The continued growth of computing demand driven by artificial intelligence (AI) has intensified energy consumption pressures and operating cost constraints in computing infrastructure. Conventional approaches dominated by hardware-level energy saving and static policies are increasingly insufficient for energy-efficiency governance under complex and dynamic operating conditions. This study reviews the research progress in AI-enabled energy-efficiency optimization for computing infrastructure from three perspectives: infrastructure layer, scheduling and operating layer, as well as cross-system coordination layer. At the infrastructure layer, we summarize key techniques including energy- and carbon-efficiency metric systems, power and thermal behavior modeling, and intelligent cooling control, and analyze the extension of life-cycle assessment to computing scenarios. At the scheduling and operating layer, we review resource scheduling and power management methods based on workload and energy prediction, reinforcement learning, and multi-objective decision making, emphasizing integrated trade-offs among energy consumption, performance, and carbon emissions under service-quality constraints. At the cross-system coordination layer, we survey research and engineering practices on carbon-signal-aware scheduling, cross-region workload migration, and coordinated optimization between computing and energy systems, highlighting the importance of simulation-based validation, phased rollout, and controlled comparisons for practical deployment. Furthermore, we summarize major challenges related to data and model trustworthiness, system stability and controllability, evaluation standards and verifiable benchmarking, and techno-economic considerations and policy coordination. Finally, we outline future directions including chip-to-system integrated energy-efficiency management, energy‒carbon co-optimization frameworks, energy-efficiency governance for edge and distributed computing, development of standardized evaluation frameworks, and integration between simulation and engineering practices, thereby providing a systematic reference for research and deployment of energy-efficiency optimization in computing infrastructure.