Study finds AI growth raised carbon emissions at China's data centers

Local green data-center policies and stronger green-technology cooperation between cities both weakened the link once they passed estimated thresholds. Neither reversed it, so AI growth still pushed emissions up under those conditions.

The effect also varied by distance. Spatial spillovers were negative between roughly 150 and 650 km and positive at longer distances, so one city's AI growth was tied to emissions elsewhere in a non-uniform pattern.

Researchers built a bottom-up account of operational carbon emissions from purchased electricity at 1,090 data centers across 162 Chinese cities, covering 2015 to 2024. They related it to a multidimensional index of AI development. The authors say earlier work had not isolated this operational-phase outcome or its mechanisms, thresholds and spatial patterns.

The headline estimate is an association: a one-standard-deviation increase in the AI index goes with an additional 68.82 million tonnes of emissions. Two mitigating factors were examined, local green data-center policies and green-technology cooperation between cities. Each weakened the association only after passing an estimated threshold, and neither reversed it.

The spatial analysis found non-monotonic spillovers. Effects on other cities were negative at distances of roughly 150 to 650 km and positive beyond that. The authors say the results identify the conditions under which AI growth raises operational emissions from computing infrastructure.