Nvidia says it will begin selling the DGX Spark mini PC, with DGX OS, for AI developers on October 15 on Nvidia.com and select third-party retailers for $3,999
(PCMag/Michael Kan) … It's not a consumer desktop, but Nvidia's foray into an AI developer-focused mini PC is finally ready to launch.
PCMagMichael Kan
Context & Ripple Effects
DGX Spark began as Project Digits, positioned around a GB10 Grace Blackwell chip and the ability to run large models locally; Nvidia later moved it into the DGX lineup and opened reservations for the renamed system.
The retail launch turns that earlier product roadmap into a purchasable developer system. Early hands-on coverage frames its 128GB unified memory and Nvidia ecosystem as central strengths, while noting trade-offs in performance and bandwidth.
First-order effects
AI developers can buy a compact DGX system with DGX OS through Nvidia and selected retailers at a stated $3,999 price, rather than waiting on reservations or relying solely on remote capacity.
Nvidia extends DGX from cloud and datacenter infrastructure into a local development endpoint, with its software environment bundled into the hardware purchase.
Second-order effects
The $3,999 system creates a clearer local-compute alternative for teams whose workloads benefit from large unified memory, while Nvidia's own DGX Cloud offering remains the scaling path for larger needs.
Workstation and mini-PC vendors targeting local AI development will need to differentiate on memory capacity, software support, performance, or price rather than treating compact form factor alone as a differentiator.
Third-order effects
If DGX-class hardware continues moving from datacenters to desks, AI infrastructure competition will increasingly center on integrated hardware-and-software environments that let developers move work between local systems and cloud capacity.
The product also illustrates a potentially durable split in AI compute: local machines for iteration and model experimentation, paired with centralized infrastructure for workloads that exceed local performance or bandwidth limits.
The trend: AI infrastructure vendors are productizing a continuum from developer desktops to cloud-scale systems, using software integration to retain users across both environments.
The Metrum AI team had a blast seeing stellar early results on Nvidia DGX Spark and integrating Metrum Insights with DGX Spark to provide a seamless environment for benchmarking and evaluation. #SparkSomethingBig @NVIDIAAI [image]
Seems like reviews for the DGX Spark ("Project Digits") started rolling in, about what you'd expect, the lackluster VRAM bandwidth brings it down I am hoping they depreciate well and I can pick one up for use as a lackluster portable development workstation. The compute is neat […
🚀 SGLang In-Depth Review of the NVIDIA DGX Spark is LIVE! Thanks to @NVIDIA 's early access program, SGLang makes its first ever appearance in a consumer product, the brand-new DGX Spark. The DGX Spark's 128GB Unified Memory and Blackwell architecture set a new standard for loc…
Personal AI computing has just become more accessible than ever before with the new DGX Spark from @nvidia - a tiny yet powerful Blackwell GPU in your hand. Check out the firsthand unboxing video and blog from the SGLang community members @yvbbrjdr and @richardczl.
🚀🚀Just spun up gpt-oss-20B with vLLM on NVIDIA's brand-new DGX Spark machine — a surprisingly powerful little box! This isn't a rigorous benchmark, but early numbers look promising on a quick test (512×128 tokens) with stable serving and smooth setup! [image]
NVIDIA DGX Spark is here! It's so exciting to make Ollama run on @nvidia DGX Spark. Super amazing to see 128GB of unified memory and the Grace Blackwell architecture. 👇👇👇 [image]
The new NVIDIA DGX Spark is here, and it supports Docker Model Runner. As a launch partner, we got a sneak peek on the new Grace Blackwell GB10 Superchip-powered workstation. It's every bit as powerful as it looks. Read our thoughts in the blog post: https://www.docker.com/...
🎉 To celebrate DGX Spark shipping worldwide starting Wednesday, our CEO Jensen Huang just hand-delivered some of the first units to @elonmusk, chief engineer at @SpaceX 🚀, today in Starbase, Texas. The exchange was a connection to the new desktop AI supercomputer's origins — [ima…
That's for the DGX Spark 😀 This is ~100X more compute per watt than the DGX-1, the first ever dedicated AI computer, that Jensen gave me at OpenAI in 2016!
🚨NVIDIA and its partners will start shipping DGX Spark — the world's smallest AI supercomputer. Early recipients are testing, validating and optimizing their tools, software and models for DGX Spark. Built on the NVIDIA Grace Blackwell architecture, DGX Spark integrates NVIDIA …
NEWS: Nvidia CEO Jensen Huang hand-delivered the world's smallest AI supercomputer — a petaflop of performance in a box that fits on a desk — to @elonmusk today. “DGX Spark packs 128GB of unified memory and is powerful enough to run models with 200 billion parameters locally.” [i…
🔋 NVIDIA and 50+ MGX partners are rolling out Vera Rubin NVL144 racks and shifting to 800 volts DC power to build gigawatt AI factories with open OCP specs. The NVL144 compute tray is 100% liquid cooled with a PCB midplane replacing cables, expansion bays for ConnectX-9 [image]