Artificial intelligence (AI) has great potential to address the carbon crisis, but unlocking its environmental benefits faces challenges. Income inequality, a key constraining factor, has received insufficient research attention. This study innovatively incorporates income inequality into the AI-carbon emissions relationship research framework. Using 2002–2019 panel data from 50 countries, we employ empirical methods to analyze how income inequality moderates the impact of AI technology on carbon emissions. Additionally, we utilize a panel threshold model to explore the nonlinear effects and heterogeneity of this moderating influence. Results show: (1) Overall, AI has a significant emission reduction effect; (2) Income inequality undermines the abatement benefits of AI, and when the income gap exceeds a threshold, the emission reduction effect of AI will be reversed; (3) The moderating effect exhibits heterogeneity across countries with different income levels, with the abatement efficacy of AI in low- and middle-income countries being more affected by inequality. The study reveals income disparity's constraint on AI's emission reduction potential, highlights the importance of considering social equity in AI development, provides a basis for designing fair and effective environmental policies, and offers insights into synergistically resolving the climate and inequality crises.