NVIDIA has launched DGX Spark, the world’s smallest AI supercomputer, marking a pivotal step in bringing supercomputing power directly to developers, researchers, and creators. CEO Jensen Huang personally delivered the first unit to Elon Musk at SpaceX’s Starbase facility in Texas, symbolizing a full-circle moment. Nine years earlier, Huang handed the first DGX-1 to Musk at OpenAI, a move that helped ignite the AI revolution. Now, DGX Spark brings that same transformative power into a compact desktop form, enabling local AI development at unprecedented scale. Built on the NVIDIA Grace Blackwell architecture, DGX Spark delivers up to one petaflop of AI performance and 128GB of unified CPU-GPU memory, making it capable of running models with up to 200 billion parameters for inference and fine-tuning models of up to 70 billion parameters entirely on the device. This level of performance, previously only available in large data centers, is now accessible on a desk, accelerating the development of agentic and physical AI applications. The system integrates NVIDIA’s full AI stack—GPUs, CPUs, high-speed networking via ConnectX-7 200 Gb/s, NVLink-C2C technology for 5x the bandwidth of PCIe Gen 5, and the CUDA ecosystem—into a single, ready-to-use platform. It comes preloaded with NVIDIA AI software, including access to models, libraries, and NIM microservices, allowing developers to immediately begin building AI agents, custom chatbots, vision systems, and more. DGX Spark is designed for real-world use across diverse fields. Early adopters include researchers at NYU’s Global Frontier Lab, who are leveraging its power for rapid prototyping of advanced AI models, especially in privacy-sensitive domains like healthcare. Creative studios like Refik Anadol’s and companies such as Zipline, Ollama, Hugging Face, and Meta are using it to refine AI workflows, optimize models like FLUX.1 and Qwen3, and develop vision-language agents. The system is now shipping globally starting October 15, available directly through NVIDIA.com and via major partners including Acer, ASUS, Dell Technologies, GIGABYTE, HP, Lenovo, MSI, and Micro Center in the U.S. This broad distribution ensures widespread access to high-performance AI computing for developers, startups, academic labs, and enterprises. The delivery to SpaceX also underscores the growing role of AI in physical systems and space technology. With the 11th test of Starship underway, DGX Spark’s arrival at Starbase signals a new era where AI development is no longer confined to the cloud or massive data centers. Instead, it’s moving into labs, studios, and engineering teams where ideas are born. NVIDIA’s vision is clear: democratize AI supercomputing. By placing petaflop-class performance in the hands of every developer, DGX Spark aims to fuel the next wave of innovation. As Huang stated, “We return to that mission — placing an AI computer in the hands of every developer to ignite the next wave of breakthroughs.” This launch is not just a product rollout—it’s a shift in how AI is developed. With DGX Spark, AI is no longer a distant, centralized resource. It’s local, accessible, and ready to transform how creators and engineers build the future.