Integrated sensing and communication (ISAC) and edge intelligence are essential components for the next generation wireless networks. ISAC provides a spectrum-efficient paradigm to collect and transfer sensing data for edge intelligence. Multi-station sensing, as a cooperative sensing scheme to perform the sensing tasks towards a number of targets, can improve the sensing efficiency and flexibility. In this paper, we propose an edge intelligence oriented ISAC via a multi-cell cooperative approach, in which multiple ISAC stations sense a group of targets with the sensing scheduling, and offload the sensing data to an edge server with sufficient computing capacity for model training. Specifically, we investigate the joint optimization of the beamforming for radar sensing, the beamforming for offloading transmission and the ISAC stations' sensing scheduling, with the objective to minimize the overall power consumption of all ISAC stations. To tackle the non-convexity of the formulated problem, we firstly propose a layered algorithm to decompose the problem into a top-layer problem that optimizes the sensing scheduling and a bottom-layer problem that optimizes the beamformings for radar sensing and offloading. Then, the bottom-layer problem is further decomposed with the block coordinate descent (BCD) method and solved efficiently. Numerical results demonstrate the performance advantage of our proposed scheme of edge intelligence oriented ISAC with multi-cell cooperative sensing and the effectiveness of our proposed algorithm.