Nvidia’s RTX Spark superchip has moved from announcement to hardware. At IFA 2026 in Berlin, WIRED got a hands-on look at the Lenovo Yoga 9n 2-in-1, the first laptop built around the chip, alongside a wave of RTX Spark mini desktops from Acer, Asus, and Lenovo.

The pitch is the same one the industry has repeated since 2024: put enough silicon on the local machine that agentic AI workflows, the kind that read your email or move files on your behalf, never have to leave the device. WIRED frames the launch as an answer to rising demand for running tools like OpenClaw locally rather than through a cloud API.

The chip itself combines an Arm-based Grace CPU, described by Nvidia as having “up to 20 cores,” with a Blackwell RTX GPU carrying “up to 6,144 cores,” the whole thing fabricated as one system-on-a-chip. Those are the manufacturer’s own figures. Based on core count alone, WIRED’s Luke Larsen estimates the integrated GPU sits somewhere between an RTX 5070 Ti and an RTX 5080 in today’s discrete laptops, an inference rather than a benchmark result.

The Yoga 9n 2-in-1 is the one Larsen actually handled: a 16-inch OLED touchscreen at 2880x1800 and 120Hz, a 360-degree hinge, a six-speaker system, and a chassis measuring 0.69 inches, which puts it within 0.03 inches of a 16-inch MacBook Pro. Memory tops out at 64GB. A companion model, the 15-inch Yoga Pro 9n with configurations up to 128GB, was announced but was not available for WIRED to inspect in person, so its dimensions and touchpad-stylus feature are described from Lenovo’s materials, not from hands-on testing.

Nvidia did not disclose battery-life figures for either machine. WIRED notes only that the Arm-based SoC design should, in theory, be more power-efficient than a conventional discrete-GPU laptop, a plausible assumption rather than a tested claim.

More RTX Spark laptops are queued for this fall: the Dell XPS 16, Asus ProArt P16, Microsoft Surface Laptop Ultra, and a single 14-inch entry, the HP OmniBook X 14. On the desktop side, Acer’s SFF RTX Spark and Asus’s ProArt Mini PC both claim “a petaflop of AI performance,” a vendor figure WIRED reports without independent verification, and both top out at 128GB of memory. Lenovo’s AMD-powered ThinkCentre X Ultra, built on a Ryzen AI Max+ Pro 495 chip, starts at $3,699 for the base configuration, not the 128GB version.

Price is the detail that actually distinguishes this launch. A 128GB MacBook Pro currently runs $6,139, and even Apple’s cheapest 24GB M5 Pro configuration is $2,349. RAM has gotten more expensive across the industry this year, and nothing in this lineup suggests RTX Spark systems will undercut that.

The on-device AI PC pitch has been running for more than two years now, across three chip generations, without a single mainstream application that makes the NPU or the local GPU the actual reason to buy the machine. The question worth asking about this new wave of Spark hardware, before the RAM upgrade is paid for, is what workload runs on it that could not just as easily run in a browser tab connected to a cloud model.

Buyers evaluating an RTX Spark machine this fall should wait for independent battery and inference benchmarks rather than the vendor’s core counts, and should identify one concrete local-only workflow, not a hypothetical one, before paying the memory premium.

WIRED (Luke Larsen), published September 3, 2026.