AI Boom Threatens Power Bills

Electrical substation with power lines at sunset
Photo: Haizhen Du / Shutterstock

U.S. data centers now consume about 4.4% of America’s electricity—roughly the output of a mid-sized state—and that share is rising fast.

Story Snapshot

  • Federal researchers say data centers used about 176 terawatt-hours in 2023, or 4.4% of U.S. power.
  • Government forecasts show demand could reach 6.7% to 12% of U.S. electricity by 2028.
  • Growth is driven by artificial intelligence, cloud services, and video, straining local grids.
  • Who pays for upgrades varies by state, but households will likely shoulder some costs.

What the New Federal Data Shows

The U.S. Department of Energy said data centers consumed about 176 terawatt-hours of electricity in 2023, equal to roughly 4.4% of national use. Lawrence Berkeley National Laboratory built this estimate using a bottom-up model that tracks equipment and cooling loads across facilities. The researchers found that demand has surged over the past decade, with rapid growth in the last few years. This baseline matters because it sets the stage for how much new generation and grid capacity utilities must plan and finance.

Lawrence Berkeley National Laboratory projects that data center electricity use could grow to between 325 and 580 terawatt-hours by 2028, or about 6.7% to 12% of U.S. consumption, depending on technology choices and efficiency gains. Those scenarios reflect heavy artificial intelligence training and inference, expanding cloud services, and more streaming. The wide range shows how much outcomes depend on server efficiency, cooling design, and whether companies add on-site power that lowers strain on local lines.

Why the Numbers Differ From “2%” Headlines

Some stories cite a lower “about 2%” share. Experts explain that differences often come from what gets counted and which year is used. The federal analysis includes on-site cooling and backups, covers more facility types, and uses 2023 as the base year, when loads were higher. The lesson is simple: the load is real, material, and growing, but the exact share can shift with scope and timing. Policymakers should read the method notes before leaning on a single headline figure.

Geography also shapes the impact. New server farms often land where land is cheap, taxes are lower, and power is available. That can mean fast growth hits smaller grids first. Local utilities then rush to add lines, substations, and sometimes new gas peakers to meet peak load. Those projects take years and billions of dollars. When planning lags demand, residents and small firms can face higher bills, brownout risks, or both, even as large customers negotiate special rates to anchor the buildout.

Who Pays and What It Means for Families

Researchers and utility watchers say cost-sharing varies by state and utility rules. But many expect households to pay a share of grid upgrades through higher delivery charges on monthly bills. One analysis finds that recent bill spikes mostly came from inflation, fuel costs, and grid work, not data centers alone, yet large new loads do add pressure in some regions. The bottom line for ratepayers is familiar: when big users plug in quickly, the system scrambles, and ordinary customers often cover part of the tab.

Efficiency and planning can help. The federal reports highlight options such as liquid cooling, server utilization improvements, heat reuse, and buying clean power near load. Siting near existing transmission and firm generation can cut delays and costs. Clearer rules for who funds upgrades can protect families and small businesses. These steps do not end the growth in demand, but they can reduce the shock and keep power reliable while the digital economy expands on U.S. soil.

Sources:

newscientist.com, eta.lbl.gov, energy.gov, eta-publications.lbl.gov, histsci.fas.harvard.edu, dl.acm.org, nationalacademies.org