Thermally induced deviations are a leading cause of geometric error in machine tools and present persistent challenges for high-precision manufacturing. This review explores how digital twins (DTs) have advanced thermal management by integrating real-time data, advanced simulation models, and adaptive control strategies. Recent developments in thermal digital twins are categorized as physics-based, data-driven, hybrid, or adaptive modeling approaches. The review highlights their contributions to the prediction and compensation of thermally induced deviations. The emphasis is on model order reduction, machine learning integration, and sensor technologies. Finally, we discuss open challenges, such as model generalization, calibration robustness, early-stage design integration, and lifecycle-wide applications, to guide future research.