Nvidia’s RTX Spark laptops are here, from $2,599 with up to 128GB of memory

Nvidia said its RTX Spark laptops are available today worldwide. Systems from ASUS, Dell, HP, Lenovo, Microsoft and MSI start at $2,599 and are built around one chip with up to 128GB of unified memory.

What is new

RTX Spark is a superchip that combines a Blackwell RTX GPU with an Nvidia Grace CPU, a design Nvidia developed with MediaTek. It runs Windows. Nvidia lists up to 128GB of unified memory and up to 1 petaflop of FP4 AI performance. The chip supports CUDA, DLSS, OptiX, Reflex, G-SYNC and TensorRT, as well as hardware-accelerated video.

The first laptops are the 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. The company says prices and configurations vary by region. Laptops are sold through partner websites and major retailers. A limited-time promotion is offered to new Perplexity users.

Local AI and agents

Nvidia is pitching the machines as hosts for AI agents that run on the device. At launch it lists support for GitHub Copilot, Hermes, OpenClaw, Perplexity, Pi and WorkBuddy, with a guided setup that avoids the command line. Nvidia also introduced Nemotron 3.5 Bolt 74B-A9B 24GB, a 74-billion-parameter model aimed at tool calling and multistep tasks.

For permissions, Nvidia points to its OpenShell software and Microsoft Execution Containers, which control what an agent can access. A feature called Nvidia PAIR lets an RTX Spark laptop serve AI inference to other computers on a home network.

For developers, Nvidia says it is shipping the first PyTorch-CUDA build for Windows on Arm and supports 97% of the top 15,000 Python packages. Windows Subsystem for Linux and Windows Developer Config handle setup. Nvidia also published Nvidia Skills, an open source GitHub repository meant to teach agents to use CUDA-X tools. It says work built locally can move to more than 150 million RTX AI PCs and to cloud GPUs.

Creative and gaming features

Nvidia says the laptops can edit 8K 4:2:2 video and render 3D scenes up to 90GB, including on battery. Native RTX Spark support for Adobe Photoshop and Premiere is in beta at launch, and Substance support is coming. Nvidia Broadcast 2.3 removes background noise from recordings. Beta RTX Video features offer up to 4K super resolution and 4x frame generation, and new ComfyUI nodes convert 15 fps AI video to 60 fps.

For gaming, launch titles include Cyberpunk 2077, Fortnite, Gears of War: E-Day, Marvel Rivals, World of Warcraft and NBA 2K27. NBA 2K27 uses DLSS 5 with 3D-Guided Neural Rendering, and Nvidia says a new model raises DLSS 5 frame rates by up to 1.4x.

Jensen Huang, Nvidia’s founder and CEO, said RTX Spark brings Nvidia AI, RTX graphics and CUDA together with Windows in one PC. Yusuf Mehdi, a Microsoft executive vice president, said these PCs give creators and developers new local compute.

Why it matters

The $2,599 starting price is a floor, not the cost of the 128GB machines the launch is built around. According to Notebookcheck, HP’s US store briefly listed the OmniBook X 14 at $2,999.99 and the OmniBook Ultra 16 at $3,199.99. Preconfigured versions with 64GB of RAM and a 2TB SSD were listed at $4,499.99 and $4,999.99, and the listings said both could be configured with up to 128GB of unified memory. HP had not officially announced pricing when those listings appeared, so final retail prices may differ.

The comparison that matters is Apple, the only other mainstream vendor selling a laptop with this much unified memory. 9to5Mac reported that at launch the top 16-inch MacBook Pro with the M5 Max, 128GB of memory and 8TB of storage cost $7,349. Lower-storage 128GB configurations cost less. AppleInsider’s price tracker lists a 14-inch M5 Max with 128GB and 2TB at $5,099, though Wccftech has since reported Apple price increases on some high-end configurations.

On those figures, a well-equipped RTX Spark laptop would sit roughly level with or below a comparable 128GB Mac, while adding CUDA and Windows software. That is a narrower claim than “cheaper than the cloud”. Whether local inference pays back depends on how many tokens a user actually generates, and that figure is not in Nvidia’s announcement. Street prices for the 128GB configurations will show how much room Nvidia’s partners have to undercut Apple.