The traditional subspace interference alignment (IA) schemes in the multicell multiuser multiple-input-multiple-output (MIMO) network align interference into the multidimensional subspace to reduce the dimension of the interference space, but the choice of subspace is random, and the difference between users is not considered. Such schemes may leave large intercell interference when the difference between users is large. To solve this problem, this article proposes the IA scheme based on subspace differentiation for multicell multiuser MIMO uplink channels. First, based on the transmit power and the distance to base stations in other cells, each user analyzes the interference value it will cause. The interference value of each user is represented by actual interference power (AIP). Within each cell, the users are reordered according to the AIP. When performing interference reduction, the signals of the users with large interference are merged into the subspace composed of the signals of the users with small interference, which can further reduce the final intercell interference. The multidimensional merge weight is assigned according to the AIP. The smaller the interference value of the dimension, the more interference it merges. The experimental simulation data show that the scheme implemented in this article can improve the proportion of the desired signals in the received signals of the receivers, thereby improving the overall system rate. When the user difference in transmit power and distance is large, the proposed scheme has a particularly obvious advantage.