As lithium batteries advance toward higher energy densities, developing electrolytes that remain stable under high-voltage conditions has become a critical bottleneck. However, electrolytes encompass a vast structural design space, complex descriptor systems, and multidimensional performance evaluation metrics, making traditional research and development both time-consuming and costly. Machine learning has opened a data-driven route for addressing pattern recognition, anomaly detection, and simulation for the accelerated discovery of high-voltage electrolytes. Here, we trace key milestones in the evolution of machine learning and, on this basis, introduce an AI for batteries (AI4B) paradigm tailored to electrochemical energy storage. AI4B emphasises the synergistic exploitation of data and algorithmic innovation to build cross-scale, multiphysics models that connect molecular-level descriptors with macroscopic interfacial phenomena, enabling a more realistic and quantitative representation of complex electrolyte reaction mechanisms. We further summarize the major advances in AI-assisted high-voltage electrolyte design and discuss complex interfacial issues, design strategies, and future research directions.