联合国大学报告：AI数据中心到2030年年耗电将达945 TWh，用水量相当于13亿人基本需求，土地占用超1.45万平方公里
英文摘要
The United Nations University Institute for Water, Environment and Health (UNU-INWEH) published a report quantifying the environmental footprint of AI’s energy use. It projects that by 2030, global data centers supporting AI will consume 945 TWh of electricity per year, use water equivalent to the basic annual needs of 1.3 billion people, and occupy over 14,500 km² of land. Training GPT-5 alone is estimated to require about 100 GWh of electricity, 1 billion liters of water, and 1.5 km² of land. The report highlights that inference accounts for 80–90% of AI’s total energy consumption. In 2025, global data centers already consumed 448 TWh of electricity, ranking as the world’s 11th largest electricity consumer if considered a single country.
中文摘要
联合国大学水、环境与健康研究所发布报告量化了AI能耗的环境足迹。报告预计到2030年，全球AI数据中心每年将消耗945 TWh电力，用水量相当于13亿人一年的基本生活用水需求，土地占用面积将超过14500平方公里。仅训练GPT-5就估计需要约100 GWh电力、10亿升水和1.5平方公里土地。报告指出，推理环节占AI总能耗的80%至90%。2025年全球数据中心已消耗448 TWh电力，若视为一个国家将排名全球第11大电力消费国。
关键要点
By 2030, AI data centers are projected to consume 945 TWh of electricity annually, with water usage equal to the basic needs of 1.3 billion people and land footprint exceeding 14,500 km².
到2030年，AI数据中心预计年耗电945 TWh，用水量相当于13亿人基本需求，土地占用超1.45万平方公里。
Training GPT-5 consumes roughly 100 GWh of electricity, 1 billion liters of water, and 1.5 km² of land, but inference dominates AI’s total energy use at 80–90%.
训练GPT-5需约100 GWh电力、10亿升水和1.5平方公里土地，但推理环节占AI总能耗的80%-90%。
In 2025, global data centers used 448 TWh of electricity, enough to rank as the 11th largest electricity consumer worldwide.
2025年全球数据中心消耗448 TWh电力，足以成为全球第11大电力消费国。
The report links each kWh of AI electricity to three environmental footprints: carbon from energy production, water from generation and cooling, and land from infrastructure and resource extraction.
报告将每千瓦时AI电力关联三种环境足迹：能源生产的碳足迹、发电与冷却的水足迹，以及基础设施和资源开采的土地足迹。