TL;DR: NVIDIA RTX Spark laptops went on sale on October 16 from ASUS, Dell, HP, Lenovo, Microsoft and MSI. They pair an NVIDIA GPU and CPU on one chip with up to 128GB of memory shared by both, so you can run large AI models and agents on the laptop itself. NVIDIA’s press release puts the starting price at $2,599. “Up to 128GB” is the top configuration, though, so check the memory in the exact model you’re buying.

What happened

NVIDIA announced on October 16 that RTX Spark laptops are “available globally today” through PC makers’ websites and major retailers, “starting at $2,599.” It adds that “pricing and configurations vary by region.”

According to NVIDIA’s press release:

  • The chip. RTX Spark combines an NVIDIA Blackwell RTX GPU and an NVIDIA Grace CPU in one “superchip,” with up to 128GB of unified memory and up to 1 petaflop of FP4 AI performance. MediaTek worked with NVIDIA on the CPU design.
  • The laptops. ASUS ProArt P14 and P16, Dell XPS 16, HP OmniBook Ultra and OmniBook X 14, Lenovo Yoga Pro 9n, Microsoft Surface Laptop Ultra, and MSI Prestige N16 Flip AI+ and Prestige N16. NVIDIA says more systems will follow.
  • Agents. Guided installers for GitHub Copilot, Hermes, OpenClaw, Perplexity, Pi and WorkBuddy, “without requiring a command line.” NVIDIA’s OpenShell runtime and Microsoft Execution Containers “help users control agent permissions.”
  • Local models. NVIDIA says its Nemotron 3.5 Bolt 74B-A9B 24GB model runs on “supported RTX Spark configurations.” A feature called NVIDIA PAIR lets an RTX Spark laptop serve AI to other Windows, Linux and macOS computers on a home network.
  • Developer tools. NVIDIA and Microsoft list the “first PyTorch-CUDA build for Windows on Arm,” a native Python build, a native Conda installer, Windows Subsystem for Linux and support for “97% of the top 15,000 Python packages.”

Two different starting prices. NVIDIA’s October 16 press release says laptops start at $2,599. NVIDIA’s October 17 blog post says “Starting at $2,399” but doesn’t say which laptop or configuration that refers to. In the US, ASUS lists the ProArt P14 from $2,599.99 and the ProArt P16 from $2,799.99. We use the press release figure until NVIDIA explains the difference.

Why it matters

RTX Spark puts up to 128GB of memory that the GPU can use into a slim Windows laptop. For a small team, local models mean client files and code can stay on the device, agents keep working offline, and everyday experiments don’t add to an API bill.

The catch is the memory you actually get. ASUS lists five ProArt P16 configurations: 24GB, 32GB, two with 64GB, and 128GB. Our analysis: on a unified-memory machine, Windows, your apps and the model all draw from the same pool. A 24GB model like Nemotron 3.5 Bolt would take the whole pool on a 24GB configuration, leaving nothing for Windows or your apps, so in practice you’d need a 32GB or larger configuration. NVIDIA doesn’t list which configurations count as “supported” for Bolt, so ask the PC maker or check its spec sheet.

NVIDIA’s developer tools for RTX Spark include what it calls the “first PyTorch-CUDA build for Windows on Arm,” so expect these laptops to run Windows on Arm, which still affects app and library support. NVIDIA’s 97% figure means some Python packages still lack support, and the release doesn’t say which ones.

What to do

  1. Pick the memory before the brand. Decide which models you want to run, add room for Windows and your apps, and only then compare laptops. The headline “up to 128GB” is the top configuration, not the starting one.
  2. Check your tools on Windows on Arm. Before you buy, confirm that the libraries, drivers and desktop apps your work depends on run on Arm, especially anything with native code.
  3. Weigh it against your cloud spend. If you only call a hosted model now and then, $2,599 buys a lot of API usage. Local hardware makes more sense for daily heavy use, private client data or offline work. Our guide to AI agent spending caps shows how to keep cloud bills predictable if you stay in the cloud.
  4. Lock down local agents. An agent on your laptop can reach your files. Use the permission controls NVIDIA names, OpenShell and Microsoft Execution Containers, and give agents access only to the folders they need.
  5. Read up on the model first. Our Nemotron 3.5 Bolt story covers what NVIDIA claims for the 24GB model and how to try it on hardware you already own.

Sources & method

Sources

Method. Last checked October 18, 2026. Specs, prices, quotes and the laptop list come from NVIDIA’s press release and blog and from ASUS’s announcement. Performance figures are NVIDIA’s own. We haven’t tested an RTX Spark laptop. Prices are US list prices and vary by region and retailer. No affiliate links on this page. How we work.

Spotted an error? See our corrections policy and tell us.