National AI Strategies and the Green Transition: Cross-Country Evidence from Staggered Difference-in-Differences and Double Machine Learning

National artificial intelligence (AI) strategies have proliferated since 2017, yet whether they advance the green transition remains unknown. We exploit the staggered adoption of national AI strategies by 59 countries within an 85-country panel over 2000–2021, combining two-way fixed effects (TWFE), Callaway–Sant'Anna difference-in-differences, and double machine learning (DML). Strategy adoption raises the renewable energy share of consumption by 1.87 percentage points in fixed-effects estimates and by 1.47 points in dynamic event-study aggregation, and reduces the CO₂ intensity of GDP (−0.026 kg per 2017 PPP dollar), a result corroborated by a significant decline in CO₂ emissions per capita. In contrast, we find no robust effect on the share of environment-related patents: pre-trends are rejected and point estimates switch sign across estimators. Green-patenting effects are concentrated, albeit imprecisely, in high-income economies, and carbon-intensity reductions are larger where renewable endowments are high. DML estimates for green patenting are sensitive to the choice of learner. National AI strategies appear to green the energy mix before they green the innovation base.