Xiaomi AI Cube – desktop terminal with three proprietary chips for local AI computing

Edge AI terminals enable running large language models directly on user devices without relying on cloud servers, increasing privacy, reducing latency, and enabling operation without internet connectivity. Such devices are suitable for applications where local processing of sensitive data is required or where stable internet connection is unavailable.

Custom chips allow optimization for specific tasks and models, better energy efficiency, and lower latency compared to general-purpose processors. Manufacturers gain independence from suppliers and can achieve better performance-per-watt ratios, which is particularly important for mobile and desktop devices.

The larger 120-billion-parameter model provides higher answer quality and better understanding of complex questions, but requires more computational power and memory. The smaller 3B model is faster and less resource-intensive, suitable for simple tasks and quick response times.