Artificial Intelligence Policy and Urban Carbon Emission Performance: Evidence from the Yangtze River Delta, China

Whether artificial intelligence (AI) policy translates into measurable urban decarbonisation remains uncertain: most evidence is firm- or robot-level, and transparent city-level policy evidence is scarce. This study treats the State Council’s “New Generation Artificial Intelligence Development Plan” and subsequent local responses as a quasi-natural experiment, using panel data on 26 Yangtze River Delta cities over 2010-2023. Four core cities (Shanghai, Nanjing, Hangzhou, Suzhou) are defined as the AI-policy treatment group, and a two-way fixed-effects difference-in-differences (DID) model estimates the effect of AI policy exposure on urban low-carbon performance, measured as a standardised inverse carbon-intensity indicator. The baseline DID coefficient is 0.493 and significant at the 5 % level, and the positive sign is stable when the onset year is set to 2018, the treatment group is expanded, the window is shortened, and Shanghai is excluded. The estimate is, however, sensitive to city-specific linear trends and to over-control specifications, so its magnitude should be read cautiously; event-study results are consistent with parallel pre-trends, with effects emerging gradually after implementation. Stepwise mechanism tests give indicative - not formal - evidence that industrial structure upgrading and green technological innovation act as transmission channels. Heterogeneity across cities is suggestive but not statistically confirmed, and cross-city spillovers are not robust across spatial weight matrices. AI policy thus appears associated with better urban carbon performance mainly through local channels, and implementation should prioritise practical low-carbon applications, green-innovation commercialisation, and locally differentiated strategies.