As the provision of computing power expands from individual cards to clusters comprising tens of thousands of cards, and as the scheduling within the computing power network advances from mere networking to cross-domain collaboration, the industry's emphasis on cost-efficiency has grown markedly. Several securities firms have conducted analyses and highlighted that the primary industrial benefit of integrated scheduling within the computing power network is its ability to substantially decrease the unit costs associated with large-scale model training and inference, with a particular focus on reducing the production expenses of AI Tokens. This is achieved through the integration of idle computing resources, optimization of task distribution, and the synergy of green energy with heterogeneous technologies.

Presently, operators have established a sophisticated model for cost reduction in the realm of computing power network scheduling. Furthermore, collaboration among domestic providers of computing power hardware and software is intensifying. Qujing Technology and Moore Threads have recently formed a strategic alliance aimed at jointly constructing a high-caliber, domestically-produced AI Token factory, with pertinent solutions already in the production phase. Liu Xianhe, General Manager of the Token Business Division at Qujing Technology, remarked that as applications based on large models progress into the stage of widespread commercial deployment, the competitive edge in infrastructure is transitioning from the performance of individual cards to the system-level efficiency in Token production.