NVIDIA and Microsoft have expanded their partnership to deliver a unified accelerated computing stack for **Agentic AI deployment** across Windows devices, computers, and local environments.
The collaboration, announced at the Microsoft Build conference, aims to provide developers with the hardware and software required to build and scale autonomous systems. Consequently, this integration will allow enterprises to run complex reasoning models locally and in the cloud.
The companies introduced RTX Spark, a platform powering Windows PCs designed for personal agents with 1 petaflop of AI performance and up to 128GB of unified memory. In addition, the DGX Station for Windows deskside supercomputer will feature the NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip, delivering 20 petaflops of performance to run models with up to 1 trillion parameters.
Accelerating Agentic AI deployment in the Enterprise
To support enterprise workflows, NVIDIA open models are now available on Microsoft Foundry to facilitate **Agentic AI deployment** across cloud networks. This integration allows businesses to use models from NVIDIA, Anthropic, and OpenAI to build secure agentic systems on Microsoft Azure. Notably, the new NVIDIA Nemotron 3 Ultra reasoning model will be available this month on Foundry managed compute to assist with coding and research tasks.
Data Integration and Physical AI
Data processing is also receiving a significant upgrade, as NVIDIA accelerated computing is now integrated into Microsoft Fabric Data Warehouse. Internal benchmarks show SQL execution running up to six times faster than CPU-powered baselines. Furthermore, Microsoft is integrating NVIDIA’s physical AI tools, including the Cosmos 3 omnimodel, into Azure to help developers simulate and deploy autonomous robotic systems.
Secure Runtime and Infrastructure Upgrades
Security remains a primary focus of the partnership through the integration of NVIDIA OpenShell into GitHub Copilot, enhancing cybersecurity protocols. This open-source runtime isolates autonomous agents in sandboxed containers, evaluating outbound calls against strict policies before accessing files or networks. Meanwhile, Microsoft announced that its Fairwater Wisconsin AI factory is now live, running hundreds of thousands of Grace Blackwell systems.
Future Outlook for Cloud Infrastructure
Looking ahead, Microsoft has validated the next-generation NVIDIA Vera Rubin platform for deployment across its Azure data centers. This platform is designed to deliver up to 10 times more inference throughput per megawatt compared to previous architectures. Consequently, this validation ensures that future cloud infrastructure can support highly complex **Agentic AI deployment** workloads while optimizing energy efficiency and operational costs.
