A new analysis finds AI data center expansion could wipe out 15 years of greenhouse gas reductions from electricity generation.
The US power sector may be on course to undo a decade and a half of hard-won emissions progress, and the culprit is the electricity appetite of artificial intelligence. A new Sierra Club analysis projects that carbon dioxide emissions from US electricity generation could climb 618 million tons by 2030, a 42% increase over 2025 levels that would push total power-sector CO2 to approximately 2.1 billion metric tons.
To put that in household terms the Sierra Club did: the added pollution is roughly equivalent to the electricity-related carbon output of more than 128 million American homes in a single year.
Where the emissions are actually coming from
The bulk of the projected increase, over 460 million tons, does not come from the traditional grid at all. It comes from what the industry calls “bring-your-own-power” natural gas plants: off-grid facilities that tech companies and data center operators are building to sidestep the notoriously slow grid interconnection queue.
Grid-connected generation is projected to add another 154 million tons, a 10% increase above baseline projections, but that figure is secondary to the off-grid problem.
Jeremy Fisher, a principal advisor at the Sierra Club, noted that the emissions growth projections have roughly doubled to quadrupled compared to estimates from just 18 months earlier.
Up to 80% of the energy generated by self-powered data center projects may come from natural gas, according to the Sierra Club analysis. Renewable energy permitting delays make it difficult to substitute cleaner sources quickly enough to keep pace with construction timelines.
