and Nvidia are still learning how they perform at scale — but it means the revenue impact is months away, not weeks.

The wider Ignite announcements show how deep the partnership runs. Microsoft is not just renting Nvidia's chips; it is using them for its own search models, offering Nvidia's open models on its Foundry platform, and adding confidential computing — a way to process sensitive data so that even the cloud provider cannot see it. The Intel EMIB deal, announced the day before, is the other side of the same coin: Microsoft is also building its own silicon, and it wants Intel's packaging technology for it.

Analyst's Take

Our call is Bullish. Azure was first to validate the system in March and first to run production systems in August — but Oracle reached customers first, with Vera Rubin generally available on Oracle Cloud since October 27, while Azure's offering is still a preview. And the 10,240-graphics-processor benchmark ran on Nvidia's GB200 NVL72 systems, not on Vera Rubin, so it shows Azure's scale on the previous generation, not on Rubin. The missing price and date are a limit, not a flaw — previews are how new hardware ships. The risk is execution: if the preview stays narrow for months, rivals with their own Rubin systems will catch up.

What to watch next is the general-availability date and the price. Those two numbers will tell the market how fast Vera Rubin becomes a real product rather than a preview. Reader poll: Will Microsoft win the Vera Rubin race? Vote: Yes, first mover, No, rivals catch up, Too early to tell.

Are you bullish or bearish on this story?

By The Numbers

10,240Nvidia GPUs Azure scaled to on DeepSeek-V3 in MLPerf Training v6.1
March 2026When Microsoft first brought up a Vera Rubin NVL72 system for validation
August 2026When the first production Vera Rubin systems were delivered to Azure datacenters
Nov 19When the Vera Rubin preview was announced at Microsoft Ignite

FAQ

What is a virtual machine?

A virtual machine is a computer that exists as software inside a bigger physical computer. Cloud customers rent virtual machines instead of buying hardware, and they can choose how much processing power, memory and storage each one has.

What is a preview, and what is general availability?

A preview is an early, limited release of a product — selected customers can try it, but it is not finished or widely available. General availability, or GA, is the full public launch, when any customer can buy it at a published price.

What is MLPerf?

MLPerf is an industry benchmark — a standard set of tests that measures how fast AI hardware trains models and runs them. Companies publish their MLPerf results so customers can compare systems from different makers on the same tasks.

What is confidential computing?

Confidential computing is a way of processing sensitive data so that it stays encrypted — scrambled and unreadable — even while it is being used, so that even the cloud provider running the hardware cannot see it. It is used for data that must stay private, like health or financial records.

👁

Source: AI Stock Wire

Get the daily briefing. AI news with a markets lens — tickers, bullish/bearish reads, zero hype. Subscribe free on Substack

More on $MSFT

All $MSFT coverage →

More from AI Agent Update