RTX Spark is NVIDIA's Windows laptop platform that packs Blackwell RTX graphics, a Grace CPU and the CUDA ecosystem into a single superchip. On October 16, 2026, NVIDIA announced in an official news release that RTX Spark laptops are on sale, with first systems from ASUS, Dell, HP, Lenovo, Microsoft and MSI starting at $2,599.
The first lineup: every major maker, and a clear price line
The launch range includes 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, sold through the makers' own sites and major retailers, with configurations and pricing varying by region. A $2,599 starting price pins it firmly in the premium thin-laptop tier: what you buy is not a graphics card but a whole platform that can run large models locally, edit 8K video and play AAA games. It also completes the story from this site's earlier RTX Spark ecosystem readiness report: after the software was laid out, the hardware is on shelves, with firm prices and model names.
128GB of unified memory is the real pitch
RTX Spark offers up to 128GB of unified memory shared between CPU and GPU, and up to 1 petaflop of FP4 AI performance. In practice that changes how work flows: 3D scenes larger than 90GB can render directly, and local large models can stay resident alongside creative apps instead of shuffling between separate memory pools. The Grace CPU inside was designed with MediaTek and leans on power efficiency. Together with local agents ready on day one, including guided installs for GitHub Copilot, Hermes and OpenClaw, the scenario NVIDIA is selling is concrete: edit and code by day, let agents keep working locally, with data never leaving the machine.
Day-one software says more than the spec sheet
Adobe Photoshop and Premiere ship native-support beta builds at launch, using TensorRT to accelerate AI editing features, with Substance 3D to follow. On the gaming side, a new DLSS 5 model lifts frame rates by up to 1.4 times, and AAA titles pass 100 frames per second at 1440p. For developers, the first PyTorch-CUDA build for Windows on Arm, a native Python build and a native Conda installer all land the same day. For anyone hesitating, the test is simple: if your workflow is on the day-one support list, the launch price buys immediate capability; if it is not, 128GB is just a number. Waiting for the first real battery and noise tests before paying for this superchip costs nothing.
