No. Windows is joining Linux on the DGX Station, not replacing it. Every NVIDIA DGX Station GB300 sold today, including the HP ZGX Fury G1n, MSI XpertStation WS300 and ASUS ExpertCenter Pro ET900N G3 (all price on request at SourceIT), ships with Ubuntu 24.04 and NVIDIA's AI developer tools. On 31 May 2026 NVIDIA and Microsoft announced "DGX Station for Windows", with OEM availability expected in Q4 2026. Its Linux tools run through Windows Subsystem for Linux (WSL). For model training, serving and containers, Linux remains the native environment. Windows mainly adds desktop management, Windows apps and a sandbox for AI agents. As of 4 Nov 2026, we have not found a Windows version of any DGX Station on sale.
What exactly did NVIDIA and Microsoft announce?
The press release is datelined "NVIDIA GTC Taipei, May 31, 2026" and headlined "NVIDIA DGX Station for Windows Puts a Trillion-Parameter AI Supercomputer on Every Enterprise Desk". It says the product was developed with Microsoft and quotes Pavan Davuluri, Microsoft's EVP for Windows and Devices. The hardware is unchanged: one GB300 superchip with up to 748GB of coherent memory, up to 20 petaFLOPS of FP4 compute and a ConnectX-8 network card. The software side has three parts:
- WSL for Linux workloads. In NVIDIA's words, "Linux workloads get the same level of manageability support through Windows Subsystem for Linux."
- NVIDIA OpenShell for agents. An open-source agent runtime that gives each agent its own isolated sandbox, built on new Windows containment features.
- Windows fleet management. Agents and the machine are managed with the security, compliance and device-management tools enterprises already use for Windows.
Some things were not said. Neither the press release nor NVIDIA's product page names a Windows edition; the page lists only "Microsoft Windows". Neither mentions Copilot. The Grace CPU is Arm, so in practice this is Windows on Arm. An NVIDIA job listing for the project referred to "Windows on Arm, WHQL, driver signing". Treat edition and feature details as unconfirmed until OEMs publish their Windows SKUs.
How do the two environments compare today?
| Ubuntu with NVIDIA AI Developer Tools | DGX Station for Windows | |
|---|---|---|
| Status (4 Nov 2026) | Shipping on all five OEM models | Announced; OEM availability expected Q4 2026; HP Japan says December 2026 or later for the ZGX Fury |
| Base | Ubuntu 24.04, kernel 6.17, NVIDIA open driver, CUDA 13.x | Windows edition not named by NVIDIA |
| Pre-installed AI stack | vLLM, SGLang, TensorRT-LLM, RAPIDS, PyTorch, NCCL, DCGM, Container Toolkit | Linux toolchains via WSL; CUDA on Windows on Arm added in CUDA 13.4 (Sep 2026) |
| Containers and NGC | Native multi-arch NGC containers | Through WSL; GPU access inside WSL on Arm not yet documented |
| Agent sandboxing | NVIDIA OpenShell playbook on Linux | OpenShell on Windows containment primitives (early preview) |
| Fleet management | Standard Linux tooling, BMC/Redfish, Ubuntu Pro for CIS hardening | Windows enterprise management (Intune and similar) |
| Desktop apps | Linux desktop; GUI needs an RTX PRO display card | Windows apps (native Arm64 or emulated x64); still needs a display card |
Sources: NVIDIA newsroom (31 May 2026), NVIDIA DGX Station for Windows page and DGX Station Development Guide (accessed 4 Nov 2026), CUDA 13.4 announcement (9 Sep 2026), HP Japan newsroom (26 Oct 2026), Microsoft Windows developer blog (2 Jun 2026). Availability dates are vendor statements, not confirmed shipments.
Why is Linux still the default for AI work?
Because almost everything that does the heavy lifting is built for it first. NVIDIA's own DGX Station software stack is Ubuntu, and its step-by-step playbooks for vLLM, SGLang, MIG partitioning, two-node clustering and OpenShell are written for Linux. The NGC framework containers for PyTorch and JAX publish Arm builds for Linux. Kubernetes and Run:ai, which appear in NVIDIA's preliminary DGX Station software diagram for team scheduling, also run on Linux. On a Windows DGX Station those tools still run, just inside WSL. That is a Linux layer hosted by Windows, not a Windows replacement for Linux.
