Data Center Electricity Consumption Statistics: TWh, GW and the AI Share

Data center electricity consumption statistics have become some of the most closely watched numbers in energy policy, because they determine how many power plants, transmission lines and grid upgrades the AI boom will require. This article in the Data Center Economy hub collects the latest figures from the International Energy Agency, US national laboratories, EPRI and national statistics offices, and explains what they do and do not tell us. For context on where these facilities are, see how many data centers there are in the world; for the physical constraint behind the numbers, read our analysis of why power is the new data center bottleneck.

How much electricity do data centers use worldwide?

Data centers consumed about 415 TWh of electricity in 2024, roughly 1.5% of global demand, according to the IEA’s Energy and AI report. The IEA’s April 2026 update found consumption rose 17% in 2025, to about 485 TWh.

To put 485 TWh in perspective, Our World in Data notes it equals roughly 1.5% of global electricity generation of about 31,800 TWh in 2025, based on Ember data. That is a small share globally, but it is growing far faster than overall demand: the IEA reported that world electricity consumption rose about 3% in 2025, so data centers expanded at nearly six times the pace of the wider system.

The trajectory is steep. Our World in Data’s summary of IEA figures puts global data center use at about 270 TWh in 2020, meaning consumption has risen by roughly 80% in five years. Generative AI has pushed hyperscalers into an unprecedented capex cycle — the IEA notes that capital spending by five of the largest technology companies exceeded $400 billion in 2025 and was expected to rise a further 75% in 2026. Our explainer on the AI capex supercycle tracks that spending company by company.

Why some estimates are much higher

Not every source agrees with the IEA. Our World in Data points out that the Energy Institute and S&P Global estimate around 790 TWh for 2025, or about 2.5% of generation — roughly 60% above the IEA figure. Most of the gap comes from cryptocurrency mining, which Our World in Data estimates at 150–200 TWh and which the IEA largely excludes from its core data center number. Methodology also differs: some sources model consumption bottom-up from server shipments, others top-down from utility data.

How much electricity will data centers use by 2030?

The IEA’s base case has global data center consumption roughly doubling to about 945–950 TWh by 2030, close to 3% of world electricity demand, with AI-focused facilities tripling to around 465 TWh — nearly as much as all conventional data centers combined.

In the IEA’s breakdown, as summarised by Our World in Data, non-AI data centers would grow from about 330 TWh in 2025 to around 480 TWh in 2030, while AI-focused facilities climb from about 155 TWh to 465 TWh. AI-focused facilities would therefore go from about one-third of data center electricity to roughly half in five years.

Two technical trends pull in opposite directions. The IEA’s 2026 update reports that the power density of AI servers increased about elevenfold between 2020 and 2025 and is expected to rise another fourfold by 2027. At the same time, the energy needed per AI task is falling by at least an order of magnitude each year as chips and models become more efficient. Whether total consumption ends up above or below the base case depends on whether efficiency gains are outrun by the growth in usage.

The IEA is regarded as one of the more conservative forecasters. Our World in Data notes projections diverge sharply after 2030, and industry forecasts tied to announced construction pipelines are often far higher. Real estate adviser JLL, for example, expects global data center capacity to nearly double from about 103 GW to 200 GW by 2030, with AI rising from about a quarter to half of workloads.

How much electricity do US data centers use?

US data centers used about 176 TWh in 2023, or 4.4% of national electricity, according to Lawrence Berkeley National Laboratory. LBNL projected 325–580 TWh by 2028 (6.7–12%), and EPRI’s February 2026 study sees 9–17% of US generation by 2030.

LBNL’s December 2024 report for the Department of Energy is the most detailed public estimate. It found US data center consumption had reached about 76 TWh by 2018 and then accelerated as GPU-accelerated servers spread, more than doubling to 176 TWh in 2023. Its 2028 range translates to roughly 74–132 GW of average load assuming 50% utilisation. The report also found that 74% of US servers sat in colocation and hyperscale facilities in 2023, a share it expects to reach 85% by 2028.

EPRI’s newer analysis, released on 26 February 2026 and reported by E&E News, puts the current share at about 4.5% and projects 9% to 17% by 2030 depending on how much of the construction pipeline is completed. EPRI said its demand estimates were about 60% higher than its own figures from two years earlier. Our World in Data’s reading of IEA data puts US data centers at roughly 5% of national electricity in 2025.

Globally, the IEA attributed about 45% of 2024 data center electricity use to the United States, 25% to China and 15% to Europe. The scale of the US build-out is covered in our United States data center investment guide.

Source: IEA Energy and AI (April 2025) and 2026 update; Our World in Data (July 2026); LBNL 2024 United States Data Center Energy Usage Report; EPRI via E&E News (Feb 2026); Ireland CSO (July 2026).

