Using ChatGPT's release as a technology shock, this study finds that generative AI increases household consumption-based carbon emissions in China, particularly indirect emissions.
Generative artificial intelligence (GenAI) brings economic benefits by increasing productivity, but it may also increase the environmental costs of household consumption. Using the public release of ChatGPT on November 30, 2022, as a shock, this study matches China Family Panel Studies (CFPS) data with city-level GenAI exposure and applies generalized difference-in-differences to assess the impact of GenAI on household consumption-based carbon emissions. We find that GenAI increases household consumption-based carbon emissions, with a greater effect on indirect emissions embodied in goods and services than on direct emissions from household energy use. To address potential endogeneity, we further apply a Bartik instrumental variable approach and use causal mediation effect models to examine the mechanisms. The results show that although GenAI-driven industrial upgrading reduces carbon emission intensity to some extent, rising household income and greater digital network use increase consumption-related emissions. Heterogeneity analysis shows that this increase is stronger among urban households, high-income households, households with higher social status, and households with internet access. This study provides new household-level evidence on the environmental impacts of GenAI and highlights the importance of coordinating sustainable consumption and carbon reduction goals while promoting technological innovation and economic growth.
Liang Fu is a joint PhD candidate at Huazhong Agricultural University and Lincoln University. His research examines artificial intelligence, agricultural economics, and climate change mitigation.