CES 2026: On-Device AI Front and Center, from Second-Gen AI PCs to Ryzen AI Halo

CES 2026 in Las Vegas was the edition where AI running directly on the device — with no data going to the cloud — stopped being a promise and became an ordinary catalogue feature.

On the PC side, the second generation of Copilot+ machines arrived from three different silicon vendors. Qualcomm showed the Snapdragon X2 Plus with an 80 TOPS neural processor, with X2 Elite variants reaching 85. AMD announced Ryzen AI 400, with a 60 TOPS NPU and Radeon 800M integrated graphics. Intel brought Core Ultra series 3, its first compute platform built on the new 18A manufacturing process. In practice: transcription, live captions, translation, and image generation all keep working offline.

The most interesting announcement for anyone who wants to run genuinely large models at home came from AMD: Ryzen AI Halo, designed to hold models up to 200 billion parameters using 128 GB of shared memory. That is the part that actually moves the bar for local AI.

NVIDIA, by contrast, spent CES 2026 on the data centre and the car: the Rubin platform, now in production, and Alpamayo, a family of open models for autonomous driving. Its desk-side supercomputer, DGX Spark, had already been on sale since October 2025 and only picked up a performance bump and some new capabilities at the show.

LG and Samsung demonstrated home robots that map a room and avoid obstacles, plus TVs with on-board AI processing for picture, sound, and voice control — still demos rather than shelf products.