This is the smallest AI supercomputer ever made
What's the story
NVIDIA CEO Jensen Huang has personally delivered the company's latest innovation, DGX Spark, to Elon Musk at SpaceX's Starbase facility in Texas. The pocket-sized powerhouse is being hailed as the smallest AI supercomputer ever built. It can run models with up to 200 billion parameters locally. Huang's visit coincided with preparations for the 11th test flight of Starship, the world's most powerful launch vehicle.
Delivery details
A full-circle moment for Huang
Huang reminisced about offering the first DGX to OpenAI years ago, calling this handoff "a full-circle moment." He joked, "Imagine delivering the smallest supercomputer next to the biggest rocket," while sharing pizza and stories with Musk and SpaceX engineers. The DGX Spark weighs just 1.2kg and is roughly as big as a hardcover book but delivers an impressive 1 petaflop of AI performance.
Tech specs
Specifications and features of DGX Spark
The DGX Spark is powered by NVIDIA's GB10 Grace Blackwell Superchip. It comes with 128GB of unified memory, NVLink-C2C connectivity, and NVMe storage - all tailored for developers, researchers, and creators who want data-center-grade performance on their desk. The supercomputer also comes with NVIDIA's full AI software stack including Cosmos, Qwen3, and NIM microservices to build custom image generators or chatbots right out of the box.
AI evolution
Transitioning from cloud to local AI
NVIDIA sees DGX Spark not just as a machine but as a mission statement. The company believes it marks the transition from cloud-bound AI to local and creative intelligence. Huang said, "DGX Spark puts a petaflop of AI performance within arm's reach of everyone." This could be a major step forward in making advanced artificial intelligence more accessible and usable for individuals and small businesses alike.