NVIDIA DGX Station Brings Trillion-Parameter AI to Enterprise Desktops

At NVIDIA GTC Taipei, the company unveiled DGX Station for Windows, described as the most powerful deskside AI supercomputer ever built for enterprise use. Designed to run always-on AI agents and handle models with up to 1 trillion parameters locally, this system marks a significant shift in how Windows-based organizations can access supercomputing-class AI.

For years, the heaviest enterprise AI workloads,training, fine-tuning, large-scale inference, and multi-agent development,have been confined to Linux-based data center infrastructure. Meanwhile, the vast majority of Fortune 500 companies rely on Windows for daily productivity, creative design, and engineering tasks. DGX Station for Windows bridges that divide by bringing NVIDIA’s GB300 Grace Blackwell Ultra Desktop Superchip directly into the Windows ecosystem for the first time.

“As enterprises scale AI agents across their organizations, they need AI infrastructure that can connect directly to the applications and workflows that power their business,” said Chris Marriott, vice president of enterprise platforms at NVIDIA. “DGX Station delivers supercomputing-class AI directly into Windows, where millions already design, engineer, research and create every day.”

Microsoft’s Pavan Davuluri, executive vice president of Windows + Devices, echoed that sentiment. “For decades, Microsoft and NVIDIA have partnered to advance the most powerful computing platforms in the world. Today, we’re taking that collaboration to the next level, scaling the full power of Windows from thin-and-light PCs to data-center-class workstations with DGX Station powered by GB300. This unlocks a new class of AI performance on Windows, the platform enterprises trust for security, manageability and compatibility.”

The hardware specifications are impressive. The DGX Station is built around the NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip, which pairs a powerful Blackwell Ultra GPU with a high-performance 72-core NVIDIA Grace CPU using the NVLink-C2C interconnect for optimal system communication. It offers up to 748GB of coherent memory and delivers up to 20 petaflops of FP4 performance. Users can also pair it with an NVIDIA RTX PRO 6000 Blackwell Workstation GPU to combine frontier AI compute with ray-traced visualization and simulation.

Networking is handled by the NVIDIA ConnectX-8 SuperNIC, which supports speeds up to 800Gb/s. This enables rapid data transfers for AI workloads and allows multiple DGX Station systems to be linked together for even larger tasks.

Enterprise AI is evolving from simple chatbot interactions to agentic inference,systems that operate continuously, reason in real time, and connect directly to enterprise applications. Developed in collaboration with Microsoft, DGX Station for Windows serves as dedicated agent infrastructure, allowing organizations to build and run agents with frontier intelligence locally. It supports models of up to 1 trillion parameters and can run hundreds of agents simultaneously.

For enterprise IT teams, the system extends the same Windows security, compliance, and fleet management tools they already use. AI agents operate within this managed environment, governed through familiar Microsoft tools. Linux workloads are also supported via Windows Subsystem for Linux, with enterprise-grade features like deployment and system updates.

Security is a major focus with the introduction of NVIDIA OpenShell, an open-source, secure-by-design runtime for autonomous agents. It creates an isolated sandbox for each agent, separating application-layer operations from infrastructure-layer policy enforcement. This means security and privacy policies are applied at the system level, out of the agent’s reach, preventing unauthorized overrides or data leaks.

The DGX Station is designed for the full spectrum of enterprise AI workflows. These include AI agent deployment and development, data science with large datasets in coherent memory, high-throughput inference on models up to 1 trillion parameters, and physical AI when paired with an additional RTX PRO Blackwell GPU. The system can serve as a dedicated supercomputer for a single developer or as a shared local compute node for entire teams, with workloads scaling seamlessly to GB300 in the data center or cloud.

Availability is expected in Q4 of this year through partners including ASUS, Dell Technologies, GIGABYTE, HP, MSI, and Supermicro.