Hyperscale Data Centers Consume Nearly 2 Percent of US Electricity, Study Finds
A peer-reviewed study reveals that 403 US hyperscale data centers consume roughly 82 terawatt-hours of power annually, relying predominantly on fossil fuels despite aggressive industry pledges toward clean energy.

In this piece
America's 403 hyperscale data centers now consume roughly 1.8 percent of the nation's electricity, according to new research published in Nature Sustainability.…
America's rapidly expanding network of hyperscale data centers now consumes roughly 1.8 percent of all electricity generated in the United States, according to a peer-reviewed study published in Nature Sustainability. Led by researchers at the Harvard T.H. Chan School of Public Health, the comprehensive analysis evaluated 403 operational facilities nationwide between mid-2024 and mid-2025. In total, the massive computational hubs accounted for approximately 82 terawatt-hours of power annually.
That annual power demand rivals the entire electrical output consumed by mid-sized states such as Indiana, Michigan, or Tennessee. Researchers estimated that depending on varying operational loads and cooling intensities, total consumption across the examined facilities ranges from 68 to 99 terawatt-hours. The findings provide one of the most detailed facility-level accountings to date of the physical footprint underpinning the modern digital economy.
Despite prominent corporate commitments by leading technology companies to achieve net-zero operations, the study revealed that the facilities remain heavily reliant on carbon-intensive energy sources. Approximately 54 percent of the electricity consumed by the examined hyperscale hubs came from fossil fuels, with natural gas supplying around 38 percent and coal providing 15 percent. Carbon-free alternatives made up the remainder, split between nuclear generation at 21 percent and renewables at 25 percent.
As a result of this grid composition, the researchers calculated that U.S. hyperscale data centers generate roughly 25.7 million metric tons of carbon dioxide emissions each year. This environmental footprint is equivalent to the total annual greenhouse gas output of mid-sized nations such as Denmark or Bolivia. The documented emissions levels represent a two- to three-fold increase compared to baseline peer-reviewed benchmarks established in 2018.
Evaluating data center power usage has historically proved difficult because operators rarely disclose facility-specific energy metrics or utility bills. To overcome this lack of transparency, the Harvard research team cross-referenced commercial databases with high-resolution satellite imagery to verify building footprints and physical dimensions. They then integrated regional grid data from the Environmental Protection Agency to map facility demand against the carbon intensity of local electrical grids.
The analysis also uncovered extreme geographic clustering, which places disproportionate strain on local utility infrastructure. More than half of all hyperscale power consumption in the study was concentrated in just four states: Virginia, Oregon, Ohio, and Iowa. In regions like Northern Virginia, where dense clusters of facilities support major cloud platforms, high power density has prompted utilities to reconsider infrastructure investments and postpone fossil fuel plant retirements.
The surge in electrical demand is driven largely by the massive computational requirements of cloud computing infrastructure and the rapid rollout of generative artificial intelligence models. Training and running advanced machine learning algorithms require thousands of specialized graphics processing units running continuously, which in turn demands substantial auxiliary power for liquid cooling and climate control. Because the study's dataset concluded in early 2025, researchers noted that current power requirements may already exceed their published figures.
The escalating power draw highlights an intensifying tension between the tech industry's artificial intelligence ambitions and broader decarbonization mandates. Power grid operators across the country are facing mounting connection backlogs and capacity constraints as data center developers seek multi-gigawatt utility interconnections. In several regional grids, utilities are weighing whether new natural gas generation will be required to keep pace with projected commercial growth.
Moving forward, researchers emphasized that establishing mandatory reporting standards for commercial data facilities will be essential for policymakers and utility planners. Independent analysts warn that without greater public visibility into facility-level energy flows, power authorities cannot accurately plan transmission expansions or protect regional reliability. The study concludes that tech firms must align their rapid capital investments with local grid capacity if they hope to fulfill their long-term environmental targets.
Primary source: Ars Technica. Newzly rewrites and structures source reporting for clarity; verify critical details at the original.