AINewsNVIDIAPC Hardware

NVIDIA’s RTX Spark Mini-PCs Are Out Now: Windows Desktops With Up to 128GB of Memory for Local AI

If you have wanted a desk-sized Windows PC that can run big AI models without sending your files to the cloud, NVIDIA says it is here. NVIDIA announced today (Nov. 18, 2026) that its RTX Spark compact desktops are available now from ASUS, Dell and MSI, with up to 128GB of unified memory and up to 1 petaflop of AI performance.

What NVIDIA says is inside

  • CPU: 20-core NVIDIA Grace
  • GPU: NVIDIA Blackwell RTX with 6,144 CUDA cores
  • Memory: up to 128GB unified memory
  • AI performance: up to 1 petaflop
  • Networking and OS: 10-gigabit Ethernet, running Windows (with Windows on Arm apps and Linux tools through WSL)

Who is selling one

MakerStatusModel
ASUSAvailable nowProArt GR1X mini PC
MSIAvailable nowEdgeMesa N AI+ series
DellAvailable nowNot named in NVIDIA’s post
Acer, GIGABYTE, HPComing soonMini PCs
MicrosoftComing soonSurface RTX Spark Dev Box
Source: NVIDIA blog, Nov. 18, 2026.

What it is meant to do

NVIDIA’s pitch is local AI. It says these PCs can run 120-billion-parameter language models, render 90GB-plus 3D scenes in real time and edit 4:2:2 video in real time. Running supported models locally keeps the code and files you process on the PC, and it avoids per-token cloud API charges.

For developers, NVIDIA lists native CUDA support, Linux tools through WSL, Hermes Agent, NVIDIA NemoClaw blueprints and Unsloth for fine-tuning. A feature called NVIDIA Switchyard, inside Hermes, routes tasks between local and cloud models to keep cloud costs down.

Creators get some of the news too. NVIDIA says Adobe Premiere gets TensorRT-RTX acceleration for Enhance Speech, Object Mask, Scene Edit Detection and Auto Reframe, with more than 50% faster turnaround on everyday H.264 deliverables. It also says Illustrator and InDesign now have native Windows on Arm versions, with Illustrator up to 30% faster than its emulated version. Blender 5.3 adds DLSS 4.5 Ray Reconstruction in the Cycles viewport, and ComfyUI gets RTX Video Frame Generation.

What we could not confirm

  • Prices. NVIDIA’s post lists none. One outlet reports MSI EdgeMesa prices, but that comes from a single source and we could not confirm it with MSI, so we are not repeating it. For context, NVIDIA’s earlier RTX Spark laptops release (Oct. 16) listed a starting price of $2,599.
  • Dell’s model. NVIDIA says Dell is shipping but does not name the system.
  • Independent results. Every performance and speed-up number above is NVIDIA’s own. We have not seen independent benchmarks yet.

PCGeek’s take

This is the most interesting thing to land on a desk this week: a small Windows box built around the same Grace and Blackwell ideas NVIDIA uses in its data centers, aimed at people who want to run models themselves. Up to 128GB of memory that the CPU and GPU share is the headline, because model size is usually what limits local AI. The catch is that it is early. Until prices and independent tests show up, treat the speed claims as NVIDIA’s pitch, and check which memory configuration you are actually buying. NVIDIA is clearly busy right now, so for the money side of its week see our earnings coverage.

Would you run your AI models on a desktop box like this instead of in the cloud? Tell us in the comments.

Follow @pcgeekinfo for more tech news.

Image: NVIDIA blog (ASUS ProArt GR1X). Credit is not a license. Sources: NVIDIA Blog, NVIDIA Newsroom.

PCgeek

Techie, YouTuber, Writer, Creator

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.