AI infrastructure and energy
IEA puts data centres at 485 TWh in 2025 and 950 TWh by 2030, while reviewers span a 40 fold range
The International Energy Agency's 16 April 2026 update has data centres using 485 TWh in 2025, rising to about 950 TWh by 2030, with demand up 17 per cent against global electricity growth of 3 per cent. A 2025 review of more than 100 studies found 2030 projections ranging from just over 200 TWh to nearly 8,000 TWh.

The International Energy Agency's Key Questions on Energy and AI, whose executive summary and accompanying charts were last updated on 15 and 16 April 2026, gives data centre electricity consumption of 485 TWh in 2025 and projects it roughly doubling to about 950 TWh in 2030. That is the figure most often quoted. The structure underneath it is more interesting than the headline.
Data centre electricity demand rose 17 per cent in 2025, against global electricity demand growth of 3 per cent, according to the IEA's news release of 16 April 2026. Within that total, AI focused sites grew about 50 per cent in 2025 and are projected to triple by 2030. The executive summary puts data centres at around 3 per cent of global electricity in 2030. Set 950 TWh against a 3 per cent share and the implied world total is roughly 31,000 TWh, which is consistent with that share attaching to data centres as a whole rather than to AI sites alone, a distinction that is frequently lost in retelling.
On emissions, the IEA projects data centre emissions doubling to around 350 million tonnes in 2035, about 2 per cent of electricity sector emissions. On the physical build, the summary states that around 15 to 27 GW of onsite natural gas may power data centres by 2030, mostly in the United States, with the explicit caveat that "uncertainties are high", and that around 20 to 25 GW of battery storage could be installed at data centres globally. It also notes that onsite generation is typically overbuilt by 30 to 70 per cent above actual demand.
Against those energy numbers sits a much steeper money number. The IEA reports that capital expenditure by five large technology companies passed USD 400 billion in 2025 and is expected to jump by a further 75 per cent in 2026. The agency's own chart page for hyperscaler capital expenditure, last updated 15 April 2026, marks the 2026 figures as estimates and does not make the underlying data available for download. The tech sector accounted for roughly 40 per cent of corporate power purchase agreements for renewables in 2025, and conditional offtake agreements for small modular reactors rose from 25 GW at the end of 2024 to 45 GW by the time of publication. Spending is compounding considerably faster than consumption, which is either a sign that efficiency is holding, or a sign that a great deal of announced capacity will not be built. The IEA lists both possibilities among its uncertainties, alongside grid connection bottlenecks, the unknown adoption path of energy intensive use cases, and sensitivity to "expectations for returns on investment".
The second half of the picture is how far apart the analysts are. Carbon Brief, in an analysis by Josh Gabbatiss published on 15 September 2025, put data centres at just over 1 per cent of global electricity demand and 0.5 per cent of CO2 emissions in 2024, and noted that 2030 forecasts range from roughly 620 TWh to more than 2,000 TWh depending on who is doing the forecasting. Carbon Brief also placed data centre growth in proportion: an increase of 530 TWh between 2024 and 2030, about 8 per cent of total global electricity demand growth, against 838 TWh for electric vehicles, 651 TWh for air conditioning and 1,936 TWh for industry. It stated plainly that "there are no comprehensive global datasets available on data-centre electricity consumption or emissions" and that all such figures are therefore estimates.
The most systematic account of that spread is Data Centre Energy Use: Critical Review of Models and Results, by George Kamiya and Vlad C. Coroama, published in March 2025 for the EDNA platform of the IEA 4E Technology Collaboration Programme, a body established under the auspices of the IEA but functionally and legally autonomous from it. Reviewing more than 100 publications since 2014 covering over 200 scenarios, the authors found that projections for 2030 range from just over 200 TWh to nearly 8,000 TWh, a factor of almost 40. Their own convergent estimate, from three independent methods including company reported data from 60 of the largest operators, is that data centres used 300 to 380 TWh globally in 2023, excluding crypto. They put AI related consumption at 10 to 50 TWh in 2023, or 5 to 15 per cent of the data centre total, and give a plausible 2030 range of 200 to 400 TWh, against other studies projecting 200 to 900 TWh.
Two findings from that review deserve wider circulation. The modelling approach used is a good predictor of a study's assessed quality, while the main author's affiliation is not. And the authors warn journalists specifically to avoid cherry picking the most extreme scenario results in either direction.
What is not settled: whether the capital expenditure now committed converts into built and connected capacity, and on what timetable. The IEA's own list of uncertainties concedes that not all announced projects will come to fruition.
Sources
Every factual claim above rests on the 6 published sources below. They are listed so you can check the reporting rather than take it on trust.
- International Energy AgencyKey Questions on Energy and AI: Executive summary
- International Energy AgencyData centre electricity use surged in 2025, even with tightening bottlenecks driving a scramble for solutions
- International Energy AgencyElectricity consumption by data centres, 2020 to 2035 (chart)
- International Energy AgencyCapital expenditures by hyperscalers (chart)
- Carbon BriefAI: Five charts that put data-centre energy use and emissions into context
- EDNA, IEA 4E Technology Collaboration ProgrammeData Centre Energy Use: Critical Review of Models and Results (Kamiya and Coroama)


