The unstructured, high-dimensional nature of point clouds poses challenges for effective feature extraction and classification. While transformer-based architectures excel at modelling complex spatial dependencies, they require substantial computational resources and large-scale pretraining. In contrast, the Mamba architecture, leveraging State Space Models (SSMs), offers greater computational efficiency and strong sequence modelling capabilities. However, its inherent linearity and temporal bias limit its ability to capture intricate spatial relationships essential for point cloud analysis. To address these limitations, we propose the Hierarchical Spatial Mamba Framework (HSMF), a novel architecture designed to enhance spatial feature learning for point cloud classification. HSMF integrates an Integrated Spatial Representation Module (ISRM) that systematically captures multi-scale geometric features at the surface, edge, and point levels, a Dynamic Scaling Learning Module (DSLM) that aggregates hierarchical spatial information through adaptive sampling, and a Mamba Backbone with Enhanced Spatial Locality that employs Morton curve-based reordering to preserve spatial coherence when mapping high-dimensional data into sequential formats compatible with SSMs. Experiments on ScanObjectNN and ModelNet40 demonstrate that HSMF achieves state-of-the-art performance, outperforming existing methods.