Deep subspace clustering has emerged as a powerful extension of conventional subspace clustering techniques, demonstrating superior performance across various applications. Nevertheless, current deep subspace clustering methods remain fundamentally limited by their coupled parametric self-expression layers that inherently require full-batch optimization, thereby restricting their scalability for large datasets. To address the limitation, we propose Anchor-Guided Scalable Deep Subspace Clustering (ADSC). Our method effectively decouples the self-expression process into two complementary components, i.e., relational prediction through a neural network head and adaptive anchor learning based on exponential moving average and memory bank. We validate ADSC through extensive experiments on twenty tabular datasets. Experimental results demonstrate that our method outperforms many state-of-the-art clustering algorithms in terms of accuracy, robustness, and scalability. Our framework establishes a new paradigm for efficient and scalable deep subspace clustering without compromising clustering accuracy.