Energy efficiency is a key challenge in next-generation Low Earth Orbit (LEO) satellite networks, especially for data-intensive applications in 6G non-terrestrial networks (NTNs). Traditional bit-wise transmission leads to excessive power consumption, making Semantic Communication (SemCom) a promising solution by transmitting meaningful information rather than raw data. This survey reviews energy-efficient SemCom for LEO satellites, covering semantic-aware compression, AI-driven power control, and joint communication-computation optimization. We also explore AI-based techniques such as reinforcement learning (RL) and federated learning (FL) for adaptive resource allocation. Additionally, we discuss trade-offs among energy efficiency, latency, and semantic accuracy, along with computational and security challenges. Finally, we highlight emerging trends, including quantum-assisted SemCom, Reconfigurable Intelligent Surfaces (RIS), and multi-modal task-oriented SemCom for 6G NTN networks.