It enables frontline workloads like model training, fine-tuning, and large-scale inference to be executed locally, reducing reliance on Linux-based cloud data centers.

NVIDIA unveils the DGX Station for Windows, a deskside AI supercomputer that lets users build, run, and connect always-on AI agents directly within Windows applications and workflows, capable of running frontier models up to a trillion parameters locally.

The platform integrates Windows security, containment primitives, and Windows Subsystem for Linux to support Linux workloads, offering enterprise manageability and fleet-wide capabilities through familiar Microsoft tools.

Availability is planned through major OEMs and partners—ASUS, Dell Technologies, GIGABYTE, HP, MSI, and Supermicro—in the fourth quarter of 2026.

The DGX Station lets frontier models with a trillion parameters run locally on standard workstations, blending datacenter-class performance with desktop components.

The DGX Station can function as a single-user workstation or a shared local resource for teams, enabling secure, local AI agent development and testing.

NVIDIA aims to bring data center–grade AI infrastructure to Windows environments, enabling local development and deployment of AI agents without mandatory cloud workloads.

The product will launch by year’s end through multiple partner systems—Dell, HP, ASUSTeK, MSI, and Micro-Star International—offering a range of configurations.

This launch aligns with NVIDIA’s broader strategy to extend its AI ecosystem from data centers to desktop and edge environments, supporting data science, high-throughput inference, and physical AI workflows.

The DGX Station is designed to integrate with enterprise IT manageability, security, and fleet management via Windows administration tools, preserving enterprise governance.

NVIDIA ConnectX-8 SuperNIC enables networking up to 800 Gb/s, allowing fast data transfers and enabling multi-DGX Station scalability for larger workloads.

Networking multiple DGX Stations with ConnectX-8 SuperNIC facilitates clustering of desktop supercomputers for even bigger workloads.