There are also practical gaps to watch. CUDA 13.4 added Windows on Arm support, but the release notes leave out CUDA-GDB and cuFile. We found no NVIDIA document confirming GPU passthrough into WSL on Arm, although the DGX Station for Windows design depends on it. Expect early Windows images to need careful testing.
Linux is not losing its crown on DGX Station. It is moving inside Windows, while Windows takes over the desk, the fleet tools and the agents.
So what does Windows actually add?
Three things matter to businesses. First, manageability: a DGX Station that enrols in the same device-management and compliance tools as your laptops is easier to govern than a one-off Linux box. Second, Windows applications: staff who need Office, line-of-business tools or a Windows-only client can use the station directly instead of through a second PC. Third, agents. NVIDIA's pitch is "hundreds of agents running simultaneously" next to Windows apps, each sandboxed by OpenShell, with policies enforced by the system rather than by the prompt. Whether you want agents working in Windows apps at all is a governance question; see Would You Trust a Trillion-Parameter AI Agent With Your Windows Desktop?.
The trade-off is compatibility. Windows on Arm runs x64 apps through Microsoft's Prism emulation, but only user-mode code; drivers must be native Arm64. Our DGX Station for Windows compatibility guide covers what that means for VPN clients, security agents and engineering software. For the same debate on smaller machines, see NVIDIA RTX Spark vs GB10: Is Windows Finally Ready to Challenge Linux for Local AI?.
Which should your team choose?
- Stay on Ubuntu if the station is a shared compute node for data scientists and engineers, if you plan to use MIG, Kubernetes or two-node clustering, or if you need it working now. This covers most purchases today.
- Plan for Windows if the station will sit on one power user's desk, if IT must manage it with existing Windows tools, or if agents need to work alongside Windows apps. Ask your OEM for the Windows SKU, its release date and whether an existing Ubuntu unit can be converted.
- Either way, budget for an RTX PRO display card if anyone will use the station as a desktop. The GB300 has no display output of its own, and adding a card lowers the GPU's power allowance (see Why Does a S$150,000 DGX Station Need Another Graphics Card?).
The underlying Arm software questions apply to both operating systems; see Your DGX Station Has 72 CPU Cores. Why Might Your Software Still Struggle?.
Buying in Singapore
SourceIT lists all five DGX Station GB300 models, each price on request: the HP ZGX Fury G1n, MSI XpertStation WS300, ASUS ExpertCenter Pro ET900N G3, GIGABYTE W775-V10-L01 and Supermicro ARS-511GD-NB-LCC. If you want Windows, tell us when you request the quotation so we can confirm each OEM's Windows SKU and timing. Browse the NVIDIA DGX Spark and DGX Station collection or read our DGX Station GB300 buyer's guide. For a quotation, use Request for Quotation on the product page, call (65) 6978 3502, email Sales@sourceit.com.sg or use our contact page.
Frequently asked questions
Can I buy a DGX Station with Windows today? Not that we have found. NVIDIA announced DGX Station for Windows on 31 May 2026 with OEM availability expected in Q4 2026. HP Japan says ZGX Fury Windows support is planned for December 2026 or later. Current units ship with Ubuntu.
Does DGX Station for Windows still run Linux tools? Yes, through Windows Subsystem for Linux. NVIDIA says Linux workloads get the same manageability support through WSL.
Which Windows edition will it use? NVIDIA has not named one; its product page says only "Microsoft Windows". Because the Grace CPU is Arm, it will be a Windows on Arm build. Confirm the edition with the OEM before ordering.
Is Copilot part of DGX Station for Windows? Neither NVIDIA's nor Microsoft's announcement mentions Copilot. The agent features described are NVIDIA OpenShell sandboxes managed with Windows security and fleet tools.
Is Linux or Windows faster on DGX Station? No published benchmark compares them yet. The GPU work is the same CUDA hardware either way; differences will come from drivers, WSL overhead and software maturity, so test your own workload once Windows images ship.
