As data centres continue to expand and energy consumption rises, improving energy efficiency and achieving low-carbon operation remain major challenges. In this context, accurate prediction of temperature and thermal load is regarded as a core element of intelligent management. In this context, research progress on temperature and thermal load prediction in data centres was systematically reviewed. The research background and significance were summarised, and the coupling mechanisms among energy consumption, cooling, and thermal environment that influence efficiency optimisation were analysed. Core concepts related to temperature and thermal load prediction were introduced, and existing work was classified into physics-driven modelling, which is characterized by mechanistic consistency and physical interpretability; data-driven approaches, which rely on monitoring data for efficient prediction; and hybrid modelling, which integrates both perspectives to balance accuracy, generalisation, and interpretability. The differences and complementarities between steady-state and transient prediction methods were compared, and their application value in energy-efficiency assessment, cooling optimisation, dynamic regulation, and risk warning was discussed. Future trends were highlighted, indicating that multi-scale coordination, cross-scenario adaptation, and integrated prediction-control were expected to become important research directions. These findings are expected to provide theoretical support and methodological guidance for the development of intelligent and low-carbon data centres.