Which US states depend most on data center electricity?

Virginia is the most exposed state, with data centers using about a quarter of its electricity today. EPRI projects that share could reach 41–59% by 2030, and expects Arizona, Indiana, Iowa, Nebraska, Nevada, Oregon and Wyoming could each exceed 20%.

Virginia’s Joint Legislative Audit and Review Commission (JLARC) found in December 2024 that the state’s data center sites drew about 5,050 MW, and that unconstrained statewide electricity demand could double within a decade, mainly because of data centers. After roughly flat demand between 2006 and 2020, that is a structural break for the state’s utilities.

Our World in Data notes that several US states already exceed 10%. The common thread among high-share states is a combination of hyperscale campuses and relatively small populations — Iowa, Oregon and Nebraska host major Google, Meta, Microsoft and AWS sites but have far less residential and industrial load than Texas or California. That makes their grids more sensitive to each new campus. Who pays for the resulting upgrades is contested; our article on whether data centers raise electricity prices lays out both sides.

What share of Ireland’s electricity do data centers use?

Data centers used 7,663 GWh in Ireland in 2025, or 23% of all metered electricity, according to the Central Statistics Office. That was up 10% from 6,973 GWh in 2024 and compares with just 5% of national consumption in 2015.

Ireland is the clearest real-world example of what happens when hyperscale demand meets a small grid. The CSO noted that data center consumption grew 10% in 2025 while all other users, including households, grew about 2%. Households accounted for 28% of total electricity use — meaning data centers now consume nearly as much as every home in the country combined.

The surge prompted grid operator EirGrid to restrict new connections in the Dublin area, and it has pushed developers toward other European markets. Our guide to Europe’s data center investment map covers how these constraints are redirecting investment to the Nordics, Spain and Italy. Across Europe as a whole, data centers used about 1.6% of electricity in 2025 based on IEA data, so Ireland is a striking outlier rather than the norm.

How much power do AI servers and accelerators consume?

AI-focused data centers used about 155 TWh in 2025 by the IEA’s estimate, roughly a third of the global data center total. Their consumption rose 50% in a single year, driven by accelerated servers packed with GPUs and custom AI chips.

The physical driver is density. The IEA reports AI server power density rose about eleven times between 2020 and 2025, with a further fourfold rise expected by 2027. Rack-scale systems built around the latest accelerators draw far more power per rack than traditional enterprise equipment, which is why operators are moving rapidly to liquid cooling — see our explainer on air, direct-to-chip and immersion cooling.

LBNL’s analysis of utilisation shows why AI loads are hard to forecast. It estimated that servers used for AI training ran at around 80% of their operational time, while inference servers ran nearer 40%. As inference becomes the dominant workload — JLL expects inference to overtake training in 2027 — average utilisation patterns and therefore consumption per installed megawatt may change.

How efficient are data centers at using electricity?

Efficiency is usually measured by power usage effectiveness (PUE), the ratio of total facility energy to IT energy. Leading hyperscale fleets operate close to the ideal of 1.0 — Google reported a fleet-wide trailing twelve-month PUE of 1.09 for both 2024 and 2025.

A PUE of 1.09 means only about 9% of energy goes to cooling, power conversion and other overhead. Older enterprise and smaller colocation facilities typically run considerably higher. Our guide to data center operating costs and PUE looks at industry averages and why power dominates the opex line.

PUE has limits as a sustainability metric. It says nothing about the carbon intensity of the electricity used, or about how much computation each kilowatt-hour delivers. That is why the IEA emphasises the falling energy per AI task, and why companies increasingly report hourly carbon-free energy matching. The tech sector accounted for about 40% of corporate renewable power purchase agreements signed in 2025, according to the IEA, and conditional offtake agreements between data centers and small modular reactor projects grew from 25 GW at the end of 2024 to 45 GW by early 2026. Our coverage of nuclear deals for data centers explores that trend.

How does data center electricity use translate into gigawatts?

Annual consumption in TWh and capacity in GW measure different things. One gigawatt of load running all year uses about 8.76 TWh; real facilities run below full load, so each gigawatt of installed capacity typically consumes less than that.

LBNL’s conversion illustrates the relationship: its 2028 US projection of 325–580 TWh corresponds to roughly 74–132 GW of capacity at 50% utilisation. Real estate analysts usually quote critical IT capacity — the power available to servers — rather than total facility draw.

Using that lens, the construction pipeline is enormous. JLL reported in August 2026 that more than 66 GW of data center capacity was under construction in North America, which it noted would require more electricity than Germany uses. Our construction pipeline statistics article breaks that pipeline down by region, while our article on data center water, land and jobs examines the rest of the local footprint.

Frequently Asked Questions