Xiaomi has introduced a prototype mini-PC called the AI Cube, built to run its large language models locally.
The device was unveiled at the XRING Chip Technology Communication Conference and is powered by three proprietary chips: XRING O3, XRING O100, and XRING D100.
The XRING O3 serves as the main AI SoC, featuring a 10-core all-big-core CPU, a 16-core G2-Ultra NX GPU, and a 200 TOPS low-power NPU.
The XRING O100 is a high-bandwidth AI acceleration chip built on 6nm 3D wafer-level stacking, with 28,672 effective data lines and near-memory computing bandwidth up to 1.22TB/s.
The XRING D100 is a 3nm chip designed for intelligent driving workloads, with a 20-core high-performance CPU, a 16-core NPU, and support for up to 160GB of memory.
The AI Cube features an aerospace-grade aluminum unibody casing with 33,874 CNC precision cutouts and can sustain up to 150W of continuous power.
The hardware is intended for local deployment of large models ranging from 3B to 120B parameters, with a toggle to switch between fast and slow system modes.
This puts it in similar territory to the Tiiny AI Pocket Lab, which also runs 120B-parameter models locally, but Xiaomi uses its own chips instead of third-party silicon.
The AI Cube remains a prototype, with no release date or price announced yet.
