Nvidia Says Local AI Is Here as RTX Spark Laptops Aim for Fall

SANTA CLARA, Calif. - Nvidia’s AI product leadership is arguing that personal, on-device AI is no longer a lab experiment, pointing to DGX Spark desktops already in homes and a coming wave of Windows RTX Spark laptops from major PC makers this fall, according to a Tom’s Guide interview with Adel el Hallak published Sept. 20, 2026.

El Hallak, described in the interview as Nvidia’s VP of product (and elsewhere as senior director of product management for Nvidia AI), said he keeps a DGX Spark at home and often runs overnight AI jobs on it. He told Tom’s Guide he feels guilty if he forgets to give the machine work before bed, noting that it is quiet and already plugged in.

Photo: NVIDIA

Photo: NVIDIA

His case for local AI is ownership and privacy. A model that runs on hardware you own keeps data on the device and avoids a monthly cloud subscription, he argued. Nvidia’s DGX Spark currently retails for about $4,699. El Hallak contrasted that one-time price with a ChatGPT Plus plan at $20 a month, or $240 a year, while stressing that the comparison is not apples to apples: frontier cloud models remain more capable than the open-source stacks most buyers would run locally on a Spark.

For many everyday tasks, he and the interviewer framed a local box with about 128GB of unified memory and Nvidia’s CUDA stack as enough, including document summary, email drafts, personal file search, and coding assistance. Unlimited local queries avoid per-token billing and work without an internet link once the model is on disk.

Photo: NVIDIA

Photo: NVIDIA

The ecosystem around DGX Spark is also shifting toward consumers. Tom’s Guide notes that Perplexity’s Portable Computer packages a local model, agent tools, app connectors, and a sandboxed runtime on DGX Spark hardware, with cloud handoff only when the user allows a step the local model cannot finish. El Hallak said harness providers are replacing command-line and pip-install setups with something closer to a Windows-style graphical install. “You’re not doing installs via a command line or a pip install,” he said. “They’re making it easier - like the old days of using Windows and setting up an application. They’re bringing it back to a GUI.”

Photo: Microsoft

Photo: NVIDIA

The larger consumer bet, in his telling, is RTX Spark notebooks that pack the same GB10-class silicon into Windows 11 laptops expected this fall from Dell, HP, Lenovo, Asus, MSI, and Acer, with Microsoft building a Surface Laptop Ultra around the chip. Those machines are not Linux-only appliances like the DGX Spark desktop. They are meant to game, create, and run ordinary Windows workloads while holding roughly 128GB of unified memory for large local models.

“Local AI is just going to bring more tokens, and tokens equate to intelligence,” el Hallak told Tom’s Guide. “When you pair intelligence with a harness, you’re able to do things, and then the runtime secures it.” He also described the destination more casually as “data centers in the house.”

Nvidia has not, in the Tom’s Guide piece, set public retail prices for partner RTX Spark laptops. Availability is framed as this fall from the named PC makers, with Surface Laptop Ultra as Microsoft’s premium entry on the same silicon.