How does clean energy reshape the relationship between artificial intelligence and carbon emissions? Evidence from renewable and nuclear energy

The high energy demand of artificial intelligence (AI) poses a significant challenge to global decarbonization efforts. This study provides novel insights into AI technology's impact on carbon emissions by examining the energy rebound effect and clean energy as key mechanisms. Using panel data from 62 countries over 1995–2023, a nonlinear mechanism identification model is constructed. The empirical results uncover a stage-dependent relationship: AI initially increases carbon emissions, but reduces them as AI technology develops. This inverted U-shaped pattern is driven by applied rather than foundational AI, with turning points indicating that reducing total carbon emissions is more difficult than lowering carbon emission intensity. The energy rebound effect mediates the AI–emissions relationship—amplifying emissions in the early stages but weakening as AI technology develops. Clean energy moderates the relationship in source-specific ways: renewable energy advances the turning point at which AI contributes to carbon emission reductions, whereas nuclear energy mitigates the initial emission-increasing effects of AI. These findings suggest that realizing the carbon mitigation potential of AI requires a coordinated strategy—curbing energy rebound risks, optimizing clean energy portfolios, and tailoring governance to the evolving stages of AI development.