NVIDIA Unveils DGX Station for Windows Desktop AI Supercomputer

At an event in San Francisco on October 7, NVIDIA announced the DGX Station for Windows, a desktop AI supercomputer designed to let developers, researchers and data science teams run large AI models locally. The system is scheduled to launch in the fourth quarter of 2026.

The DGX Station is powered by the NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip, combining a 72-core Grace CPU with a Blackwell Ultra GPU. It features up to 748GB of unified memory, including 252GB of HBM3e GPU memory and 496GB of LPDDR5X CPU memory, and delivers up to 20 PFLOPS of AI performance, supporting local execution of models with roughly 1 trillion parameters.

Compared with traditional high-performance workstations, the DGX Station packs datacenter-class hardware, unified memory and NVIDIA’s AI software stack into a desktop form factor. Developers can use it for model fine-tuning, inference, agent development and generative AI testing, avoiding the latency, cost and privacy concerns associated with uploading data to the cloud.

NVIDIA has also tied the system to the Windows ecosystem, allowing developers to use local large-model capabilities and explore AI agents, Copilot and hybrid intelligent applications in a familiar environment. The device has a maximum power draw of 1,600W and requires a 20A circuit.