Skip to content
ContentLora

    Tip: press / anywhere to search.

    Explainer

    AI data centers and electricity, explained

    The IEA estimates that data centers used about 485 TWh of electricity in 2025, after 17% growth that year, and expects that to roughly double by 2030.[1][2] AI is the fastest-growing part: AI-focused data centers used 50% more power in 2025, and the power density of AI servers rose 11-fold between 2020 and 2025.[3][4] Supply is a constraint: new power lines, transformers and gas turbines can take years to deliver.[5][6]

    Editor reviewedUpdated Explainer47 claims · 72% primaryEnergy and climate, Artificial intelligence, Semiconductors
    Reading level: 5 passages differ; the rest is the same for everyone.

    This explainer covers how much electricity AI data centers use, why, where the power comes from, cooling and water, and grid limits. For competing forecasts, read How much power will AI need?. For dated 2026 news, see the AI data center power tracker. Choose a reading level in each section.

    How much power AI uses

    Data centers, the buildings that run cloud services and AI, used about 415 TWh of electricity in 2024, around 1.5% of the world’s total.[7] That rose 17% in 2025, to about 485 TWh.[2][1] The IEA expects it to roughly double to 950 TWh by 2030.[1]

    AI is the fastest-growing part. Data centers focused on AI used 50% more electricity in 2025 than in 2024.[3] Gartner, a research firm, estimates that AI servers account for 31% of data center electricity in 2026, up from about 20% a year earlier.[8]

    The IEA puts global data center consumption at 415 TWh in 2024 (about 1.5% of world electricity) and 485 TWh in 2025, with a central projection of 950 TWh in 2030.[7][1] Consumption grew about 12% a year from 2017, more than four times faster than total electricity use.[9] AI-focused sites grew 50% in 2025 and are projected to triple by 2030.[3][10]

    Gartner’s estimate for 2025 is lower, at 447 TWh, rising to 565 TWh in 2026.[11] It has AI-optimised servers going from 95 TWh in 2025 to 175 TWh in 2026, overtaking conventional servers in 2027.[8] Globally, data centers are still only about one-tenth of electricity demand growth to 2030 in the IEA’s view. In the United States they are nearly half.[12][13]

    Why AI chips need so much electricity

    AI runs on servers packed with specialised chips, such as the GPUs made by NVIDIA. These servers keep getting more power-dense: the IEA says AI server power density rose 11 times between 2020 and 2025.[4] By 2027, one server rack the size of a large refrigerator could draw as much power at peak as about 65 homes.[14]

    A single chatbot question uses little energy on its own. Even if every web search became a simple AI query, the IEA estimates the total would be less than 1% of data center electricity.[15]

    The IEA expects electricity use by accelerated servers to grow 30% a year in its base case, against 9% for conventional servers. Accelerated servers account for almost half of the net increase in data center consumption.[16] The power density of AI servers rose 11-fold from 2020 to 2025 and is set to rise another fourfold by 2027.[4]

    AI loads also behave differently. Server load can swing by more than half of rated capacity within a second.[17] AI data centers are shifting to high-voltage direct-current power distribution, and the IEA estimates 20-25 GW of batteries could be installed in data centers by 2030 to handle the swings.[18][19] Chip supply matters too. The IEA says bottlenecks across the AI value chain make its faster near-term growth scenarios less likely, and it notes a shortage of high-bandwidth memory expected to last until at least the end of 2027 (see the memory wall and AI chip supply concentration).[20][21]

    Cooling and water

    AI chips turn power into heat that must be removed: the IEA says an AI data center must remove heat equal to about 30 gas boilers for each server rack.[22] Cooling uses about 7% of the power in efficient large data centers and over 30% in less efficient ones.[23]

    Some cooling systems use water. Evaporative systems let water escape as steam. Closed-loop systems recirculate it.[24] A Berkeley Lab estimate put direct cooling water at 17 billion gallons for US data centers in 2023.[25] Google’s site in Council Bluffs, Iowa used 1 billion gallons in 2024, its highest.[26]

    Cooling’s share of site load ranges from about 7% in efficient hyperscale facilities to over 30% in enterprise data centers.[23] Gartner forecasts cooling electricity rising 22.6% in 2026 to 195 TWh, according to Tom’s Hardware.[27] On water, the LBNL estimate is 17 billion gallons (64 billion litres) of direct use in 2023, possibly doubling or quadrupling by 2028.[25] Indirect use, through the water consumed by power plants, was about 12 times direct use.[28] Evaporative designs consume water; closed-loop designs with air-cooled chillers recirculate it.[24]

