Nvidia Unveils DGX Spark — The World’s Smallest AI Supercomputer

​A Supercomputer That Fits on Your Desk
Nvidia just launched the DGX Spark, which it calls the world’s smallest AI supercomputer. The compact machine packs Nvidia’s new GB10 Grace Blackwell superchip and advanced ConnectX-7 networking, giving it the power of a data center inside a desktop-sized device. It’s designed to bring enterprise-level AI computing to smaller companies and independent developers — no cloud rental or massive server bills required.

Bringing AI Power to Everyone
The DGX Spark is part of Nvidia’s broader push to democratize AI hardware. Built with partners like ASUS and Dell, the system lets users train and run AI models locally with up to 128GB of memory. Nvidia CEO Jensen Huang said the Spark continues the company’s mission that began with the 2016 DGX-1, which helped power OpenAI’s early breakthroughs. “With DGX Spark, we’re putting AI in the hands of every developer to ignite the next wave of innovation,” Huang said.

Designed for Real AI Work — Not Gaming
While it might look like a futuristic desktop, the Spark isn’t for gaming. It runs on Nvidia’s Linux-based DGX OS and full AI software stack, purpose-built for model training and deployment. Users can even link two Sparks together to handle AI models of up to 405 billion parameters, rivaling some of the world’s largest supercomputers.

Nvidia’s Expanding AI Empire
​The Spark joins Nvidia’s lineup alongside the more powerful DGX Station, which runs on the GB300 Grace Blackwell Ultra chip. Together, they round out Nvidia’s dominance in the AI hardware market — one strengthened by major supply deals with OpenAI, xAI, Google, Amazon, Meta, and Microsoft. Nvidia remains at the center of the global AI boom, supplying the computing muscle behind nearly every major model on the planet.