NVIDIA Unveils DGX Station for Windows, Trillion-Parameter AI on Demand

NVIDIA has announced the DGX Station™ for Windows, a deskside AI supercomputer capable of running AI models with up to 1 trillion parameters. The system, powered by the cutting-edge NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip, is designed to integrate advanced AI capabilities directly into Windows-based enterprise workflows. The product will be available in Q4 2026 through partners like ASUS, Dell, and HP.

This development marks a significant shift in AI accessibility, integrating robust AI infrastructure into the widely adopted Windows ecosystem. Historically, enterprise AI workloads—such as training and inference of large models—have been confined to Linux-based data center deployments. By bringing these capabilities to Windows, NVIDIA aims to streamline AI integration for Fortune 500 companies that rely heavily on Windows for day-to-day operations.

“As enterprises scale AI agents across their organizations, they need infrastructure that connects directly to their existing workflows,” said NVIDIA’s Chris Marriott, VP of Enterprise Platforms. “DGX Station delivers supercomputing-class AI directly into Windows, enabling millions of developers, researchers, and designers to build and deploy AI faster.”

Technical Highlights

The DGX Station for Windows is built around the GB300 Grace Blackwell Superchip, which combines a 72-core NVIDIA Grace™ CPU with a Blackwell Ultra GPU via NVLink™-C2C interconnect. The system offers up to 748GB of coherent memory and supports 20 petaflops of FP4 performance. For enhanced visualization and simulation, it can be paired with an NVIDIA RTX PRO™ 6000 Blackwell Workstation GPU.

Networking capabilities are driven by the NVIDIA ConnectX®-8 SuperNIC™, which supports speeds of up to 800Gb/s. This ensures rapid data transfer for AI workloads and facilitates the connection of multiple DGX Stations for larger-scale tasks.

Enterprise AI on the Desktop

The DGX Station for Windows is designed to serve a variety of enterprise workflows, including:

• AI Agents: Build and run multiple frontier AI agents in parallel, connecting them directly to enterprise applications.
• AI Development: Pretrain and fine-tune large AI models locally, with access to Linux toolchains via Windows Subsystem for Linux.
• Data Science: Eliminate data movement bottlenecks with 748GB of memory for data preparation and machine learning tasks.
• AI Inference: Run real-time inference on trillion-parameter models.
• Physical AI: Use ray-traced visualization and simulation to bridge physical and virtual environments.

To enhance security, NVIDIA also introduced OpenShell™, an open-source runtime that isolates AI agents into secure sandboxes, ensuring enterprise-grade privacy and compliance. This is particularly relevant for industries handling sensitive data.

Market Context

NVIDIA’s move to democratize access to trillion-parameter AI aligns with its broader strategy to distribute AI capabilities beyond centralized data centers. The DGX Station for Windows reflects an effort to address the needs of enterprises looking to deploy AI locally without relying on cloud infrastructure. This is especially critical as AI usage evolves toward real-time, always-on agents embedded in enterprise workflows.

The launch follows the March 2026 debut of the GB300-powered DGX Station at GTC 2026. Priced around $97,000 per unit, the system targets high-end enterprise users and research institutions. As of May 30, NVIDIA shares (NASDAQ: NVDA) traded at $211.14, giving the company a $5.15 trillion market cap, underscoring investor confidence in its dominant position in AI hardware and software.

Availability

The DGX Station for Windows will be available in Q4 2026 through major OEMs, including ASUS, Dell, HP, and MSI. Enterprises looking to integrate local AI supercomputing into their workflows can expect a seamless transition, leveraging NVIDIA’s AI software stack and Microsoft’s enterprise management tools.

For more details, watch NVIDIA CEO Jensen Huang’s keynote at NVIDIA GTC Taipei.

Image source: Shutterstock