NVIDIA Starts Shipping AI Supercomputer DGX Spark for Physical AI Development

• NVIDIA has begun shipping DGX Spark systems globally, enabling local development of agentic and physical AI.
• The compact platform integrates NVIDIA’s full AI stack, now available from major OEMs and partners.

NVIDIA has started shipping DGX Spark, a new desktop AI supercomputer designed to support the next generation of physical and agentic AI development. Starting October 15, developers worldwide can access the system through NVIDIA.com and leading hardware partners, including Acer, ASUS, Dell Technologies, HP, Lenovo, and MSI.

DGX Spark includes the Grace Blackwell GB10 Superchip and delivers 1 petaflop of compute with 128GB of unified memory, enabling developers to run and fine-tune large models locally. The system integrates NVIDIA GPUs, CPUs, networking, CUDA libraries, and preinstalled software, reducing reliance on cloud infrastructure. Supported tools include FLUX.1 for image generation, Qwen3 for chatbot development, and NVIDIA Cosmos for building multimodal vision agents and summarization tools.

Early adopters of DGX Spark include research institutions and industry leaders such as Meta, Microsoft, Hugging Face, JetBrains, and NYU’s Global Frontier Lab, which are using the system to test models, validate tools, and build privacy-sensitive applications.

🌀 Tom’s Take:

Traditional PCs and workstations weren’t built for today’s AI demands. DGX Spark puts a full AI supercomputer on the desktop, stacked, local, and ready to support the next wave of agentic and physical AI development.

Source: NVIDIA