Hardware

Nvidia Debuts RTX Spark Superchip in New AI PCs

Hardware makers at the IFA 2026 show unveiled the first laptops and mini PCs powered by Nvidia's RTX Spark superchip, enabling developers to run complex agentic AI workflows entirely locally.

WIRED AI3 days agoHardware
Image: WIRED AI

At the IFA 2026 tech show in Berlin, hardware manufacturers showcased the first computers powered by Nvidia's Arm-based RTX Spark superchip. This system-on-a-chip integrates a Grace CPU featuring up to 20 cores with a Blackwell RTX GPU boasting up to 6,144 cores. This architecture shares memory across the chip, delivering local graphics performance estimated to sit between a discrete laptop-grade RTX 5070 Ti and RTX 5080. The design allows for ultra-thin form factors, such as the newly revealed Lenovo Yoga 9n 2-in-1. This 16-inch laptop measures 0.69 inches thick, features a 2880 x 1800 resolution 120-Hz OLED touch display, a 9.2-megapixel webcam, and supports up to 64 GB of memory.

Lenovo also introduced the Yoga Pro 9n, a 15-inch model measuring 0.66 inches thick with a 2560 x 1600 OLED screen, up to 128 GB of memory, and a Force Pad touchpad that supports the Yoga Pen Gen 2. Other upcoming RTX Spark laptops include the Dell XPS 16, Asus ProArt P16, Microsoft Surface Laptop Ultra, and the 14-inch HP OmniBook X 14. For desktop users, the Acer SFF RTX Spark and Asus ProArt Mini PC pack the same superchip and up to 128 GB of memory into Mac Mini-sized chassis, promising a petaflop of AI performance.

Nvidia faces competition from AMD, whose Ryzen Max+ Pro 495 chip powers the Framework Desktop and Lenovo's new ThinkCentre X Ultra. The ThinkCentre X Ultra offers 128 GB of memory for agentic AI models at an expected starting price of $3,699. While Nvidia has not released official pricing for RTX Spark systems, they will likely command a premium. For comparison, Apple's M5 Pro MacBook Pro with 24 GB of RAM starts at $2,349, while a 128-GB model costs $6,139.

For AI practitioners, these local systems represent a major shift toward decentralized development. Having up to 128 GB of unified memory and a petaflop of local performance allows developers to run complex agentic AI workflows, such as OpenClaw, entirely on-device. This local execution eliminates latency, cuts cloud computing costs, and ensures strict data privacy when handling sensitive financial or personal information.

This is our own summary of reporting by WIRED AI

More in Hardware