    Where the power comes from

    In the IEA’s 2025 outlook, renewable power meets most of the extra demand from data centers to 2035. Natural gas and nuclear add smaller, roughly equal amounts.[29] Tech companies are also signing deals to keep nuclear plants running or make them bigger. Meta announced agreements in January 2026 supporting up to 6.6 GW by 2035.[30] In October 2026 Google contracted 3,590 MW from Constellation, including 890 MW from uprates at existing nuclear plants.[31]

    The IEA’s 2025 base case adds over 450 TWh of renewable generation, 175 TWh of gas and about as much nuclear to serve data centers to 2035.[29] Its 2026 update tracks a US pipeline of onsite gas plants built because grid connections are slow. About one-fifth had started land clearing or construction, and 15-27 GW could be running by 2030.[32] Reliable onsite supply needs 30% to 70% more generation than the load, and most operators still prefer a grid connection.[33] Long-duration storage is part of the picture too: a Google–Xcel agreement includes a 30 GWh Form Energy iron-air battery (see how grid-scale storage works).[34] Microsoft signed a fusion power agreement with Helion in 2023 for a plant expected online by 2028.[35]

    Grid limits

    New transmission lines can take four to eight years in advanced economies, and waits for transformers and cables doubled in the three years to 2025.[5] Gas turbine orders jumped 70% in 2025, which the IEA says exposed chokepoints in equipment supply chains.[36] The IEA estimated in 2025 that about 20% of planned data center projects could face delays unless grid risks are addressed.[37]

    Data centers cluster. Nearly half of US capacity sits in five regional clusters, and half of the US data centers under development are in existing large clusters.[38] Peak load is hard to plan for, because sites fill with servers gradually and often oversize their grid connections at first.[39] Sudden load drops are a reliability risk: NERC issued its highest-level alert in May 2026 after data centers unexpectedly dropped load or swung demand.[40] About 3,800 MW of data center load tripped offline in northern Virginia on July 22, 2026, according to PJM.[41] Regulators are responding with ride-through rules, large-load tariffs and interconnection reforms, tracked in the AI data center power tracker.[42][43][44]

    NVIDIA

    NVIDIA designs GPUs, CPUs and networking for AI data centers. Its history, products, finances and recent developments as of November 2026.

    More in Energy and climate

    Developing in Science & Energy

    Go deeper

    Questions readers ask

    How much electricity do data centers use?

    The IEA estimates data centers used about 485 TWh in 2025, up from 415 TWh in 2024, when they accounted for around 1.5% of the world's electricity. It projects about 950 TWh in 2030, roughly 3% of global demand.[1][7]

    Does a single AI chatbot query use a lot of power?

    Individually, no. The IEA estimates that if every conventional internet search were done as a simple AI text query, it would use less than 4 TWh a year, under 1% of data center consumption. Most of the growth is in AI-focused data centers, whose electricity use rose 50% in 2025.[15][3]

    Which countries use the most data center electricity?

    In 2024 the United States accounted for 45% of global data center electricity use, China 25% and Europe 15%, according to the IEA.[45]

    Do data centers use a lot of water?

    Some do. A 2024 Berkeley Lab estimate put direct cooling water at 17 billion gallons for US data centers in 2023, with indirect use through power generation about 12 times higher. Evaporative cooling consumes water, while closed-loop systems recirculate it.[25][28][24]

    Where does the power for AI data centers come from?

    In the IEA's 2025 base case, renewables meet most of the extra demand to 2035, followed by natural gas and nuclear. In the United States, some developers are building onsite gas plants because grid connections are slow.[29][32]

    Will AI data centers raise my electricity bill?

    It depends on the local grid. The IEA says new load can raise prices where supply is tight but can lower them where there is spare capacity. The EIA forecasts 2026 wholesale prices in the PJM region 41% above 2025.[46][47]

    Sources

    · 47 claims from 10 publishers

    Each numbered claim is a statement we checked against the sources listed with it. Tap a number in the text to see its sources in place.

