RTX Spark Laptops Land: $2,599 Entry, 128GB Ceiling, 1 Petaflop Target
Microsoft and its OEM partners have put a price on local agentic computing. The first laptops powered by Nvidia's RTX Spark chip, led by the Surface Laptop Ultra alongside models from Dell, Lenovo, MSI, HP and Asus, start at $2,599 and launch as early as October 16th.
The entry point, detailed by The Verge, is $2,599 for an 18-core CPU configuration with 24GB of unified memory and 512GB of storage. That positions RTX Spark systems above mainstream ultrabooks and squarely in mobile workstation territory. A 48GB Microsoft Surface Laptop Ultra model is listed at $3,999.99 on Best Buy. At the top of the stack disclosed so far, a configured Asus ProArt P16 with 20-core chip, 128GB of unified memory and 2TB of storage costs $6,999.99.
Silicon options center on two bins. Specs outlined by Windows Central list an 18-core RTX Spark configuration with 5,120 GPU cores and a 20-core configuration with 6,144 GPU cores for the Surface Laptop Ultra. Microsoft describes the system as using a new NVIDIA chip that combines an ultra-efficient CPU with a powerful RTX GPU, according to Microsoft. The Surface Laptop Ultra supports up to 128GB of unified memory.
The OEM footprint is broad from day one. Nvidia lists the Dell XPS 16, HP OmniBook Ultra 16, Lenovo Yoga Pro 9n and MSI Prestige N16 Flip AI among RTX Spark laptop models. Nvidia had previously said RTX Spark laptops and compact desktops will be available this fall from ASUS, Dell, HP, Lenovo and Microsoft. For developers, Microsoft also offers a Surface RTX Spark Dev Box with 128GB of unified memory.
Nvidia positions RTX Spark as providing 1 petaflop of AI performance for what it calls the world's first Windows PCs purpose-built for personal agents, as noted on the Nvidia blog. Microsoft scheduled a special Windows and Surface event on October 7 covering the Surface Laptop Ultra and RTX Spark.
The broader context here is memory capacity as product segmentation. The jump from 24GB at entry to 48GB and 128GB at the high end is not about multitasking headroom in the conventional sense. For local agent runtimes, unified memory determines what can stay resident without spilling to storage or to the cloud, including model weights, KV cache, retrieval indexes and concurrent background agents. Nvidia and Microsoft are effectively selling residency.
In my view, the pricing will limit initial uptake to developers and technical professionals who already pay workstation prices to keep sensitive inference local. That is a rational beachhead. Low latency, offline operation and data locality matter more for agentic workflows than for prior assistant features. If the software delivers persistent, background agents that can use local tools reliably, the larger memory SKUs will look less like luxury and more like minimum viable configuration. The long arc favors more local capability, and this generation tests how much buyers will pay to get it early.
