The invention discloses a deep learning computing system and method for cloud terminal edge computing fusion and relates to the fields of cloud computing, edge computing and artificial intelligence technology. According to the method, deep learning computing is distributed to a cloud terminal, a pipeline and an edge side, the cloud terminal is responsible for basic model training of a large historical data computing quantity, personalized model distribution is performed according to the demand of the edge side, nodes from the cloud terminal to the edge side deploy a deep learning model after learning, the deep learning model is used for completing inference, the edge side continuously feeds back an inference result, and then the inference result is uploaded to the cloud terminal to continuously optimize the model. Compared with the traditional mode that training and inference are both performed on the cloud terminal, in the whole deep learning computing process, the computing model iscontinuously optimized, personalized model distribution is performed according to the demand of the edge side, bandwidth is effectively utilized, network transmission efficiency is guaranteed, and real-time service execution efficiency is improved.