    1. [1]

      The IEA's April 2026 projection has data centre electricity use roughly doubling from 485 TWh in 2025 to 950 TWh in 2030, about 3% of global electricity demand. confirmedas of 2026-04-16· forecast

    2. [2]

      The IEA estimates that global electricity demand from data centres grew 17% in 2025, in line with its earlier projections. confirmedas of 2026-04-16

    3. [3]

      The IEA estimates that electricity use by AI-focused data centres grew 50% in 2025. confirmedas of 2026-04-16

    4. [4]

      The IEA estimates that the power density of AI servers rose 11-fold between 2020 and 2025 and is set to rise a further fourfold by 2027. confirmedas of 2026-04-16

    5. [5]

      The IEA says new transmission lines take four to eight years in advanced economies, and wait times for transformers and cables doubled in the three years to 2025. confirmedas of 2025-04-10

    6. [6]

      The IEA said in 2025 that turbine deliveries for new gas-fired plants faced lead times of several years, which could push some commissioning beyond 2030. confirmedas of 2025-04-10

    7. [7]

      The IEA estimates that data centres used around 1.5% of the world's electricity in 2024, or 415 TWh. confirmedas of 2025-04-10

    8. [8]

      Gartner estimates that AI-optimised servers account for 31% of data centre electricity use in 2026, up from about 20% a year earlier, rising from 95 TWh in 2025 to 175 TWh in 2026 and 258 TWh in 2027, when they would overtake conventional servers. confirmedas of 2026-07-09· forecast

    9. [9]

      The IEA says global data centre electricity use grew about 12% a year from 2017 to 2024, more than four times faster than total electricity use. confirmedas of 2025-04-10

    10. [10]

      The IEA projects that electricity use by AI-focused data centres will triple between 2025 and 2030, much faster than data centres overall. confirmedas of 2026-04-16· forecast

    11. [11]

      A Gartner forecast reported in July 2026 projects that global data centre electricity use will reach 565 TWh in 2026, up 26% from 447 TWh in 2025. confirmedas of 2026-07-09· forecast

    12. [12]

      Globally, the IEA projects data centres account for around one-tenth of electricity demand growth to 2030, less than industrial motors, air conditioning or electric vehicles. confirmedas of 2025-04-10· forecast

    13. [13]

      The IEA projects that data centres account for nearly half of US electricity demand growth between 2025 and 2030. confirmedas of 2025-04-10· forecast

    14. [14]

      The IEA says a single advanced server rack, the size of a large refrigerator, could by 2027 have peak power demand equal to about 65 households. confirmedas of 2026-04-16· forecast

    15. [15]

      The IEA estimates that if all conventional internet searches were done as simple AI text queries, they would use less than 4 TWh a year, under 1% of current data centre consumption. confirmedas of 2026-04-16

    16. [16]

      In the IEA's 2025 base case, electricity use by accelerated servers, driven mainly by AI, grows 30% a year, against 9% for conventional servers, and accounts for almost half of the net increase in data centre consumption. confirmedas of 2025-04-10· forecast

    17. [17]

      The IEA says AI data centres see repeated swings in server load of more than 50% of rated capacity within a second. confirmedas of 2026-04-16

    18. [18]

      The IEA says AI data centres are shifting towards high-voltage direct-current power distribution, building on technology from electric vehicles and renewables. confirmedas of 2026-04-16

    19. [19]

      The IEA estimates that around 20-25 GW of battery storage could be installed in data centres worldwide by 2030 to manage AI load swings. confirmedas of 2026-04-16· forecast

    20. [20]

      The IEA judges that bottlenecks across the AI value chain make its more aggressive near-term scenarios less likely, despite booming investment and project pipelines. confirmedas of 2026-04-16· interpretation

    21. [21]

      The IEA reported in April 2026 that a shortage of high-bandwidth memory, used in AI chips, had developed over the previous six months and was expected to last at least until the end of 2027. confirmedas of 2026-04-16

    22. [22]

      The IEA says an AI data centre must remove heat equivalent to about 30 natural gas boilers per server rack. confirmedas of 2026-04-16

    23. [23]

      The IEA says cooling accounts for about 7% of electricity use in efficient hyperscale data centres and over 30% in less efficient enterprise data centres. confirmedas of 2025-04-10

    24. [24]

      Evaporative cooling systems consume water by venting it as steam, while closed-loop systems use air-cooled chillers and recirculate the water. reportedas of 2025-08-19

    25. [25]

      The Conversation reported that a 2024 LBNL report estimated US data centres used 17 billion gallons (64 billion litres) of water directly for cooling in 2023, a figure that could double or quadruple by 2028. reportedas of 2025-08-19

    26. [26]

      Google reported that its Council Bluffs, Iowa data centre consumed 1 billion gallons (3.8 billion litres) of water in 2024, the most of any of its sites. reportedas of 2025-08-19

    27. [27]

      Tom's Hardware reported that Gartner forecasts electricity used by data centre cooling to rise 22.6% in 2026 to 195 TWh. reportedas of 2026-07-09· forecast

    28. [28]

      According to the same LBNL estimate, indirect water use through the electricity that powers US data centres was about 12 times their direct cooling use in 2023. reportedas of 2025-08-19

