Point cloud transformer is a hot topic in the field of point cloud understanding. Some existing studies have explored the possibility of the combination of point cloud deep learning and transformer models, with the outstanding results. However, in specific tasks, such as point cloud classification and segmentation tasks, there is still room for improvement in accuracy. Motivated to improve the capture of semantic and discriminative features of point clouds by transformer, we propose a novel point cloud transformer-InfoPCT, which combines the self-attention mechanism and mutual information maximization that facilities deep learning on point clouds. Moreover, a mutual information maximization-based loss function is employed, so that the point cloud transformer can be trained simultaneously with the mutual information maximization module without pre-training. Compared with both CNN-based and transformer-based existing works, our superior results indicate the efficacy of InfoPCT.