Against the backdrop of the rapid development of the digital economy, internet dispatching platforms such as food delivery and ride-hailing services have become key urban infrastructure. However, they generally face the core contradiction between dynamic demand fluctuations and rigid service capacity constraints. This paper decomposes the dispatching system into a two-stage closed-loop structure of “waiting and service”. Combining queuing theory principles, AI empowerment characteristics, and introducing user loss aversion psychology and reference utility features, a configuration model covering basic capacity and safety capacity is constructed to explore optimal capacity strategies under profit-oriented and welfare-oriented orientations. Numerical examples verify the model’s effectiveness. Results show that the optimal capacity consists of basic capacity and safety capacity, with the two-stage safety capacity maintaining a specific matching ratio. Moreover, AI empowerment reduces the basic capacity demand in the waiting stage but requires simultaneous optimization of service stage capacity to avoid new bottlenecks. Consequently, platform positioning and user behavior characteristics significantly affect capacity configuration efficiency. The research conclusions provide theoretical support and practical guidance for dispatching platforms to achieve refined operations and balance efficiency with user experience.