Applied Spatial Analysis and Policy География, планирование и развитие

Artificial Intelligence and Carbon-Constrained Energy Efficiency: Direct Effects, Digital Infrastructure Thresholds, and Spatial Effects

Xu, Bo, Li, Ruifa, Zou, Gang, Hu, Yingde
Springer Nature · 2026-09-02 · DOI 10.1007/s12061-026-09986-x

Аннотация

Artificial intelligence is reshaping urban energy systems at an unprecedented pace, yet its net contribution to carbon-constrained energy efficiency remains an unsettled empirical question. The mechanisms of impact, the role of digital infrastructure, and the spatial patterns of energy savings constitute a critical but underexamined research frontier. Drawing on panel data from 297 Chinese prefecture-level cities over 2011–2024, we employ a parallel multiple mediation model, a panel threshold model, and a spatial Durbin model within a unified framework. We present our main findings as follows. First, AI development exerts a positive effect on carbon-constrained energy efficiency. Raising AI_level by one standard deviation improves carbon-constrained energy efficiency by roughly 10.3% of the sample mean. Second, industrial upgrading and green innovation together explain 38.2% of the total effect, with the remaining 61.8% attributed to the direct effect of AI. Third, digital infrastructure exerts a single threshold effect. The positive association between AI and energy efficiency is stronger in cities with more developed digital infrastructure. Fourth, AI development contributes to energy efficiency gains in neighboring cities via spatial spillovers, yet the size of these indirect effects is modest. The spatial distribution of AI exhibits high persistence, with annual transition probabilities below 4%. Fifth, the positive association is heterogeneous across regions, policy environments, and socioeconomic conditions. This study advances the literature by quantifying the relative importance of direct versus indirect pathways, identifying the threshold role of digital infrastructure, and documenting the spatial persistence and limited spillovers of AI development. These findings offer implications for differentiated regional strategies and infrastructure investment priorities.

Читать у издателя →