Its Vera Rubin NVL72 rack-scale systems are now running in Microsoft Azure data centers, providing infrastructure for model training, inference and long-running agents. The companies are also working to extend confidential-computing protections across bare-metal and rack-scale systems.

The announcement reaches beyond the newest hardware. Microsoft Foundry is expanding access to Nvidia’s Nemotron 3.5 model family, including models for reasoning, retrieval, document parsing and content safety. Foundry Managed Compute will support custom model weights, while Microsoft’s managed post-training service handles infrastructure for organizations adapting models to their own data.

Deployment options are also moving closer to where companies produce data. Foundry Local is adding Nemotron support for Azure Local, and Nvidia’s Jetson Thor and Orin edge systems are planned for industrial uses such as robotics and machine vision. Windows PCs with RTX Spark hardware will demonstrate local large-language-model applications.

Several elements remain previews or future products. Nemotron Bolt and Transcribe are described as coming soon, and GPU-accelerated Windows 365 streaming has only a limited preview before a planned public preview in early 2027. Organizations evaluating the stack still need concrete regional availability, capacity and pricing; the announcement describes a broad roadmap rather than one uniformly available service.