China Debuts First Mobile Vehicle-Mounted 'Token Factory' for AI Inference
China has unveiled its first mobile vehicle-mounted "token factory," a containerized computing unit designed to provide flexible AI inference power. The system integrates domestic hardware and energy management to dynamically match AI workloads with renewable energy sources. This project explores a new "computing-electricity synergy" model, allowing mobile data centers to utilize cheap, excess renewable energy directly at substations. By doing so, it can reduce overall computing costs by approximately 25% while helping power grids absorb surplus green energy. Housed in a standard 40-foot container, the unit can deliver 50 to 100 PFLOPS of computing power under a 120 kW power supply, which is enough to run two domestic large language model inference tasks. It supports up to nine racks and can be fully deployed and networked within 6 to 8 hours.
## BACKGROUND
"Computing-electricity synergy" refers to the coupling and dynamic matching of computing resources with power supply to optimize energy efficiency. Meanwhile, a "token factory" in AI refers to infrastructure optimized specifically for generating tokens during AI inference workloads, where throughput and cost-efficiency are critical.