NVIDIA has announced the new DGX Station for Windows, a deskside supercomputer designed to run artificial intelligence agents locally.
Historically, heavy-duty enterprise workloads required powerful systems in data centers running on Linux. However, the majority of Fortune 500 companies use Windows for everyday productivity, design, and engineering applications. This new hardware brings data-center class infrastructure directly to these users.
Hardware Specifications of DGX Station
The hardware configuration includes up to 748GB of coherent memory. This memory consists of 252GB of high-speed HBM3e GPU memory and 496GB of LPDDR5X system memory. Meanwhile, the system delivers up to 20 petaflops of FP4 performance.
Users can pair the system with an NVIDIA RTX PRO 6000 Blackwell Workstation GPU. This setup enables ray-traced visualization and simulation alongside frontier computing. The architecture connects the GPU to a 72-core NVIDIA Grace CPU using the NVLink-C2C interconnect.
For high-speed networking, the system features the NVIDIA ConnectX-8 SuperNIC. This component supports data transfer speeds of up to 800Gb/s. Consequently, it enables fast data transfers for complex workloads and allows users to connect multiple systems together.
Artificial Intelligence Capabilities
This deskside system can run local artificial intelligence models with up to 1 trillion parameters. Consequently, developers can execute hundreds of autonomous agents simultaneously on the DGX Station without data center reliance.
This capability supports tasks ranging from pretraining to high-throughput inference.
Security and Enterprise Integration
The system integrates directly into the Windows platform to support existing enterprise applications. Specifically, it supports NVIDIA OpenShell on Windows, which is an open-source runtime for autonomous agents. This runtime creates an isolated sandbox for each agent, which separates application-layer operations from system-level security policies.
Market Availability and Pricing
Several manufacturing partners will deliver systems based on this architecture. Specifically, ASUS, Dell Technologies, GIGABYTE, HP, MSI, and Supermicro will release their versions. Among these partners, MSI has developed the XpertStation WS300 based on the DGX Station design.
While official pricing from NVIDIA is not published, partner systems have established market rates. The MSI system is priced at $84,999.99, while other high-end configurations are expected to reach up to $125,000. These systems will become available starting in the final quarter of 2026.
Source: X (@TheGalox_) — researched




