What the Unified NVIDIA–Microsoft AI Stack Actually Is
The unified NVIDIA–Microsoft AI stack is a coordinated hardware, software and cloud platform that lets developers build, run and scale agentic AI across Windows PCs, local infrastructure and Azure services using one consistent toolchain and security model. Announced at Microsoft Build by Jensen Huang and Satya Nadella, the collaboration connects RTX Spark Windows PCs, DGX Station for Windows, Azure Local, Microsoft Foundry, Fabric and GitHub Copilot into a single Windows AI stack. For enterprise AI infrastructure planners, the shift is that Windows is treated as a managed endpoint for local agents, large-model inference and hybrid agentic AI deployment rather than only as a client OS. According to NVIDIA, the agentic AI moment “requires more than good models” and depends on fast hardware, secure runtimes and a responsive data layer that now span client, local and cloud environments.
RTX Spark and DGX Station: Turning Windows PCs into Agent Platforms
At the device level, RTX Spark brings agentic AI deployment to everyday Windows laptops and small desktops. NVIDIA describes RTX Spark systems as the first Windows PCs purpose-built for personal agents, delivering 1 petaflop of AI performance, up to 128GB of unified memory and all‑day battery life while keeping full AI and graphics performance unplugged. Systems from major PC makers, including Microsoft Surface, are expected later this year, alongside a Surface RTX Spark Dev Box tuned for local agent workloads. Higher up the stack, DGX Station for Windows extends the same idea to deskside enterprise AI infrastructure. Powered by the NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip, it offers up to 748GB of coherent memory and 20 petaflops of FP4 performance to run frontier models of up to 1 trillion parameters directly on Windows enterprise environments.
OpenShell and Execution Containers: Securing Long‑Running AI Agents
Agentic AI deployment depends on agents that can act autonomously over long periods without exposing data or systems. NVIDIA OpenShell is the secure runtime layer in the unified stack, designed specifically for autonomous agents running across RTX Spark and DGX Station for Windows. Microsoft is bringing OpenShell to Windows on top of Microsoft Execution Containers, a policy-driven execution environment that controls what an agent can access at runtime. That combination means developers can define the permissions and data boundaries for each agent, then apply those policies consistently from local PCs to on‑premises servers and Azure. OpenShell is also appearing inside GitHub Copilot and Foundry Agent Service, which gives enterprises a single approach to governance for hosted and local agents. This security focus is less visible than new hardware, but it is central to making Windows a reliable endpoint for serious enterprise AI workloads.
Foundry, Nemotron and Fabric: Linking Models and Data Across Cloud and Local
On the cloud side, the NVIDIA Microsoft partnership centers on Microsoft Foundry and Fabric as the enterprise AI infrastructure layer. Foundry now hosts NVIDIA’s open models, Anthropic and OpenAI models, plus Hermes special agents, so enterprises can compose systems of cooperating agents on Azure with identity and governance built in. NVIDIA Nemotron 3 Ultra, a new open frontier reasoning model for long‑running agents, is available on Foundry managed compute alongside Nemotron 3.5 ASR and Nemotron 3.5 Content Safety. In parallel, Microsoft Fabric Data Warehouse now includes NVIDIA accelerated computing, with Microsoft’s internal benchmarks showing SQL execution up to 6x faster than a CPU-only baseline and up to 7x faster than three unnamed leading cloud data warehouse providers for high‑concurrency workloads. This makes it easier for agents to query live enterprise data instead of relying on static snapshots.
Why This Redefines the Windows AI Stack for Enterprises
For enterprises, the unified stack reframes Windows PCs from being the edge of the network to being first‑class AI endpoints in a broader agentic architecture. RTX Spark laptops can run personal agents, DGX Station for Windows can host always‑on enterprise agents next to line‑of‑business apps, Azure Local can keep sensitive workloads on‑premise, and Foundry plus Fabric extend those same agents into the cloud. NVIDIA’s CUDA‑X libraries, Agent Toolkit and NemoClaw blueprints give developers a consistent way to build agents that span coding, research, physical AI and data-heavy workflows. At the same time, Microsoft’s integration of NVIDIA’s physical AI tools with Azure’s Physical AI Toolchain connects simulation, robotics and autonomous systems to this shared platform. This unified NVIDIA Microsoft partnership signals a Windows AI stack where agentic AI deployment is no longer fragmented by device class or location.
