Abstract Achieving carbon peak and carbon neutrality is an inherent requirement for countries to promote green recovery and transformation of the global economy after the COVID-19 pandemic. As "a smoke-free industry", its agglomeration may have significant impacts on CO2 emission reduction. This study uses nightlight data to calculate the CO2 emissions of 268 cities in China from 2005 to 2017 and then employs dynamic spatial Dubin model and intermediary effect model to explore the impact and transmission mechanism of producer services agglomeration (PSA) on CO2 emissions. Furthermore, the dynamic threshold model is used to analyze the nonlinear characteristics of PSA on CO2 emissions under different degrees of government intervention. The findings indicate that: (1) Generally, China's CO2 emissions are path-dependent in the time dimension, showing a "snowball effect". PSA not only significantly inhibits local CO2 emissions, but also reduces CO2 emissions in adjacent areas through spatial spillover effect; (2) PSA can indirectly curb CO2 emissions through economies of scale, technological innovation and industrial structure upgrading, Heterogeneity analysis shows that there are significant differences in the impact of PSA on CO2 emissions in different regions, time nodes and sub-industries; (3) The impact of PSA on China's CO2 emissions has an obvious double threshold effect under different degree of government intervention. Accordingly, the Chinese government should increase the support for producer services, dynamically adjust industrial policies, take a moderate intervention, and strengthen market-oriented reform to reduce CO2 emissions so that the goal of "carbon peak and carbon neutrality” can be early achieved.