Skip to content
AI.info

The Pulse

Microsoft widens access to Nvidia Vera Rubin systems on Azure and adds Nemotron 3.5 models to Foundry

Nvidia and Microsoft are expanding access to AI infrastructure and models across Azure, local systems and Windows PCs. The companies will demonstrate local language models on RTX Spark PCs and open a limited preview of GPU-accelerated Cloud PC streaming.

Microsoft widens access to Nvidia Vera Rubin systems on Azure and adds Nemotron 3.5 models to Foundry

Nvidia and Microsoft are expanding access to Nvidia-powered Azure virtual machines and AI models, while preparing demonstrations of language models running locally on Windows PCs. Nvidia described the updates in a November 19 post about the companies' work at Microsoft Ignite.

Source blogs.nvidia.com

The collaboration spans cloud infrastructure, on-premises systems and industrial devices. On Windows PCs powered by Nvidia RTX Spark, the companies will demonstrate how developers can run large language model applications directly on the device.

Windows and cloud access

The companies are also bringing Direct Stream capabilities to GPU-accelerated Windows 365 Cloud PCs for graphics-intensive applications. The capabilities support up to two 4K streams at 60 frames per second, with low latency even over challenging networks, according to Nvidia.

Sign-ups are open for a limited preview, and a public preview is planned for early 2027. The local PC demonstrations are separate from the Cloud PC streaming capabilities.

Vera Rubin NVL72 systems are running in Azure data centers, with access to virtual machines powered by the systems expanding. The infrastructure is designed for workloads including reasoning applications, model training and inference.

Microsoft and Nvidia are working to extend confidential computing to bare-metal and rack-scale AI infrastructure. Nvidia said its confidential computing and attestation technology lets customers verify an environment before releasing sensitive models, data or keys; the companies are working to extend those protections across interconnected infrastructure. They are also collaborating on power efficiency using Nvidia MaxLPS.

In MLPerf Training v6.1, Azure scaled to 10,240 Nvidia GPUs using GB200 NVL72 systems on the DeepSeek-V3 model, which Nvidia said was the largest scale in that round. For MLPerf Inference v6.1, the company cited Signal65 as saying Azure's GB300 NVL72 systems led cloud submissions on DeepSeek-R1 interactive throughput by over 10%.

Model customization

Microsoft Foundry is expanding the Nvidia models and services available to developers. Its serverless Programmed Training offering lets developers control post-training workflows while Foundry manages the underlying infrastructure. Foundry Managed Compute adds support for deploying custom model weights.

Nvidia said EY used Programmed Training to adapt Nemotron open models for classifying corporate purchasing records. On EY's evaluation dataset, the work improved accuracy by 55% and lowered cost by 15x compared with proprietary frontier models, according to the post.

The expanded Nemotron 3.5 lineup on Foundry Managed Compute includes Super, designed for long-running agents working across coding, research and business workflows, and Lightning, for applications requiring fast responses and tool use. Other models cover retrieval, document parsing and content safety. Bolt, a reasoning model, and Transcribe, for speech recognition, are coming soon.

Local and scientific workloads

Expanded Nemotron 3.5 support in Foundry Local brings reasoning, coding and tool-using agent capabilities to Azure Local environments. Retrieval models can find evidence across text, document images, charts and tables.

Azure Local will also support Nvidia RTX PRO 6000 Blackwell GPUs from hardware partners for on-premises deployments. The platform is expanding to Nvidia Jetson Thor and Jetson Orin for industrial applications including machine vision and robotics, while Nvidia physical AI models, including Cosmos, are being brought to Foundry Local.

Nvidia's BioNeMo Agent Toolkit and ALCHEMI Toolkit are expanding into the Microsoft Discovery catalog for research in life sciences, drug discovery, materials and chemical science. An initial integration uses RFD4-Proteina, developed with the University of Washington's Institute for Protein Design, to generate candidate protein structures and amino acid sequences for further evaluation and laboratory validation.

Microsoft AI also uses Nvidia Dynamo software for reinforcement learning rollout inference during post-training. The software coordinates inference across a GPU fleet and helps manage requests, model updates and recovery from failures.

Sources

  1. Original source · Blog post NVIDIA blogs.nvidia.com

Companies in this story

Related stories

Explore

More articles