Defining the Unified Agentic AI Stack
NVIDIA and Microsoft’s unified stack for agentic AI deployment is a combined hardware, software and data platform that lets developers design, run and manage autonomous AI agents consistently across Windows PCs, local infrastructure and cloud services, so the same agentic application can span personal devices, enterprise data centers and hyperscale platforms without being rewritten for each environment. This stack is the focus of their expanded NVIDIA Microsoft partnership announced at Microsoft Build, where Jensen Huang joined Satya Nadella’s keynote to highlight a shared Windows AI stack. The strategy goes beyond powerful laptops: Microsoft aims to treat Windows as a managed endpoint for local agents, large-model inference and hybrid AI infrastructure that connects cleanly to Azure Local, Microsoft Foundry and Fabric. In practice, it links RTX Spark devices, DGX Station for Windows and GPU-accelerated data services into one edge to cloud AI continuum.
RTX Spark and DGX Station: Reinventing Windows for Agents
On the client side, RTX Spark systems are Windows PCs purpose-built for personal agents, delivering 1 petaflop of AI performance and up to 128 GB of unified memory for on-device reasoning. NVIDIA says these systems will arrive from vendors including Microsoft Surface, ASUS, Dell, HP, Lenovo and MSI in the fall, with a Surface RTX Spark Dev Box tuned for local model and agent workloads. At the enterprise desk, DGX Station for Windows moves the same concept to a deskside AI supercomputer powered by the NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip. According to NVIDIA, DGX Station for Windows offers up to 748 GB of coherent memory and 20 petaflops of FP4 performance, supporting AI models with up to 1 trillion parameters while still fitting into Windows enterprise management and Linux AI toolchains via Windows Subsystem for Linux.
OpenShell and Secure Agentic AI Deployment Across Environments
Security and control are central to running autonomous agents at scale, so the unified Windows AI stack includes NVIDIA OpenShell, a secure-by-design runtime for agentic AI deployment. OpenShell is coming to Windows on top of Microsoft Execution Containers, a policy-based execution layer that governs what an agent can access at runtime. This makes Windows more than an operating system; it becomes part of a managed security perimeter for agents that may act over long periods, touch sensitive data or call external tools. The same OpenShell runtime also underpins RTX Spark and DGX Station for Windows, aligning local and enterprise environments with cloud services like GitHub Copilot and Microsoft Foundry. With one runtime spanning edge to cloud AI, developers can test agents locally, then move them into hosted or on-premises deployments without redesigning security or governance logic for each target platform.
Foundry, Nemotron and Fabric: Cloud and Data Foundations for Agentic AI
In the cloud, Microsoft Foundry and Azure provide the model and orchestration layer for enterprise agentic systems built on the NVIDIA Microsoft partnership. Hosted agents in Foundry Agent Service now include NVIDIA models, Anthropic’s Claude family and OpenAI models, plus Hermes special agents, so teams can compose multi-model systems with built-in identity and governance. NVIDIA’s Nemotron 3 Ultra reasoning model, Nemotron 3.5 ASR and Nemotron 3.5 Content Safety are available on Foundry managed compute, while CUDA-X libraries like cuDF, cuOpt and NeMo become domain-specific skills that agents can call through NVIDIA Agent Toolkit and NemoClaw blueprints. On the data side, Microsoft Fabric Data Warehouse now integrates NVIDIA accelerated computing. Microsoft reports SQL execution up to 6x faster than a CPU baseline and up to 7x faster than three other leading cloud data warehouses for high-concurrency workloads, helping agentic workflows keep pace with continuous querying.
From Edge to Cloud AI: What This Means for Enterprise Builders
The expanded NVIDIA Microsoft partnership effectively turns Windows devices, on-premises servers and Azure into a single, continuous platform for agentic AI deployment. RTX Spark PCs handle personal agents and offline workloads, DGX Station for Windows supports large frontier models at the deskside, and Azure Local plus Foundry Local extend the same stack into data centers and edge sites. In parallel, Microsoft Fabric and Microsoft Planetary Computer Pro bring GPU acceleration to data and physical AI workloads, including NVIDIA Cosmos 3 and Earth-2 models for simulation, forecasting and autonomous systems. For enterprises, the shift is less about hardware launches and more about a consistent development and deployment story: one Windows AI stack, one security and runtime model, and one edge to cloud AI fabric. That coherence is what makes building long-running, autonomous agents across devices and environments feasible instead of experimental.
