How Much Electricity a Data Center Uses

The best national number comes from the Lawrence Berkeley National Laboratory. Its 2024 United States Data Center Energy Usage Report, written by Arman Shehabi and colleagues and released by the Department of Energy on December 20, 2024, estimated that US data centers used 176 terawatt-hours in 2023. That was about 4.4 percent of all electricity used in the country. Ten years earlier, in 2014, the figure was 58 TWh.

For 2028, the report gives a range instead of a single number: roughly 325 to 580 TWh, or 6.7 to 12 percent of projected US electricity use. The width of that range is the point. The low end assumes slower growth in AI servers and better efficiency. The high end assumes the buildout continues at full tilt. Anyone who quotes only one end is choosing a side, whether they know it or not.

The International Energy Agency looks at the whole world. Its Energy and AI report put global data center use at about 415 TWh in 2024, roughly 1.5 percent of world electricity, and projected about 945 TWh by 2030. The IEA also says data centers account for nearly half of US electricity demand growth between now and 2030. So the national share is still modest, but the growth is concentrated, and it lands in particular places.

Power versus energy

Most confusion about data centers starts here. A developer says a campus is "300 megawatts." A council member repeats it as "300 megawatts a year." That phrase means nothing.

A megawatt measures power, the rate at which electricity is used right now. A megawatt-hour measures energy, the amount used over time. A 1 MW load running for one hour uses 1 MWh. Running for a full year (8,760 hours) it uses 8,760 MWh. Your water bill works the same way: the pipe size is one thing, the gallons on the meter are another.

A campus rating can mean several different things, and the difference matters:

• Utility service request: the maximum the customer asked the utility to be ready to deliver.
• IT load: the power the servers themselves draw, before cooling and other overhead.
• Facility load: IT load plus cooling, lighting and power conversion losses.
• Buildout target: what the site might reach after ten years and several buildings, often far above year one.

When someone gives you a megawatt number, ask which of those it is and for what year.

The homes-equivalent math, shown

Most people picture homes more easily than megawatt-hours. The Energy Information Administration reports that the average US residential customer bought about 10,791 kWh in 2022, or about 899 kWh a month. That is the yardstick used below.

Check the result against someone else's math. The Idaho Conservation League described the proposed Gemstone Technology Park near Kuna, planned at 600 to 800 MW, as equal to 390,000 to 520,000 homes. That works out to 650 homes per megawatt, which matches the 80 percent figure above (about 649 per MW). When two independent calculations land in the same place, you can use the number with some confidence.

Campus size	Annual energy at 80% load factor	Homes-equivalent (10,791 kWh/yr)
20 MW	140,160 MWh	about 13,000
100 MW	700,800 MWh	about 64,900
300 MW	2,102,400 MWh	about 194,800
1,000 MW (1 GW)	7,008,000 MWh	about 649,400

The IEA says a typical AI-focused data center uses as much electricity as 100,000 households, and that the largest ones under construction will use 20 times as much. Those are global households, which use less than American ones, so do not mix the IEA comparison with EIA numbers in the same sentence.

Where homes-equivalent breaks down

The comparison is about energy over a year. It says nothing about peak demand, which is what forces a utility to build power plants and lines. An average home draws about 1.2 kW averaged over the year (10,791 kWh / 8,760 h), but much more on a hot afternoon with the air conditioner running. A data center draws close to the same amount at 3 a.m. as at 5 p.m. That steadiness cuts both ways. It can use plant capacity that would otherwise sit idle overnight. It also never backs off during the hours when the grid is tightest, unless a contract makes it.

Reading a load forecast

Every regulated utility forecasts future demand in its integrated resource plan, filed with the state commission. Idaho Power's 2025 plan, as reported by the trade newsletter Clearing Up in July 2025, forecast peak load growing about 1,700 MW, or 45 percent, by 2045, with nearly 1,000 MW of that in the next five years. The plan proposed more than 4 GW of new resources and conservation.

IDACORP's first-quarter 2026 investor slides compare forecasts from three plans. The five-year retail sales figure was 2.6 percent in the 2021 plan, 5.5 percent in the 2023 plan and 8.3 percent in the 2025 plan. The slide labels it a "5-Year Forecasted Annual Growth Rate," so 8.3 percent is a yearly rate for 2025 through 2029. Compounded over five years that is about 49 percent growth (1.083 to the fifth power is about 1.49). Strip off the label, as a quick summary often does, and "sales will grow 8.3 percent" understates the forecast by a factor of six. Always find the label in the source document.

Forecasts also carry a known problem with large loads. A developer may shop the same project to three utilities in three states. Each utility puts it in its queue. Add them up nationally and you count one building three times. Texas responded in 2025 by requiring large-load customers to disclose duplicate requests elsewhere in the state (covered in the grid lesson). When a forecast jumps, ask the utility how much of the increase is signed contracts, how much is customers who paid for studies, and how much is inquiries.

Watch which IEA report a number comes from. Electricity 2024 (January 2024) estimated data centres, AI and crypto at about 460 TWh in 2022 and said use could pass 1,000 TWh in 2026, "roughly equivalent to the electricity consumption of Japan." Energy and AI (2025) projects about 945 TWh by 2030 and says data centre use has grown about 12 percent a year since 2017, more than four times faster than total electricity use. Goldman Sachs Research projected in May 2024 that data center power demand would grow 160 percent by 2030. Pierson's The Hidden Driver of Your Power Bill uses these figures; match each to its report.