There is no reliable global total for electricity or water used by AI data centres alone. The best available figures cover data centres overall, which serve AI alongside cloud, storage and networking workloads: the International Energy Agency (IEA) estimates 485 terawatt-hours (TWh) of electricity use in 2025 and about 560 billion litres of water consumption annually. Those totals have different reporting boundaries, and neither should be read as an AI-only figure.
How much electricity do data centres use?
The IEA’s 2026 update estimates that data centres worldwide used 485 TWh of electricity in 2025. Its earlier Energy and AI report put 2024 use at 415 TWh, or about 1.5% of global electricity consumption that year. Both figures cover the data-centre sector, not AI alone. IEA, Energy and AI (2025); IEA, Key Questions on Energy and AI: Executive summary (2026).
As an Amazon Associate I earn from qualifying purchases.
The IEA’s 2026 central projection reaches about 950 TWh in 2030, roughly twice its 2025 estimate. This is a forecast for all data centres, not a measured outcome or an AI-only prediction. It depends on factors such as AI adoption, efficiency improvements, supply chains and limits on infrastructure.
PC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchWhy an AI-only electricity total is hard to give
Data centres commonly handle mixed workloads. Public figures do not consistently separate electricity used for AI from electricity used for cloud services, storage, networking and other computing. A facility described as AI-focused still needs a clearly defined boundary before its consumption can be compared with sector totals.
#1 Best Overall
How much water do data centres use?
The IEA’s 2025 estimate is about 560 billion litres of water consumed annually across global data centres. Its 2030 base case is about 1,200 billion litres. These estimates include direct water use at facilities as well as indirect water consumption associated with energy supply and semiconductor manufacturing; the 560-billion-litre figure is not a cooling-tower-only total. IEA, Energy and AI (2025).
For 2023, the IEA attributes roughly two-thirds of data-centre water consumption to primary energy supply and electricity generation, about one-quarter to direct cooling, and the remainder to semiconductor and microchip manufacturing. These are global modeled estimates, not a meter reading from one facility.
Rank #2
Withdrawal is not the same as consumption
Water withdrawal is water taken from a source. Water consumption is the portion not returned to that source, for example because it evaporates. A reported number is meaningful only if its metric and boundary are clear: onsite cooling, total facility operations and upstream supply-chain water are not interchangeable measures.
The Tool Desk
Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Why local conditions change the impact
A modest-looking global share can still matter where a large facility connects to a constrained electricity grid or draws from a stressed watershed. The local effect depends on site demand, climate, cooling design and the electricity supply mix. In the IEA’s example, a 100 MW U.S. hyperscale data centre consumes about 2 million litres of water per day in total, with more than 60% of that amount indirect. That example is not a universal rate for data centres of that size.
Cooling approaches also involve trade-offs. The IEA describes direct expansion cooling as substantially less water-intensive than some airside economiser and adiabatic systems with water-cooled chillers. Direct liquid and immersion cooling can reduce direct water consumption, but no single approach is established as the right choice for every site. Local climate, design and energy supply affect the result.
How to interpret company water and efficiency figures
Company metrics can describe a particular operator’s facilities, but they do not automatically represent the global fleet or the full indirect water footprint estimated by the IEA.
| Reported figure | What it covers | What it does not establish |
|---|---|---|
| Microsoft FY25 PUE: 1.17 | Global metric for data centres Microsoft fully owns and controls that were operational for 12 months; FY25 ran from July 1, 2024, to June 30, 2025. | A universal benchmark or a measure of water consumption. |
| Microsoft FY25 WUE: 0.27 litres per kWh | Global water-use-efficiency metric for the same qualifying facilities. | The IEA’s upstream-inclusive sector water total, or a figure that can be compared without aligning definitions. |
| Google: 4.5 billion gallons replenished in 2024; freshwater-consumption replenishment rose from 18% in 2023 to 64% | Water replenishment figures in Google’s 2025 Environmental Report. | Gross withdrawals, onsite consumption, or a direct measure of water use at a particular data centre. |
PUE is a facility energy-efficiency metric; WUE expresses water use relative to electricity. Their meaning depends on the operator’s definitions and reporting boundary. Microsoft’s figures apply to its specified set of fully owned and controlled sites, not every facility it uses. Microsoft Datacenters, “Measuring energy and water efficiency for Microsoft datacenters”. Google’s replenishment numbers are a different kind of claim from water consumed onsite. Google, 2025 Environmental Report.
Recommended Free Tools
Can you calculate the water or electricity used by one AI prompt?
Not from these global totals. The IEA does not provide a single exact per-prompt figure in the cited reports, and dividing sector-wide use by an assumed number of prompts would not produce a dependable answer. A credible estimate would need to specify the model, hardware, workload, data-centre overhead, location, electricity mix and water-accounting boundary. Training and inference should be treated separately only when a source provides separate figures.
Be cautious with viral claims that one query uses a fixed amount of electricity or water if they omit those details. A per-query estimate can vary with the system and assumptions used, and global data-centre totals cannot be assigned to an individual AI interaction.
Quick Recap
What makes these estimates uncertain?
- Mixed workloads: sector electricity and water totals are not consistently divided into AI and non-AI use.
- Different boundaries: direct cooling, total facility operations and water used upstream to supply electricity or manufacture chips describe different things.
- Uneven disclosure: the IEA notes that water-efficiency reporting is less common than energy or emissions reporting, adding uncertainty to modeled water estimates. IEA, Energy and AI (2025).
- Forecasts are scenarios: the IEA’s 2030 values are projections, not measured future totals.
- Local impacts differ: global averages cannot show whether a particular grid or watershed can accommodate a facility’s demand.
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
