With the advent of deep learning, state space models have made significant progress in point cloud analysis tasks. However, existing methods mainly focus on the spatial structure of points and pay little attention to the semantic information of point clouds and the higher-order hyperedge relationships between points. To address this issue, this paper proposes a hybrid attention mechanism-based Mamba architecture, PointMHA, which combines Transformers’ expressive power with Mamba’s efficient modeling characteristics to improve point cloud classification performance. The core design of the model is a hybrid feature aggregator that combines two attention mechanisms: content-based attention utilizes the locality of points in the feature space to cluster sampled points with similar features into the same class. In contrast, hyperedge attention dynamically adjusts the importance of each hyperedge through learnable hyperedge attention to capture high-order structural relationships between points, thereby improving feature representation capabilities. The final fused features are input into the Mamba module for sequence modeling, which is used for downstream classification tasks. Experimental results show that PointMHA achieves classification accuracy rates of 93.6% and 89.5% on the ModelNet40 and ScanObjectNN datasets, significantly outperforming existing state-of-the-art methods.