Lei Jun, founder, chairman, and CEO of Xiaomi, has just introduced two prototypes powered by the next-generation Xuanjie chip, including a high-speed AI demonstration terminal and a personal AI supercomputer terminal.
Xuanjie O100 prototype: A high-speed AI demonstration terminal designed for large models running on the device. It uses collaborative computing between the Xuanjie O3 and O100 chips, removes the traditional smartphone imaging module, completely redesigns the motherboard, and adds an active air-cooling system capable of dissipating up to 10 watts of power. With Xiaomi MiMo deployed on the device, it achieves an exceptionally high measured inference speed of 295 Tokens per second.
Xiaomi AI Cube prototype: The exterior uses a unibody aviation-aluminum design with 33,874 precision CNC ventilation holes, enabling sustained high-performance output of 150 watts. Inside, the Xuanjie O3, O100, and D100 are fully integrated, along with 80GB of unified memory. It can deploy a 120B large-parameter model and a 3B on-device model, allowing users to switch between fast and slow systems. Whether for frontend development or complex programming tasks, this personal AI supercomputer terminal can execute them quickly and accurately.
According to the roadmap, Xuanjie O3 will launch in the Xiaomi 18 Fold, while the other two chips will enter commercial use in 2027.
ITHome also photographed the two prototypes at Xiaomi's Xuanjie chip technology briefing. More images are shown below:
