A policy determining method is disclosed, and includes: obtaining, based on a digital twin entity, first data and a label corresponding to the first data; and training, based on the first data and the label corresponding to the first data, to obtain an artificial intelligence AI model. The AI model is configured to obtain an air interface access policy and/or a user plane path selection policy of a plurality of terminal devices in a network or in an area. This helps resolve a problem of the air interface access policy and user plane path selection of the plurality of terminal devices in the network or in the area. This can further ensure that network energy consumption is the lowest when overall average service experience of the terminal devices reaches a specific threshold.