NVIDIA and Microsoft Extend AI Infrastructure From Azure to Windows PCs

NVIDIA and Microsoft are expanding their accelerated computing partnership across Azure, industrial systems and Windows PCs, adding access to Vera Rubin infrastructure, open models and local AI deployment tools. The companies also highlighted Microsoft’s use of NVIDIA software and GPUs for its own models and services.
- Vera Rubin NVL72 systems are running in Azure data centers.
- Microsoft Foundry is adding NVIDIA open models and customization capabilities.
- EY reported 55% higher evaluation accuracy and 15x lower cost in one Nemotron use case.
- Azure Local is expanding support for NVIDIA GPUs and Jetson edge platforms.
- Windows 365 Direct Stream public preview is planned for early 2027.
Vera Rubin infrastructure arrives in Azure
NVIDIA Vera Rubin NVL72 systems are running in Microsoft Azure data centers, providing infrastructure for demanding reasoning workloads, agentic applications and large-scale model training and inference. The companies are also working to extend confidential computing to bare-metal and rack-scale AI infrastructure.
According to NVIDIA, confidential computing and attestation will help customers verify the computing environment before releasing sensitive models, data or keys. The companies say this work is intended to protect proprietary and regulated information as AI workloads use increasingly interconnected accelerators.
NVIDIA and Microsoft are also collaborating on power efficiency through NVIDIA MaxLPS. In MLPerf Training v6.1, Azure scaled to 10,240 NVIDIA GPUs with GB200 NVL72 systems on the DeepSeek-V3 model, which NVIDIA described as the largest scale in that round. In MLPerf Inference v6.1, Azure’s NVIDIA GB300 NVL72 systems led cloud submissions on DeepSeek-R1 interactive throughput by over 10%, according to Signal65.
Open models gain customization and deployment options
The partnership is expanding the models and services available through Microsoft Foundry. Microsoft’s 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, connecting model customization with Microsoft’s managed services.
EY used Programmed Training in Microsoft Foundry to adapt NVIDIA Nemotron open models for classifying corporate purchasing records. NVIDIA says the work improved accuracy on EY’s evaluation dataset by 55% and lowered cost by 15x compared with proprietary frontier models.
The expanded NVIDIA Nemotron 3.5 lineup includes Super for long-running agents across coding, research and business workflows, and Lightning for applications requiring fast responses and tool use. Nemotron Embed and Rerank support information retrieval, Nemotron Parse extracts structured information from documents, and Nemotron Content Safety is designed to detect harmful content. Nemotron 3.5 Bolt and Transcribe are coming soon, according to NVIDIA.
AI moves closer to enterprise data and devices
Organizations will have more options to run models and agents outside the cloud. Expanded support for Nemotron 3.5 in Foundry Local brings reasoning, coding and tool-using agents to Azure Local environments. Azure Local will also support NVIDIA RTX PRO 6000 Blackwell GPUs from hardware partners, as well as NVIDIA Jetson Thor and Jetson Orin platforms for industrial uses including machine vision, robotics and autonomous operations.
On Windows PCs powered by NVIDIA RTX Spark, NVIDIA and Microsoft will demonstrate local large language model applications. For Cloud PC customers, Direct Stream capabilities are coming to GPU-accelerated Windows 365 Cloud PCs, supporting up to two 4K streams at 60 frames per second. A limited preview is open for sign-ups, with public preview planned for early 2027.
The companies are also bringing NVIDIA BioNeMo Agent Toolkit and NVIDIA ALCHEMI Toolkit into the Microsoft Discovery catalog. At the same time, Microsoft AI uses NVIDIA Dynamo for reinforcement learning rollout inference during post-training, coordinating inference across a large GPU fleet and helping manage model updates and failures. The announcement positions the partnership as a broader effort to supply infrastructure, models and deployment tools across cloud, enterprise and local computing.


