Carbon emission reduction effects of computing power development: evidence from double machine learning

Achieving the “dual carbon” goals constitutes a systemic endeavor requiring coordinated societal efforts. As a new productive force in the digital economy era, computing power development should actively contribute to advancing carbon neutrality. This study aims to clarify the quantitative relationship between computing power development and regional carbon emissions, explore the core pathways and heterogeneous characteristics of computing power-driven carbon reduction, and provide scientific support for leveraging computing power to empower green transformation. Leveraging panel data from 30 Chinese provinces (2011–2022), this study employs dual machine learning for benchmark regression and conducts robustness tests through variable substitution, capping, algorithm replacement, and instrumental variables. It empirically examines how computing power development affects regional carbon emissions and unveils the underlying mechanisms. The findings demonstrate that computing power development exerts significant carbon emission reduction effects, which remain robust across multiple sensitivity analyses. Multidimensional examination reveals that computing power infrastructure emerges as the principal driver, while industrial and technological dimensions exhibit limited contributions. Mechanism analysis identifies dual pathways through energy consumption structure optimization and industrial structure upgrading. Heterogeneity analysis indicates stronger emission reduction effects in western regions, inland areas, regions with weaker environmental regulations, and less economically developed areas. This study systematically reveals the decarbonization potential of computing power and provides empirical evidence, offering policy insights for regional green transformation. It further proposes recommendations for resource sharing, industrial cultivation, and differentiated regional policies.