IEA energy and AI report: Key Questions on Energy and AI, 2026
In April 2026, the International Energy Agency updated its estimate of how much electricity data centers and AI take. This IEA energy and AI report follows the previous edition, published in 2025 as “Energy and AI”. By the new data, the world’s data centers used 485 TWh in 2025, and their use will roughly double by 2030. Below are the main figures, the method, the limits and the conclusions for a company in Russia.
Key figures
485 TWh
used by the world’s data centers in 2025; by 2030, use will roughly double to 950 TWh, about 3% of world electricity demand
415 TWh
data center use in 2024 (416 TWh in the table); 360 TWh in 2023
50%
growth of AI data center use in 2025; electricity demand of all data centers grew by 17%
465 TWh
will be used by AI data centers by 2030, more than a threefold rise
under 10%
of all growth in world electricity demand to 2030 will come from data centers, a little under 10%
$715 billion
of capital expenditure announced by the largest technology companies for 2026, up 75% after more than $400 billion in 2025
an order of magnitude
is the minimum yearly fall in energy per AI task, but video, reasoning and agents need hundreds or thousands of times more energy per request than plain text
426 TWh
will be used by US data centers in 2030 (224 in 2025), China 277 TWh (117), Europe 114 TWh (72), Base scenario
The report in brief
- Published by
- International Energy Agency (IEA)
- Published on
- 16 April 2026
- Original title
- Key Questions on Energy and AI
- Length and language
- 138 pages, in English, CC BY 4.0 license (annex A: CC BY-NC-SA 4.0)
- Checked
- 16 October 2026
- Next edition
- The IEA promises to update its forecasts regularly; there is no new demand forecast after April 2026, and the report “Powering Artificial Intelligence” of 14 October 2026 deals with connecting data centers to grids
“roughly doubling from 485 TWh in 2025 to 950 TWh in 2030”IEA, p. 10, Executive Summary
How the report was made
The IEA model counts physical shipments of servers to data centersAI data center. AI data centers are buildings with thousands of servers on graphics cards and other accelerators, with strong power supply and cooling. Neural networks are trained and run there., above all AI accelerators, from IT industry data and forecasts. It adds company reports and satellite monitoring of the construction of large AI centers. The base estimate for 2025 is 485 TWh (table A.1, p. 108).
The forecast to 2030 comes in several scenarios: Base, Lift-Off, High Efficiency and Headwinds; the estimates to 2035 are exploratory. The IEA says the Base scenario stayed close to the path of the 2025 report (p. 10, p. 24).
What the report does not measure
Industry data is incomplete: there are no full statistics on how often and how deeply AI is used worldwide (pp. 9–10, p. 18). The IEA calls its 2035 estimates exploratory because of high uncertainty.
The report text gives 950 TWh by 2030, and table A.1 gives 945 TWh; the difference comes from rounding (annex A, p. 107). The regional tables do not show Russia and Eurasia separately.
What it means for a company in Russia
Russia in the report: the report has no data on Russia. It belongs to the “Eurasia” group in the region definitions (annex B, p. 121), but this group is missing from the data center tables (tables A.2-A.4, pp. 109–110).
- The energy for one AI task falls by at least an order of magnitude a year, but reasoning and agents need hundreds or thousands of times more energy per request than plain text (p. 9). Do not use a heavy model or an agent where a simple request is enough. The budget by item is covered in our article how much AI adoption costs.
- AI data centers will use about 465 TWh by 2030 (p. 26); by our calculation from the report’s figures, that is about half of the 945 TWh of all data centers (table A.1). If you put AI on your own servers, calculate power, cooling and backup supply in advance: the requirements are on our page data center for AI.
- The US (426 TWh) and China (277 TWh) will take the largest share of use by 2030, and there are no separate figures for Russia (table A.4). For a company in Russia, the question is not only capacity but also where the data is stored and what the law allows. This is covered in our section law and data.
What to compare it with
- World Bank: AI foundations 2025: where data centers are by country group, a different source and a different method
- State of AI Report 2026: AI computing and investment through the eyes of an industry review