What a Unified Agentic AI Stack Means
A unified agentic AI stack is an integrated set of hardware, models, runtimes and data tools that lets developers build, deploy and manage autonomous AI agents consistently across personal devices, edge servers and cloud infrastructure. NVIDIA and Microsoft are joining their platforms to create such a stack for Windows, Azure and local deployments. Announced at Microsoft Build during a keynote with Satya Nadella and Jensen Huang, the collaboration ties together RTX Spark PCs, DGX Station for Windows, Microsoft Fabric, Azure Local, Microsoft Foundry and GitHub Copilot into one Windows AI stack. The goal is to treat every Windows endpoint as a managed home for edge AI agents, long-running reasoning and hybrid inference. Instead of treating AI as a cloud-only feature, the NVIDIA Microsoft partnership pushes agentic AI deployment down to the PC while aligning it with the same tools and security model used in the data center.
Reinventing Windows Devices for Agentic AI
On the client side, NVIDIA RTX Spark systems mark a shift from traditional PCs toward RTX AI development machines built for personal agents. According to NVIDIA, “RTX Spark is a new beginning, powering the world’s first Windows PCs purpose-built for personal agents, with 1 petaflop of AI performance and up to 128GB of unified memory.” OEM systems from Microsoft Surface, ASUS, Dell, HP, Lenovo and MSI are expected later this year, alongside a Surface RTX Spark Dev Box tuned for local model and agent workloads. For enterprise desksides, DGX Station for Windows extends the same idea with the GB300 Grace Blackwell Ultra Desktop Superchip, up to 748GB of coherent memory and up to 20 petaflops of FP4 performance, allowing frontier-scale models and always-on agents to run locally while remaining under standard Windows management.
OpenShell and Secure Edge AI Agents
To make edge AI agents practical on Windows, the stack includes NVIDIA OpenShell, a secure-by-design runtime for autonomous agents. On client and deskside systems, OpenShell runs on top of Microsoft Execution Containers, giving administrators policy-based control over what each agent can access at runtime. This moves Windows from being just the operating system on the box to being part of the security and governance fabric for agentic AI deployment. With OpenShell integrated into GitHub Copilot, developers can experiment with autonomous workflows while targeting the same constrained runtime used in production. For hardware and IT teams, the message is clear: a Windows PC can now be a managed endpoint for local agents, large-model inference and hybrid workflows that span PC, edge and cloud. That makes edge AI agents a first-class part of enterprise architecture instead of isolated experiments.
Foundry, Nemotron and Fabric: Cloud-Scale Agentic AI
In the cloud, Microsoft Foundry and Azure Local extend the same Windows AI stack to large-scale agentic systems. Enterprises can compose agents using NVIDIA, Anthropic and OpenAI models, plus Hermes special agents, all available through the Foundry Agent Service with built-in identity and governance. NVIDIA’s Nemotron 3 Ultra reasoning model, Nemotron 3.5 ASR and Nemotron 3.5 Content Safety add long-running reasoning, speech and safety capabilities, while CUDA-X libraries such as cuDF, cuOpt, AI-Q and NeMo are exposed as skills through the NVIDIA Agent Toolkit and NemoClaw blueprints. Meanwhile, Microsoft Fabric Data Warehouse now runs on NVIDIA accelerated computing, delivering up to 6x faster SQL execution than a CPU baseline for high-concurrency workloads. This faster data layer lets agents continuously query, plan and act over enterprise data without waiting on slow analytics pipelines, aligning cloud performance with on-device RTX AI development.
From PC Computing to Agent-Based Workflows
The combined Windows AI stack signals a broader shift from app-centric PCs toward agent-based workflows spanning devices, edge and cloud. Developers can design an agent once, then run it on an RTX Spark laptop, scale it on a DGX Station for Windows, and extend it into Azure via Foundry or Azure Local, all while keeping a consistent toolchain. Physical AI via Cosmos 3 and Microsoft’s Physical AI Toolchain shows how the same stack can drive robots, autonomous vehicles and industrial systems that perceive, reason and act in the physical world. Data pipelines accelerated in Fabric make it realistic for agents to stay always-on and data-aware. For enterprises, this NVIDIA Microsoft partnership is less about a single new device and more about a Windows AI stack that treats every endpoint as part of a distributed, agentic AI deployment platform.
