Defining a Unified Agentic AI Stack from Edge to Cloud
Agentic AI deployment describes building and running autonomous software agents that can perceive context, reason over long periods, and perform tasks across devices, local servers and cloud platforms through a single, consistent technology stack that spans models, runtimes, hardware and data infrastructure. NVIDIA and Microsoft are turning this idea into a unified platform that stretches from gaming-class RTX Spark PCs to DGX Station for Windows, Azure Local and Microsoft Fabric. Instead of treating Windows as a simple endpoint, the partnership treats it as a managed agent platform for large-model inference and hybrid AI infrastructure. Developers can now target the same agentic AI stack for Windows AI agents on laptops, deskside supercomputers, local edge AI computing and Azure-hosted services. This reduces fragmentation between development environments and production deployments, and it prepares Windows to host both personal assistants and enterprise-grade, always-on AI agents.
RTX Spark PCs and DGX Station: Windows as an Agent Platform
On the client side, RTX Spark PCs are the first Windows systems built specifically for personal agents, combining gaming-class graphics with dedicated agentic AI deployment support. NVIDIA says RTX Spark delivers 1 petaflop of AI performance and up to 128 GB of unified memory in laptops and small desktops from Surface, ASUS, Dell, HP, Lenovo and MSI. For heavier Windows AI agents, DGX Station for Windows pushes the same model to enterprise deskside machines. Powered by the GB300 Grace Blackwell Ultra Desktop Superchip, it provides up to 748 GB of coherent memory and 20 petaflops of FP4 performance to run models with up to 1 trillion parameters locally. These systems bring CUDA, RTX, DLSS and TensorRT into a Windows environment, so developers can build, tune and run agents on the same platform used by gamers and enterprise users.
OpenShell, Security and Local‑First Windows AI Agents
Security and control are central to agentic AI. NVIDIA OpenShell is the secure runtime at the heart of this stack, providing policy-driven control over what Windows AI agents can access at runtime. Microsoft is integrating OpenShell on top of Microsoft Execution Containers so enterprises can set precise permissions for files, networks and tools. This gives hardware makers and engineering teams a path to treat Windows as a managed endpoint for autonomous agents and large-model inference. Local-first AI ideas, such as the emerging class of Blue AI Worker-style agents, benefit from this design. They can run on RTX Spark laptops or DGX Station for Windows, reducing cloud calls, latency and bandwidth costs by using onboard GPU compute. By aligning Windows management, security and AI runtimes, the NVIDIA Microsoft partnership aims to make local edge AI computing a first-class option rather than a fall-back when the network is unavailable.
From Azure Local to Microsoft Fabric: Data and Hybrid Agentic AI
Beyond devices, the same NVIDIA Microsoft partnership extends agentic AI deployment into Azure Local, Microsoft Foundry and Microsoft Fabric. On Azure, Anthropic’s Claude models now run natively on NVIDIA GB300 Blackwell Ultra systems, and NVIDIA Nemotron 3 Ultra is offered as an open reasoning model tuned for long-running agents across coding, research and enterprise workflows. According to NVIDIA, CUDA-X libraries such as cuDF, cuOpt, AI-Q and NeMo are exposed as domain-specific skills to these agents through the NVIDIA Agent Toolkit and NemoClaw blueprints. In Microsoft Fabric Data Warehouse, NVIDIA accelerated computing has delivered SQL execution up to 6x faster than a CPU-only baseline for high-concurrency workloads, allowing data platforms to keep pace with agents that continuously query and reason over live information. This shared stack allows developers to compose hybrid agents that split work between local RTX systems and Azure services.
Beyond Planning: Agentic AI for Gaming, Enterprise and Physical Systems
Agentic AI in this stack moves beyond simple planning to long-running, autonomous task execution across gaming, enterprise and physical AI. On Windows, gamers and creators gain personal agents that can co-exist with high-performance graphics on RTX Spark PCs, while enterprises can run always-on agents tied into Windows applications on DGX Station for Windows. In the cloud, NVIDIA’s open model portfolio on Microsoft Foundry now spans agentic, physical and scientific AI, including Cosmos 3 for world simulation and action generation, and Earth-2 weather models exposed through Microsoft Planetary Computer Pro and Foundry. These models support agents that can perceive, reason and act in simulated or real environments, from industrial systems to autonomous machines. By aligning edge devices, local infrastructure and Azure in a single agentic AI deployment story, NVIDIA and Microsoft are laying the groundwork for agents that follow users and workloads wherever they run.
