---

NVIDIA and Microsoft have used Microsoft Build to push Windows further into the centre of local and cloud-based agentic AI development, joining RTX Spark PCs, DGX Station for Windows, Azure Local, Microsoft Foundry and GitHub Copilot in a single accelerated computing stack.

The companies said the collaboration is intended to let developers build, run and scale agents across Windows devices, Azure cloud services and local deployments. For hardware developers and enterprise engineering teams, the relevant shift is not just faster laptops; it is an attempt to make Windows a managed endpoint for local agents, large-model inference and hybrid AI infrastructure.

NVIDIA Microsoft stack spans PC to cloud

At the client end, RTX Spark systems are aimed at running personal agents directly on Windows. NVIDIA says the platform offers 1 petaflop of AI performance and up to 128 GB of unified memory, with systems expected from Microsoft Surface, ASUS, Dell, HP, Lenovo and MSI in the US autumn. Microsoft has also introduced the Surface RTX Spark Dev Box for developers, with 128 GB of unified memory and a 100 W thermal envelope for local model and agent workloads.

At the higher end, DGX Station for Windows moves the same argument to enterprise deskside systems. The machine is based on the NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip, with up to 748 GB of coherent memory and up to 20 petaflops of FP4 performance. NVIDIA says it can run AI models of up to 1 trillion parameters locally, while also supporting Windows enterprise management and Linux AI toolchains through Windows Subsystem for Linux.

That positioning follows NVIDIA’s earlier desktop AI push, as previously reported by eeNews Europe when Nvidia launched a desktop AI computer. The difference this time is the closer Microsoft integration: Windows is not just the operating system on the box, but part of the security, management and developer runtime story.

OpenShell and local agents

The NVIDIA Microsoft stack also includes OpenShell, NVIDIA’s secure runtime for autonomous agents. Microsoft says NVIDIA is bringing OpenShell to Windows on top of Microsoft Execution Containers, a policy-driven execution layer for controlling what an agent can access at runtime. That is the less glamorous part of the announcement, but arguably the more important one: agentic AI without containment is a liability generator with a nice demo.

NVIDIA also said its open models, including Nemotron 3 Ultra and Cosmos 3, are being added to Microsoft Foundry, while accelerated computing is being built into Microsoft Fabric Data Warehouse. Microsoft internal benchmarks cited by NVIDIA put GPU-accelerated Fabric SQL execution at up to 6 times faster than a CPU-powered baseline.

Local, hybrid and sovereign AI

Beyond PCs and workstations, Microsoft is bringing Foundry Local on Azure Local to NVIDIA RTX PRO 6000 Blackwell Server Edition systems. That is aimed at manufacturers, energy companies, government users and other organisations that want AI workloads close to their data rather than fully dependent on remote cloud infrastructure.

The broader direction is clear enough: NVIDIA wants its AI hardware and software stack to follow developers from laptop to workstation to cloud, while Microsoft wants Windows, Azure and Foundry to remain the control plane for those workloads. Whether the market wants always-on local agents is still an open question, but the two companies are now giving developers a more coherent path to test the idea on real hardware.

---

---