Seven hundred observations State-year: the growth in the number of establishments is not associated with a measurable increase in electricity demand or emissions from the sector.
A study published on September 15 in Frontiers in Environmental Science tackles a question rarely asked so head-on: is the proliferation of data centers reflected in the energy statistics of American states? The authors put together a panel of seven hundred state-year observations, covering the fifty states from 2010 to 2023, and sought an association between the number of establishments linked to data centers and two variables — aggregate electricity consumption and emissions from the electricity sector.
The result is negative in both cases. The estimated coefficient for the number of establishments in log is −0.0266, without statistical significance; no significant association appears on the emissions side either. In other words, over the period studied, the footprint of data centers remains drowned out by the noise of the other determinants of a state's electricity demand.
The authors are careful not to conclude that the problem does not exist. They point out that data centers accounted for about 4 to 4.4% of US electricity consumption in 2023, and that this share could reach 12% by the end of the decade. The period analyzed stops precisely before the acceleration of constructions related to generative AI: what the study documents is the absence of a signal in the historical data, not the absence of a future effect.
Read the full article on the original source
Go to the source →