NVIDIA DGX Spark is a small desktop computer from NVIDIA for building and running AI models and agents on your own desk instead of in the cloud.
In plain words
It is built around the GB10 Grace Blackwell Superchip, which NVIDIA says delivers as much as 1 petaFLOP of AI computing at FP4 precision. It comes with 64 GB or 128 GB of shared memory and NVIDIA's software stack. NVIDIA says the 64 GB version handles models of up to 100 billion parameters and goes on sale on 23 October from PC makers, starting at $4,999. Up to four units can be linked to work on bigger models.
A quick example
Imagine a small team that wants a coding assistant that never sends its code to an outside service. They download an open model and an inference tool onto a DGX Spark under a desk, and the assistant runs there all day.
Where you'll meet it
- Running AI agents and large open models locally, without cloud servers.
- Fine-tuning models of up to 70 billion parameters at a desk.
Related
Sources
- DocsNVIDIA DGX Spark, NVIDIA · read 21 Oct 2026
- DocsNVIDIA DGX Spark 64GB Gives Developers More Ways to Build and Scale Local AI, NVIDIA · read 21 Oct 2026
Checked by WhatIsAI against the sources above · 21 Oct 2026. Spotted an error? Report it.