Abstract: The traditional physical modeling methods for CNC machine tools, plagued by low computational efficiency and insufficient accuracy in nonlinear dynamic response modeling, struggle to meet the real-time decision-making requirements of intelligent manufacturing. This study focuses on the evolutionary trajectory of surrogate modeling techniques and their practical applications in CNC machine tools. It systematically explores the technical iteration from single-fidelity models to multi-fidelity hybrid modeling, thoroughly investigating the paradigm shifts among traditional mathematical methods, machine learning, and deep learning-based surrogate models. The findings reveal that the current surrogate modeling framework has established a robust paradigm addressing multidimensional complex working conditions, with key advancements in three critical capabilities: multi-source data fusion mechanisms, adaptive sampling strategies, and multi-objective optimization architectures. These breakthroughs fully unlock the potential for synergistic optimization of machining precision and energy efficiency. The research highlights the core enabling role of surrogate models in the intelligent transformation of CNC machine tools, offering theoretical insights and practical guidance to advance digital engineering in the field of machine tool technology.