NVIDIA and Microsoft are expanding their full-stack collaboration to help organizations build, customize and deploy intelligent applications across Azure data centers, enterprise infrastructure, industrial systems and Windows PCs. Announced at Microsoft Ignite, the expanded partnership includes Azure virtual machines powered by NVIDIA Vera Rubin NVL72 systems, broader access to NVIDIA open models through Microsoft Foundry, and new options for running AI closer to enterprise data.
Microsoft is also using NVIDIA technology to accelerate its own models and services, including NVIDIA Dynamo for post-training inference and the NVIDIA RTX PRO 6000 Blackwell GPU for the Bing search model.
NVIDIA Vera Rubin NVL72 systems on Azure
The NVIDIA Vera Rubin NVL72 system running in an Azure data center is designed for demanding inference workloads, AI agent applications, and large-scale model training and inference.
Microsoft and NVIDIA are also working to extend confidential computing to bare-metal systems and rack-scale AI infrastructure. The initiative is intended to protect sensitive models and data during processing, supporting organizations that handle sensitive or regulated information.
NVIDIA Confidential Computing and digital certificates help customers validate computing environments before exposing sensitive models, data or keys. As AI workloads increasingly span interconnected accelerators, enterprises are seeking to extend these protections across their AI factory infrastructure.
To help data centers achieve higher performance within available power limits, NVIDIA and Microsoft are also collaborating on power efficiency using NVIDIA MaxLPS.
The Vera Rubin preview on Azure is based on joint NVIDIA-Microsoft design work and performance optimizations, highlighted by recent NVIDIA Blackwell-based MLPerf results. In MLPerf Training v6.1, Azure scaled to 10,240 NVIDIA GPUs using GB200 NVL72 systems for the DeepSeek-V3 model, making it the largest system in that round.
In MLPerf Inference v6.1, Signal65 reported that the NVIDIA GB300 NVL72 system on Azure led cloud providers in DeepSeek-R1 interactive throughput by more than 10%.
Customizing open models with Microsoft Foundry
Enterprise AI agents must reason through complex tasks, locate relevant company information and understand business documents. NVIDIA and Microsoft are expanding the models and services available through Microsoft Foundry to help developers add these capabilities to business applications.
Microsoft’s serverless program training service allows developers to control post-training workflows while Foundry manages the underlying infrastructure. This enables enterprises to focus on using business data to evaluate model responses and improve model behavior.
Foundry Managed Compute adds support for deploying custom model weights and helps connect model customization with deployment through Microsoft’s managed services.
EY used programmatic training from Microsoft Foundry to adapt the NVIDIA Nemotron open model for classifying corporate purchasing records. According to the study, the customized model increased accuracy on evaluation datasets by 55% and reduced costs by 15 times compared with proprietary frontier models. The results illustrate how companies can adapt open models for specialized business tasks.
The expanded NVIDIA Nemotron 3.5 lineup in Foundry Managed Compute supports a range of enterprise applications. Nemotron 3.5 Super is designed for long-running agents working across coding, research and business workflows, while Nemotron 3.5 Lightning targets applications that require fast responses and tool use.
Other models support information retrieval, document processing and safety tasks. Nemotron Embed and Rerank help agents find relevant information, Nemotron Parse extracts structured data from documents, and Nemotron Content Safety detects harmful content. Nemotron 3.5 Bolt, designed for high-throughput, low-latency inference, and Transcribe for speech recognition are coming soon.
Bringing AI closer to enterprise data
NVIDIA and Microsoft are expanding deployment options beyond the cloud so organizations can run models and agents closer to their data while retaining control of their infrastructure and operations.
Expanded support for Nemotron 3.5 in Foundry Local brings reasoning, coding and tool-using agents to Azure Local environments. Search models can help applications find evidence across text, document images, graphs and tables.
Azure Local also supports NVIDIA RTX PRO 6000 Blackwell GPUs through hardware partners, giving organizations additional options for on-premises AI deployments.
Azure Local extends to NVIDIA Jetson Thor and NVIDIA Jetson Orin, an edge AI platform for industrial applications such as machine vision, robotics and autonomous operations. NVIDIA physical AI models from Cosmos have also been brought to Foundry Local.
For Windows PCs equipped with NVIDIA RTX Spark, NVIDIA and Microsoft will introduce a local large language model experience at Ignite. The demonstration will show how developers can run AI applications directly on their devices.
For organizations using cloud PCs, the companies plan to bring Direct Stream capabilities to GPU-accelerated Windows 365 cloud PCs. The features support demanding graphics applications with up to two 4K streams at 60 frames per second and low latency, including over difficult networks. A limited preview is available for sign-up, with a public preview expected in early 2027.
Connecting scientific AI with Microsoft Discovery
For scientific research, the NVIDIA BioNeMo Agent Toolkit and NVIDIA ALCHEMI Toolkit are being added as extensions to the Microsoft Discovery Catalog. The integration brings life sciences, drug discovery, materials science and chemical science models to Microsoft’s agent-based research platform.
The BioNeMo Agent Toolkit allows scientists to run computational biology workflows in Microsoft Discovery using natural-language instructions. Agents coordinate the tools and model calls needed to complete the work.
The initial integration uses the RFD4-Proteina model, developed in collaboration with the University of Washington’s Protein Design Institute. The model generates candidate protein structures and amino acid sequences for further evaluation and laboratory validation.
Accelerating Microsoft models and services
Alongside expanding customer access to NVIDIA technology, Microsoft uses NVIDIA software and accelerated computing in its own AI development and services.
Microsoft AI uses NVIDIA Dynamo to support post-training reinforcement-learning rollout inference. This computationally intensive process produces experiences that are used to improve the model.
Dynamo orchestrates inference across large GPU fleets to optimize performance and help manage requests, model updates and failure recovery. The deployment is based on joint engineering between Microsoft AI and NVIDIA.
NVIDIA’s Ignite demonstrations will cover enterprise applications, robotics, high-speed data processing and local computing using hardware such as the NVIDIA GB300 NVL72 rack and RTX Spark devices. The company will also offer sessions, hands-on labs and booth theater presentations.
Source: blogs.nvidia.com


