Abstract
Under global pressure to reduce carbon emissions, understanding how the level of artificial intelligence (AI) use affects corporate carbon emissions (CCE) is crucial for achieving a green transition. This study, based on provincial panel data from China, employs a combination of two-way fixed effects (TWFE) empirical analysis and structural equation modeling (SEM) to identify the nonlinear effects of AI use levels on corporate carbon emission intensity and their transmission pathways, and conducts robustness and regional heterogeneity tests. The empirical results show that: first, the relationship between AI use levels and corporate carbon emissions exhibits a significant inverted U-shaped curve-in the early stages of development, due to high energy consumption in computing and deployment, AI adoption may temporarily increase carbon emissions; however, after exceeding a critical point, further deepening of use significantly reduces carbon emissions. Second, SEM analysis reveals several key mediating channels: improving green innovation efficiency (GIE), enhancing energy utilization efficiency (EUE), promoting Scientific innovation (SI), and driving industrial structure upgrading (ISU). These pathways collectively amplify the emission reduction effect of AI. Third, regional heterogeneity analysis shows that the AI emission reduction effect is significantly stronger in the eastern region than in the central and western regions. Finally, this study emphasizes the policy implications: while promoting the use of AI, energy structure and incentive mechanisms should be optimized, and differentiated policies should be formulated according to regional characteristics to achieve an AI-driven sustainable low-carbon transformation.