Nvidia DGX Station Brings Data Center AI Power to the Desktop

Nvidia unveiled the DGX Station at GTC 2026, a desktop supercomputer designed to run AI models with up to one trillion parameters locally, without relying on cloud infrastructure. This machine, packing 748GB of coherent memory and delivering 20 petaflops of AI performance, represents a significant leap in personal computing, potentially rivaling the impact of the original Mac Pro for creative professionals.

Addressing the AI Infrastructure Challenge

The DGX Station arrives at a critical juncture for the AI industry. While the most powerful AI models demand extensive data center infrastructure, developers and enterprises increasingly prioritize keeping their data, agents and intellectual property secure and local. Nvidia’s DGX Station directly addresses this tension, offering a six-figure investment that bridges the gap between cutting-edge AI and individual engineers’ workspaces.

Unpacking the DGX Station’s Performance

At the heart of the DGX Station is the GB300 Grace Blackwell Ultra Desktop Superchip, which integrates a 72-core Grace CPU and a Blackwell Ultra GPU via Nvidia’s NVLink-C2C interconnect. This connection provides 1.8 terabytes per second of coherent bandwidth, seven times faster than PCIe Gen 6, enabling seamless data sharing between the CPU and GPU. Twenty petaflops of compute performance – 20 quadrillion operations per second – would have placed this machine among the world’s top supercomputers less than a decade ago, comparable to the Summit system at Oak Ridge National Laboratory in 2018, which occupied a space equivalent to two basketball courts.

The Importance of Unified Memory

The 748GB of unified memory is a key feature. Trillion-parameter models, like GPT-4, are massive neural networks that require complete loading into memory for operation. Insufficient memory renders processing speed irrelevant. The DGX Station’s coherent architecture eliminates latency issues typically associated with transferring data between CPU and GPU memory pools.

NemoClaw: The Operating System for Personal AI

Nvidia designed the DGX Station for the next wave of AI: autonomous agents capable of reasoning, planning, coding, and continuous task execution. Central to this vision is NemoClaw, a new open-source stack combining Nvidia’s Nemotron open models with OpenShella, a secure runtime that enforces policy-based security, network, and privacy controls for autonomous agents. Nvidia CEO Jensen Huang has positioned NemoClaw – and its broader platform, OpenClaw – as “the operating system for personal AI,” drawing parallels to Mac and Windows.

Local Processing for Always-On Agents

Cloud instances are inherently on-demand, but always-on agents require persistent compute, memory, and state. A local machine running 24/7 with local data and models within a secure environment offers an architectural advantage over rented cloud GPUs. The DGX Station can function as a personal supercomputer or a shared compute node for teams, and supports air-gapped configurations for highly sensitive environments.

Architectural Continuity: From Desktop to Data Center

Applications developed on the DGX Station seamlessly migrate to Nvidia’s GB300 NVL72 data center systems without code rearchitecting. This architectural continuity streamlines the development process, eliminating the time and resources often lost rewriting code for different hardware configurations. Nvidia offers a vertically integrated pipeline: prototype on the desktop, then scale to the cloud when ready.

Expanding the DGX Ecosystem

Nvidia has also expanded the capabilities of the DGX Spark, the Station’s smaller counterpart, with new clustering support. Up to four Spark units can now operate as a unified system with near-linear performance scaling, creating a “desktop data center” without the demand for traditional rack infrastructure.

Early Adoption and Industry Applications

Initial customers of the DGX Station represent industries rapidly integrating AI into their core operations. Snowflake is utilizing the system to locally test its Arctic training framework. EPRI, the Electric Power Research Institute, is leveraging AI-powered weather forecasting to enhance electrical grid reliability. Medivis is integrating vision language models into surgical workflows. Microsoft Research and Cornell University are deploying the systems for large-scale AI training.

Availability and Pricing

Systems are available for order now and will ship in the coming months from ASUS, Dell Technologies, GIGABYTE, MSI, and Supermicro, with HP joining later in the year. While Nvidia has not disclosed pricing, the components and historical DGX pricing suggest a six-figure investment.

Model Agnostic Support

The DGX Station supports a wide range of models, including OpenAI’s gpt-oss-120b, Google Gemma 3, Qwen3, Mistral Large 3, DeepSeek V3.2, and Nvidia’s own Nemotron models, demonstrating its versatility and open ecosystem approach.

Nvidia’s Broader AI Strategy

The DGX Station is part of a larger Nvidia strategy to provide AI compute at every scale. This includes the Vera Rubin platform, the Vera Rubin NVL72 rack, and the Vera Rubin Space Module for orbital data centers. Nvidia is also forging partnerships with companies like Adobe, BYD, Nissan, and Mistral AI, and has introduced Dynamo 1.0, an open-source inference operating system.

The Future of AI Infrastructure

For years, cloud GPU instances have been the default for serious AI perform. While the cloud remains crucial, the DGX Station offers a viable local alternative for specific workloads, such as fine-tuning models on proprietary data, running inference for internal agents, and prototyping. This expands Nvidia’s market reach while reinforcing its cloud business. The DGX Station represents a shift towards a future where a supercomputer resides on every desk, powering an agent that operates continuously.