Environmental outcomes of artificial intelligence

emissions to account for trade-embedded carbon, and we assess adaptation capacity using the ND-GAIN readiness index. AI is measured using an AI capital stock indicator capturing tangible and intangible AI-related assets. Empirically, we estimate fixed-effects models with Driscoll-Kraay standard errors and strengthen causal interpretation for the emissions outcomes using an IV-2SLS design that instruments domestic AI with geography-filtered patenting shocks from major global innovation hubs. We then use panel local projections to trace the dynamic responses of emissions and climate readiness to unexpected AI-growth shocks. The results show that higher AI stock is associated with lower emissions, and the IV estimates support emissions-reducing effects in medium-run stock specifications. The dynamic evidence is more nuanced: territorial emissions rise after an AI-growth shock, peak around the medium-run horizon, and then ease, while consumption-based emissions decline more persistently. AI shocks are also followed by steady improvements in climate readiness. Together, the findings suggest that AI can support decarbonisation and resilience in advanced economies, but its environmental payoff depends on the deployment path and the energy system that absorbs AI-related demand.