How AI Data Centers Could Push US Power Emissions Back Up

A Sierra Club analysis projects that carbon dioxide emissions from U.S. power plants could be 42% higher in 2030 than in 2025. The cause is new fossil-fuel plants built to meet the electricity needs of data centers. The Verge reported the analysis on October 29, 2026.
The projected increase is 618 million tons by the end of the decade. That number assumes the planned AI projects actually get built.
The Verge put the figure in household terms. It is roughly the pollution from the electricity used by 128,788,523 homes in one year. The same report said the added emissions would essentially wipe out 15 years of U.S. greenhouse gas reductions.
Net U.S. emissions fell from a peak of about 6.58 billion metric tons in 2007 to about 5.35 billion metric tons in 2025.
Bloomberg published its article on October 28, 2026. It cited the same Sierra Club analysis and said the boom would cancel out 15 years of declines in power-sector emissions. Bloomberg put the power sector's 2030 emissions under buildout projections at 2.1 billion metric tons of carbon dioxide.
The key question is where the electricity comes from. Reuters reported on July 1, 2026 that gas plants planned to serve U.S. data centers were estimated to generate 143 gigawatts of electricity. A gigawatt is a measure of how much power a plant can deliver at once. Those plants were estimated to produce 662 million tons per year of greenhouse gases.
On the demand side, the Sierra Club's Dirty Truth Report cites a Rhodium Group estimate. It says data-center demand could grow by 338 percent by 2030 under a high-growth scenario.
Other research points the same way, at a different scale. A June study by academic researchers forecast that electricity demand from data centers and crypto mines could push power-sector emissions up 28% by 2030. That comparison was with a scenario that had no data-center growth.
The Environmental Defense Fund, in a separate report, found the U.S. is planning to build more fossil-fuel capacity than it has in a decade. A Sierra Club article also states that a new study shows emissions grow when the oil and gas industry uses AI. That thread reaches beyond the power grid.
These figures are projections, not measured emissions. Each one depends on assumptions about how much of the planned pipeline gets built and which fuels power it. The 42% figure, the 28% academic estimate and the 2.1 billion ton projection help show a likely range, but none of them is an observed result.
The broader context here is that the emissions question depends less on data-center demand itself than on the power plants that supply it. A power plant is built to last for decades, much like a house. The choices about where to build and which fuel to burn over the next few years will shape the 2030 totals more than any single efficiency gain inside a data center.
There is a more hopeful reading in the same numbers. U.S. net emissions have already fallen substantially from their 2007 peak, which suggests the power sector can change direction when decisions about new capacity change. The same building cycle that could add gas plants could also add lower-emitting power. The demand that creates the problem also creates the buying power that could pay for cleaner alternatives. Which path the country takes depends on choices not yet made, and the next construction cycle will show which way they go.


