<p>Multi-view subspace clustering (MVSC) is a crucial and widely used method, as it effectively combines multi-view data to uncover significant patterns. However, its high computational cost limits its application to large-scale datasets. To address this, Anchor-based MVSC methods have been proposed, with adaptive anchor point selection being the most popular. However, these methods still have room for improvement in three areas: (1) most adaptive anchor selection methods learn anchors and construct anchor graphs in the original space, where redundancy and noise are inevitable, especially in high-dimensional data. (2) Adaptive anchor selection methods often result in anchors that lack representativeness and are unevenly distributed. (3) Existing methods always fail to consider the overall structure of the dataset during anchor learning. To address these issues, we propose a new multi-view subspace clustering method called Large-Scale Multi-View Subspace Clustering with Latent Centroid Anchor Guidance (MVSC-LCA). Specifically, we integrate latent intact space learning, anchor search, and anchor graph construction into a unified framework, achieving mutual reinforcement and joint optimization. Unlike existing methods that learn anchors and construct anchor graphs in the original space, we perform anchor learning and graph construction in the latent intact space. Additionally, we design a novel regularization term that drives the anchor set towards the latent centroid, thereby fully leveraging the overall structure of the dataset during the anchor learning process, making the anchor set more representative. Finally, extensive experiments on eight benchmark datasets with large-scale data demonstrate the effectiveness and superiority of MVSC-LCA existing state-of-the-art clustering methods.</p>