Nvidia’s RTX Spark turns the Windows laptop into a local AI agent

After a week of testing, Nvidia’s Blackwell-powered RTX Spark looks less like a MacBook Pro rival than a new kind of Windows computer built to run AI agents, creative tools and games locally.

After a week with Nvidia’s RTX Spark on a Surface Laptop Ultra, one conclusion stands out: this is not a conventional attempt to dethrone Apple’s MacBook Pro, but something stranger and potentially more consequential. Tom’s Guide describes the machine as a new kind of Windows PC built around locally hosted AI, with 128GB of unified memory and Nvidia’s Blackwell-based silicon designed to keep personal agents running on the device itself. That framing aligns with Nvidia’s own launch messaging, which positions RTX Spark as a class of Windows PCs built for on-device AI work rather than cloud dependence.

The first practical test was not glamorous. The reviewer used an agentic setup to play The Sims 4, build a house and find a job, only to watch the system become confused, stall and eventually start a kitchen fire. The lesson was less about failure than about workflow: the article argues that local AI agents respond better to patience, structure and reinforcement than to the habits most people have developed with browser-based chatbots. In that sense, the machine demands a different style of interaction, closer to coaching than searching.

By the second day, the device had been turned into a kind of automated assistant. The reviewer linked the agent to Google Workspace and a browser-control script, then used it to sort emails, remove junk, draft replies and assemble a workday to-do list. Tom’s Guide says the experience highlighted a clear speed advantage over trying the same Hermes Agent workflow on a MacBook Pro, while Nvidia’s platform advantage stems from CUDA support and the wide use of Nvidia-optimised open-source models in local AI. In practice, the benefit was not only raw performance but also the ability to offload repetitive administrative work before the day began.

The machine’s creator credentials were equally striking. Tom’s Guide reports that ComfyUI image generation, fed by a rough sketch of Medusa, produced a stylised cartoon image in about 20 seconds, followed by an image-to-video sequence. Nvidia has said RTX Spark systems are intended for personal AI agents, content generation and local automation, and the review suggests those claims are not just theoretical. The article also says the system outpaced AMD Ryzen AI Max+ in the reviewer’s testing by roughly two to three seconds per image generation, with the MacBook Pro trailing as well.

Gaming, too, looks unusually ambitious for an Arm-based system. The reviewer found that a large GPU appears to solve much of the compatibility problem that often dogs Arm laptops, with 007 First Light, NBA 2K27 and other titles running well at high settings. When Red Dead Redemption 2 refused to launch because of a compatibility issue, the reviewer asked the local agent to intervene; according to Tom’s Guide, it rewrote configuration code to bypass the block and get the game running. The same GPU muscle also improved video calls, with Nvidia Broadcast being used to upscale a webcam feed and clean it up in real time.

Still, the article stops short of calling the device a MacBook Pro killer, and for good reason. The battery lasted around 10 hours in the reviewer’s browser test, which is respectable for a Windows machine but still below Apple’s best-in-class endurance, while Apple’s M5 Pro and M5 Max chips remain faster on CPU-heavy tasks. The more important point, Tom’s Guide argues, is that comparing it directly with a MacBook misses the point: RTX Spark is trying to define a new category of computer, one that treats local AI as a first-class feature rather than an add-on. Nvidia and Microsoft have both framed the platform in similar terms, saying it is meant to bring agentic AI into native Windows workflows with up to 128GB of unified memory and full CUDA and RTX support.