NVIDIA has started shipping DGX Spark, a compact powerhouse described as the world’s smallest AI supercomputer. The system brings petascale AI capability to a desktop-friendly form factor, making high-end AI computing accessible to developers worldwide.
Source: NVIDIA Newsroom
DGX Spark delivers 1 petaflop of AI performance and 128 GB of unified memory in a compact design. It allows developers to run inference on models up to 200 billion parameters and fine-tune models up to 70 billion parameters locally, without depending on large cloud environments.
Built on the Grace Blackwell superchip architecture that combines GPUs, CPUs, networking, and NVIDIA’s full AI software stack in one ready-to-use platform.
Supports local AI workflows such as agent development, vision and language models, and custom fine-tuning.
Ships through major OEM partners, including Acer, ASUS, Dell, GIGABYTE, HP, Lenovo, and MSI.
DGX Spark is built for developers who want enterprise-grade compute power in smaller, more flexible spaces.
AI workloads are quickly outgrowing the capacity of traditional PCs and workstations. DGX Spark fills that gap by delivering data-center-grade performance in a form that fits into labs, startups, and studios.
For developers building agentic systems, autonomous agents, or edge AI applications, DGX Spark offers lower latency, complete control, and stronger data privacy compared to cloud-only setups.
This launch continues NVIDIA’s AI journey that began with the DGX-1 in 2016, hand-delivered to OpenAI, and now extends to making supercomputing accessible to every developer.
Adoption and optimization of DGX Spark by frameworks like Hugging Face and other AI ecosystems.
The rise of compact supercomputing as a new hardware trend in AI R&D.
Competitive responses from AMD, Intel, AWS, and Google in hybrid and edge AI computing.
How smaller research teams and institutions use DGX Spark to compete with cloud-first AI labs.