Xiaomi, AI Cube prototype: Multi-chip local LLM powerhouse (me.mashable.com)

Xiaomi has unveiled its prototype AI Cube, a compact desktop device designed to function as a powerful on-premise supercomputer capable of running complex large language models without relying on external cloud servers. Powered by Xiaomi’s custom silicon, including three interconnected processors—XRING O3, XRING D100, and XRING O100—the AI Cube boasts impressive specs like a memory bandwidth of 1.22 TB/s, support for up to 160GB unified memory, and the ability to locally deploy AI models with up to 120 billion parameters. This innovation challenges existing market leaders like Apple's Mac Studio, aiming to provide a more efficient, locally-controlled solution for AI processing. The significance of the AI Cube lies in its potential to disrupt the landscape of local AI computing, traditionally dominated by high-cost cloud solutions. By achieving such high computational capabilities within a compact form factor and maintaining low power consumption (150 watts), Xiaomi addresses key challenges in running frontier AI models locally. This move positions the AI Cube not just as an alternative for enterprise and creative professionals, but also as a strategic asset in Xiaomi's broader smart ecosystem ambitions, integrating smart home and automotive applications. With plans for commercial availability by 2027, the AI Cube could reshape competitive dynamics by offering an attractive, privacy-focused solution that mitigates reliance on cloud infrastructure.