Microsoft Azure opens NVIDIA Vera Rubin systems to customers
At Microsoft Ignite, Azure opened machines with NVIDIA Vera Rubin NVL72 systems. EY tells an open model cut its cost 15x.
NVIDIA and Microsoft extended their work at Microsoft Ignite. Customers get Azure virtual machines that use NVIDIA Vera Rubin NVL72 systems. Microsoft Foundry, its AI service for developers, adds NVIDIA open models and tools to adapt them. The two companies also add alternatives to operate AI on site, at the edge and on Windows PCs.
What is new
Foundry Managed Compute accepts custom model weights. Nemotron 3.5 models, such as Super for long agent tasks, are on Foundry. Azure Local, the Microsoft system for on site use, gets NVIDIA RTX PRO 6000 Blackwell chips. A small preview of Windows 365 Cloud PC streams is open. The public preview is planned for the start of 2027.
The numbers
EY adapted a Nemotron model to label company purchase records. NVIDIA tells that precision grew 55% and cost fell 15x against closed models. In the MLPerf test, Azure scaled to 10,240 NVIDIA GPUs, the largest in that test.
What is not known
The source gives no price or date for Vera Rubin on Azure. It calls this a preview. Only the EY test gives a test result for adapted models.
Insight: NVIDIA is the common layer from cloud to sovereign plans
In October, NVIDIA added Confidential Bare Metal. It also works with the European Broadcasting Union and South Korea on sovereign AI, which is AI that a country or a company controls. It is also in Azure, Azure Local and Windows PCs. This seems to be a plan to make NVIDIA software the same layer in each location where a customer keeps data. Find the cost to change chip maker.
Sources
Posted