SIAgentNews.com
← All news
DevelopingChipsBusiness

Nvidia and Microsoft launch Vera Rubin systems and Nemotron models in Azure

The partnership expands to include new rack-scale AI hardware, specialized Nemotron models for various tasks, and expanded edge computing support.

· First reported by MLQ.ai

15-foldcost cut versus proprietary frontier models

The short version

  • Vera Rubin NVL72 rack-scale systems are now available in preview within Azure data centers.
  • Nemotron 3.5 Super and Lightning models joined Foundry Managed Compute for agents and fast responses.
  • Azure Local will soon support RTX Pro 6000 Blackwell GPUs and Jetson platforms for robotics.
  • EY reported a 15-fold cost cut using Programmed Training to adapt Nemotron models for purchasing records.

Why it matters

This move signals a shift toward making high-performance AI infrastructure more accessible for specific enterprise tasks like document parsing and robotics. It also highlights a push toward 'confidential computing' to protect sensitive data during AI processing.

Next-generation AI hardware

Nvidia and Microsoft officially introduced the Vera Rubin NVL72 rack-scale systems during the Microsoft Ignite event. These systems are now running in Azure data centers, where they are currently available in a preview phase for users. Access to this hardware will gradually expand through Azure virtual machines.

The Vera Rubin architecture represents a significant step in infrastructure scaling for large language models. By integrating these systems into Azure, the companies are providing a foundation for high-demand AI applications. This hardware allows for more efficient processing of complex workloads that require massive computational power.

A key technical addition to this rollout is the extension of confidential computing to bare-metal and rack-scale AI systems. This technology uses Nvidia's attestation to verify the environment before any sensitive data, models, or keys are released. It ensures that private information remains secure even while being processed by powerful AI hardware.

Expanding the Nemotron suite

The partnership also expanded the availability of Nvidia's Nemotron 3.5 family on Foundry Managed Compute. This includes Nemotron 3.5 Super, which is designed for long-running agents, and Nemotron 3.5 Lightning, which is optimized for fast responses and tool use. These models are intended to help developers build diverse applications with different performance requirements.

Beyond the primary models, the update adds several specialized tools for specific data tasks. These include Nemotron Embed and Rerank for information retrieval, Nemotron Parse for extracting data from documents, and Nemotron Content Safety. The companies also announced that Nemotron 3.5 Bolt and a Transcribe speech model will be available soon.

To help businesses deploy these models, Microsoft added Programmed Training, which is a serverless post-training service. This allows users to customize models more easily without managing the underlying infrastructure. The companies highlighted how EY used this service to adapt Nemotron models for classifying purchasing records, achieving a 55% accuracy gain on their test set.

Edge and on-premises growth

The updates extend beyond the cloud to include Azure Local, which is Microsoft's version of Azure for on-premises environments. This service will now support RTX Pro 6000 Blackwell GPUs from various hardware partners. It also expands to include Jetson Thor and Jetson Orin platforms, which are specifically designed for robotics and machine vision applications.

Foundry Local is also receiving wider support for the Nemotron 3.5 models. Additionally, Nvidia's Cosmos physical AI models are scheduled to arrive on this platform. These updates aim to provide consistent AI capabilities for companies that need to run models locally or on specialized hardware rather than in a centralized cloud.

The companies are also focusing on local language-model experiences on standard Windows PCs. They demonstrated these capabilities using Nvidia's RTX Spark chip, which has recently appeared in mini desktops from manufacturers like Asus, Dell, and MSI. This suggests a trend toward bringing more sophisticated AI processing directly to the user's desktop hardware.

Cloud PC and scaling milestones

For remote work environments, the companies introduced Direct Stream for GPU-accelerated Windows 365 Cloud PCs. This feature supports up to two 4K streams at 60 frames per second, which is useful for high-resolution visual tasks. While a limited preview is currently open for sign-ups, the public preview is not expected until early 2027.

Nvidia also shared significant scaling milestones regarding the Azure infrastructure. They reported that Azure scaled to 10,240 GPUs on GB200 NVL72 systems in the MLPerf Training v6.1 benchmark. This represents the largest scale achieved in that specific round of testing, showcasing the capacity of the current partnership.

Finally, the update confirmed that Bing search models are now running on RTX Pro 6000 Blackwell GPUs. This integration highlights how the new hardware is being deployed for high-traffic, real-world applications. The ongoing rollout of these systems indicates a move toward more specialized and efficient hardware for diverse AI needs.

What to watch

  • The release date for Nemotron 3.5 Bolt and the Transcribe speech model.
  • The public launch of Direct Stream for Windows 365 Cloud PCs in 2027.
  • How many more enterprises adopt Programmed Training to cut costs on frontier models.

Try it

Check the Azure portal to see if Vera Rubin systems are available in your region's preview list.

Sources and links

First reported by MLQ.ai. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.

Liked it? Say why

Add a line for other readers.

No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.

More AI news