NVIDIA and Microsoft have expanded their long-standing collaboration to offer enterprises new ways to customize, deploy, and secure AI models across both cloud and on-premises environments. The initiative brings advanced infrastructure, confidential computing, and tailored AI experiences from the Azure cloud down to industrial systems and Windows PCs.
What Happened
On November 19, 2026, at Microsoft Ignite, NVIDIA and Microsoft announced a broad extension of their accelerated computing partnership. The companies revealed wider access to Azure virtual machines powered by NVIDIA Vera Rubin NVL72 systems, enabling large-scale AI training, inference, and deployment for enterprise applications. Through Microsoft Foundry, customers can now leverage NVIDIA open models and customization tools, including recent releases such as Nemotron 3.5, with options to locally deploy enterprise AI on Azure Local, edge platforms such as NVIDIA Jetson Thor and Orin, and even Windows PCs via NVIDIA RTX Spark. Confidential computing, enhanced power efficiency, and deeper integration for scientific research through Microsoft Discovery were also highlighted among the updates.
Key Takeaways
- Azure now offers VMs powered by NVIDIA Vera Rubin NVL72 for enterprise AI workloads.
- Microsoft Foundry expands support for NVIDIA Nemotron 3.5 open models and programmatic customization.
- Confidential computing protections available for sensitive AI workloads at bare-metal and rack scale.
- Enterprises can run AI on premises with Azure Local and hardware including RTX PRO 6000 Blackwell GPUs and NVIDIA Jetson platforms.
- Windows PCs with NVIDIA RTX Spark can support local generative AI model use cases.
- Direct Stream brings GPU acceleration to Windows 365 Cloud PCs, with public preview in early 2027.
The Details
Wider Azure support for NVIDIA Vera Rubin NVL72 equips enterprises with infrastructure tailored for demanding applications in agent reasoning and large model deployment. MLPerf Training v6.1 benchmarks showed Azure scaling to 10,240 NVIDIA GPUs with this system, and MLPerf Inference v6.1 found Azure’s NVIDIA GB300 NVL72 leading in cloud-based interactive throughput.
Microsoft Foundry’s enterprise AI offering now extends all the way from customization to deployment, supporting Programmed Training for post-training optimization and Foundry Managed Compute for deploying custom model weights. The new Nemotron 3.5 model lineup covers everything from robust, long-running enterprise agents (Super), to fast-responding applications (Lightning), with specialist models for embedding, document parsing, safety, and forthcoming reasoning and transcription needs. Case highlights include EY boosting model accuracy and reducing costs by adapting Nemotron with Foundry tools.
Deployment options now span Azure’s cloud, on-premises data centers, and edge environments. Azure Local and Foundry Local enable reasoning and tool-using agents close to enterprise data, across text and image sources. Hardware support grows to include RTX PRO 6000 Blackwell for on-premises use and Jetson Thor/Orin for industrial edge. On Windows, RTX Spark-powered PCs can run generative AI locally, demonstrated for developers at Ignite. Windows 365 Cloud PCs will add Direct Stream GPU acceleration for 4K, low-latency graphics application streaming – with previews available for sign-up and a wider rollout planned for early 2027.
Scientific AI also gains integration, as toolkits like NVIDIA BioNemo and ALCHEMI become available in Microsoft’s Discovery catalog, supporting natural-language-accessible workflows for life sciences and materials research. These allow scientists to generate protein structure candidates and work with specialized computational biology models directly in the Microsoft ecosystem.
Background and Context
NVIDIA and Microsoft have built their partnership around AI acceleration and cloud computing, with Microsoft Azure long serving as a preferred platform for large-scale GPU deployment and research. As enterprise demands evolve to require not only large-scale cloud models but also customizable, secure, and local deployments, the two companies have moved to address needs spanning data privacy, industrial edge computing, and scientific research collaborations. The Vera Rubin and Blackwell GPU lines reflect NVIDIA’s ongoing hardware innovation, while Foundry and Azure Local point to Microsoft’s growing role in AI model lifecycle management.
BytesWall Analysis
This expansion positions NVIDIA and Microsoft as key enablers for enterprises moving beyond generic AI models towards deeply customizable, vertically integrated solutions. By uniting top-performing infrastructure (Vera Rubin NVL72, Blackwell GPUs) with flexible deployment (cloud, local, edge, and PC), the partnership targets a range of industries: corporate IT, industrial automation, life sciences, and more. Confidentiality features help regulated or data-sensitive sectors embrace AI, while co-developed power efficiencies reduce operational costs and barriers to scaling.
Compared to purely cloud-based or fixed-model solutions, the approach lets businesses tune AI to proprietary data and workflows — as the EY example illustrates — unlocking significant accuracy and cost gains. The inclusion of robust local and edge options addresses growing demand for data residency, privacy, and latency-sensitive use cases. The deep MLPerf benchmark results further cement NVIDIA as an infrastructure leader, while Microsoft’s control of both cloud and productivity platforms ensures broad customer reach.
Nevertheless, this full-stack complexity raises integration and operational challenges. In-house teams will need sufficient expertise in model training, deployment, and security. Competing platforms, notably from other cloud providers and emerging AI chipmakers, are also racing to address these needs with their ecosystems. BytesWall believes that Microsoft and NVIDIA still set the pace for scalable, enterprise-ready AI infrastructure today. Decision-makers should evaluate how their data, security, and customization requirements align with these evolving, hybrid cloud-to-edge offerings.
What to Watch Next
- How quickly will enterprises transition from cloud-only AI workflows to hybrid and edge deployments?
- Will confidential computing protections scale seamlessly as AI factory complexity grows?
- How will rival cloud providers and hardware vendors respond with their own end-to-end AI stacks?
Read more in our AI News Hub
Source: read the original announcement. BytesWall adds summary and analysis.