NVIDIA's RTX Spark Laptops Ship: Blackwell RTX and 128GB Unified Memory in Windows PCs From $2,399

NVIDIA's RTX Spark laptops are now available, the company said, putting its Blackwell RTX graphics architecture and the full NVIDIA AI software platform into thin, premium Windows portables. Six manufacturers are shipping machines at launch: ASUS, Dell, HP, Lenovo, Microsoft and MSI. Entry pricing starts at $2,399, which places the line above mainstream consumer laptops and alongside high-end creator and gaming notebooks.

The headline specification is memory. RTX Spark systems carry up to 128GB of unified memory, a configuration that lets the GPU and CPU draw from a single pool rather than splitting work across separate RAM allocations. For local AI workloads, that changes what fits on the machine: larger language models and diffusion pipelines that would otherwise require a cloud endpoint can run on the device, subject to the usual tradeoffs in speed and power draw.

NVIDIA is positioning the laptops around three uses — running AI models locally, content creation, and gaming — rather than gaming alone. That framing mirrors how the company has marketed its desktop GPUs over the past two years, as buyers weigh whether to rent inference from a cloud provider or own the hardware outright. Bundling the full NVIDIA AI stack, rather than just the graphics driver, is the part aimed squarely at developers and studios.

The lineup is broad for a launch. ASUS, Dell, HP, Lenovo and MSI cover the bulk of global PC volume, while Microsoft's participation gives the platform a first-party Windows option. Spreading across six vendors in one cycle is unusual for a new NVIDIA laptop generation and suggests the company wants shelf space in retail and enterprise channels quickly, not a slow rollout through a single flagship model.

At $2,399, RTX Spark does not compete on price with mainstream Windows laptops, many of which now ship with dedicated NPUs for lighter AI tasks. The pitch instead rests on raw GPU compute and memory capacity, which matter most to people running models locally rather than calling an API. Whether buyers accept that premium is the central commercial question for the line.

What to watch next is real-world performance data. Independent benchmarks on memory bandwidth, tokens-per-second for popular open models, and battery life under sustained AI load will decide whether 128GB of unified memory is a working tool or a spec-sheet number. Availability of the top memory configuration across all six vendors, and at what price, is the other figure to track.