NVIDIA DGX Spark - voted 'The Best Inventions of 2025' by Time Magazine
NVIDIA DGX Spark is redefining AI accessibility by condensing datacenter-grade power into a desktop device. Equipped with 1 petaflop of compute, 128GB memory, and the Grace Blackwell GB10 Superchip, it supports AI models with up to 200 billion parameters. Compact yet powerful, it merges GPUs, CPUs, networking, and NVIDIA’s AI software stack—unlocking advanced AI development for labs and startups in a desktop friendly form factor.
The Next Big Wave
Back in 2016, its' predecessor the NVIDIA DGX-1 was delivered to Elon for a small startup called OpenAI, the creators of ChatGPT. It kickstarted the AI evolution and started the first splash of breakthroughs. The demands of AI workloads are quickly forcing teams to move to the cloud or data centers to have enough power to drive through. Jensen Huang wants change that by placing AI computer in every hands of every developer - making the tools accessible to everyone.
For the Developers
NVIDIA DGX Spark is a new kind of desktop AI computer built to bring datacenter power straight to developers’ desks. Powered by the Grace Blackwell GB10 Superchip and the Grace Blackwell architecture, it delivers 1,000 TOPS of performance and 128GB of unified memory—enough to fine-tune models up to 70B parameters and run inference on 200B. And for those pushing even further, NVIDIA ConnectX networking and the NVIDIA Link Cable allow two systems to scale up to 405B-parameter models.
Preinstalled with the NVIDIA AI software stack, DGX Spark mirrors enterprise AI factory environments, letting developers start building immediately. Running NVIDIA DGX OS on Ubuntu Linux, it supports familiar tools like PyTorch, Jupyter, and Ollama, plus access to NVIDIA NIM and NVIDIA Blueprints. From refining FLUX.1 models to building AI chatbots with Qwen3 or deploying vision agents with NVIDIA Cosmos, it’s a powerful platform that makes advanced AI development fast, flexible, and local.
