The cheapest way into Nvidia’s RTX Spark laptops costs $2,599. For that money, buyers get a machine that is supposed to run big AI models without ever calling a cloud server.
The laptops reached stores worldwide on October 16, 2026, with models from ASUS, Dell, HP, Lenovo, Microsoft and MSI. Each one is built around a single N1X superchip that pairs an Arm-based Grace processor with a Blackwell graphics unit. Both halves draw on the same pool of up to 128GB of unified memory, and Nvidia rates the platform at up to 1 petaflop of FP4 AI performance. The machines run Windows 11.
Not every model has 128GB
That memory figure applies only to the top configuration. Shattered, which tracked launch-day listings, counted 17 laptop configurations at pre-order, with memory options of 24GB, 32GB, 48GB and 128GB. Seven of the 15 it examined used the smaller chip. There is a pricing wrinkle too: Nvidia’s press release gives $2,599 as the starting price, while a blog post a day later says “starting at $2,399.” Tech Times put the first-wave range at roughly $2,599 to $6,700.
ShiftDelete, a Turkish tech outlet, described the hardware as 14 mm thick and about 1.4 kg, and said the shared memory lets large language models and heavy 3D projects keep running on battery. Creators can work with raw 4:2:2 camera footage or 90 GB 3D scenes.
The local AI side
Nvidia is pairing the hardware with models built for the machine. Nemotron 3.5 Bolt 74B-A9B 24GB is a 74-billion-parameter model tuned for tool calling and multistep tasks. A smaller sibling, Nemotron 3.5 Bolt E4B 4GB, is meant for in-game characters. By Nvidia’s own account, it scores about 40% higher in average quality than its predecessor while producing up to 80% fewer tokens than peers. Those numbers come from Nvidia’s own evaluation, so treat them accordingly.
ShiftDelete added that the company has released Nemotron 3.5 Transcribe, an open-weights tool for transcription, speaker diarization and translation across 40 languages, available on Hugging Face. The Liquid Foundation Models family, LFM3, now runs natively on RTX Spark too, aimed at local agents handling chores such as email automation and document analysis. A new control in the NVIDIA App lets users decide whether gaming or background compute gets priority on the GPU.
What reviewers found
The first production reviews split down the middle. The graphics half impressed. The Tech Times roundup of reviews, as summarised by Shattered, reported 6,060 for the N1X in 3DMark Time Spy Extreme, against 4,689 for AMD’s Ryzen AI Max+ 395. The processor cores drew complaints: the roundup said they are Arm’s 2024-generation designs and that single-thread results trail Apple’s M5 generation.
Battery life was the bright spot. HotHardware reportedly measured 14 hours and 35 minutes of video playback on the Surface Laptop Ultra.
Memory bandwidth is the open question for anyone planning to chat with local models all day. Reported figures range from 273GB/s to about 300GB/s, and Nvidia has not published a laptop-specific number. Neither Nvidia nor Microsoft has released a head-to-head decode-speed comparison against Apple hardware.
Call of Duty is not coming to RTX Spark until 2027.
Sources
- RTX Spark Review: 128GB Memory Wins, CPU Lags [2026]
- NVIDIA Launches RTX Spark Platform for AI-Powered Laptops
