Artificial Intelligence and Green Water-Use Efficiency in China: Provincial Evidence from 2011–2024

Whether artificial intelligence (AI) alleviates or aggravates resource constraints is an open and increasingly consequential question: AI carries a substantial energy and water footprint, yet it may also raise resource-use efficiency through prediction, intelligent optimization, and better allocation. Using a balanced panel of 30 Chinese provinces over 2011–2024, this study measures green water-use efficiency (GWUE) with a Super-SBM model and estimates the association between AI development and GWUE using two-way fixed effects, complemented by an instrumental-variable strategy and interpretable machine learning. We find that AI development is associated with a significant improvement in GWUE: a one-standard-deviation increase in the AI index corresponds to a 3.8% rise in GWUE. The relationship is consistent with two channels—green technological innovation and environmental-governance capacity. The association is stronger where per-capita water use is higher, and differs across regions of differing industrial water dependence and computing infrastructure. XGBoost–SHAP analysis further suggests a nonlinear transition around an AI development level of approximately 0.12, beyond which AI’s marginal contribution to predicted GWUE becomes increasingly positive. These findings indicate that AI does not automatically become a green technology; its resource-efficiency dividend depends on the resource and technological environment in which it is deployed.