Particle.news

Harvard Study Shows U.S. AI Data Centers Use Vast Power and Produce Large CO2 Emissions

The finding heightens pressure on grids, pushes calls for cleaner power, and prompts state regulatory talks as hyperscale capacity expands rapidly.

Overview

  • The peer-reviewed Harvard study, published Nov. 18–19, 2026, found 403 hyperscale U.S. data centers consumed about 82 terawatt-hours of electricity and produced roughly 25.7 million tonnes of CO2 between May 2024 and April 2025.
  • The study estimated roughly 54 percent of that power came from fossil fuels, a share that made the facilities' emissions larger than the annual CO2 output of some small countries.
  • Lead author Gianluca Guidi told reporters that the U.S. hyperscale fleet has more than doubled since the study period and that about 2,000 additional facilities are planned or under development.
  • The International Energy Agency projects global data center electricity use could double by 2030 with related emissions rising sharply, increasing demands for transmission lines, transformers, and other grid upgrades.
  • The report has already prompted a voluntary COP31 pledge to clean up AI infrastructure that major U.S. tech firms have not signed and has triggered state-level responses such as talks in Oklahoma about a possible moratorium and concern over rising local bills.