    29. [29]

      In the IEA's 2025 base case, renewables supply over 450 TWh of extra generation for data centres to 2035, natural gas 175 TWh (notably in the United States) and nuclear about the same as gas. confirmedas of 2025-04-10· forecast

    30. [30]

      On January 9, 2026 Meta announced nuclear agreements with Vistra, TerraPower and Oklo supporting up to 6.6 GW of new and existing clean power by 2035, including uprates and life extensions at the Perry, Davis-Besse and Beaver Valley plants. confirmedas of 2026-01-09

    31. [31]

      On October 6, 2026 Google and Constellation announced a 20-year agreement for 890 MW of new capacity from uprates at existing nuclear plants in PJM, with first deliveries expected in 2028, alongside a long-term supply agreement for another 2,700 MW, for 3,590 MW in total. confirmedas of 2026-10-06

    32. [32]

      The IEA says US developers constrained by slow grid connections are pursuing onsite gas generation, that about one-fifth of tracked projects had started land clearing or construction, and that around 15-27 GW of onsite gas may power data centres by 2030, mostly in the United States. confirmedas of 2026-04-16· forecast

    33. [33]

      IEA analysis finds that reliable onsite gas power for data centres requires building 30% to 70% more generation than the load, and that most data centres still prefer to connect to the grid. confirmedas of 2026-04-16

    34. [34]

      A 2026 agreement between Google and Xcel Energy includes a 300 MW / 30 GWh Form Energy iron-air battery, described as the largest battery project by energy capacity announced to date. reportedas of 2026-03-04

    35. [35]

      On 10 May 2023 Helion signed the first fusion power purchase agreement, to supply Microsoft from a plant expected online by 2028 and targeting at least 50 megawatts after a one-year ramp-up, with Constellation as power marketer. confirmedas of 2026-10-10

    36. [36]

      The IEA says gas turbine orders surged 70% in 2025, exposing chokepoints in energy equipment supply chains. confirmedas of 2026-04-16

    37. [37]

      The IEA estimated in 2025 that around 20% of planned data centre projects could face delays unless grid risks are addressed. confirmedas of 2025-04-10· forecast

    38. [38]

      The IEA says nearly half of US data centre capacity is in five regional clusters, and half of US data centres under development are in existing large clusters. confirmedas of 2025-04-10

    39. [39]

      The IEA notes that actual data centre peak load is often uncertain, because sites fill with servers gradually and often oversize their grid connections at first. confirmedas of 2026-04-16

    40. [40]

      On May 4, 2026 NERC issued a Level 3 alert, its highest, requiring grid entities to take seven actions by August 3 after data centres unexpectedly dropped load or swung demand rapidly. reportedas of 2026-05-05

    41. [41]

      About 3,800 MW of data centre load tripped offline in northern Virginia on July 22, 2026 after a fault on a 230-kV line, the largest such event in PJM's history, according to the grid operator. reportedas of 2026-08-12

    42. [42]

      On July 9, 2026 the Public Utility Commission of Texas unanimously approved rules requiring large computational loads in ERCOT to stay connected through grid disturbances; ERCOT said it had seen 28 trips of at least 100 MW since 2023. reportedas of 2026-07-13

    43. [43]

      The Smart Electric Power Alliance counted 77 large-load tariffs pending or in place across 36 US states as of March 2026; state regulators approved 29 in 2025, against 14 from 2018 to 2024. reportedas of 2026-03-31

    44. [44]

      Utility Dive reported that on June 18, 2026 FERC issued six show cause orders finding grid operators' large-load interconnection rules appeared inadequate, naming five issues including cost shifts, co-location and services for flexible loads. reportedas of 2026-06-22

    45. [45]

      The United States accounted for 45% of global data centre electricity use in 2024, China 25% and Europe 15%, according to the IEA. confirmedas of 2025-04-10

    46. [46]

      The IEA says data centre load growth can raise electricity prices where supply is tight, but can lower them where there is spare supply, by using power plants and grids more efficiently. confirmedas of 2026-04-16· interpretation

    47. [47]

      In its October 2026 Short-Term Energy Outlook, the EIA forecast 2026 wholesale electricity prices in the PJM region 41% higher than in 2025, the largest increase of any region it tracks. confirmedas of 2026-10-06· forecast

    Revision history (1)
    1. Page created.

    Created Nov 7, 2026. Last reviewed by an editor on Nov 8, 2026. Next scheduled review: Apr 16, 2027.

    Cite this page

    "AI data centers and electricity, explained." ContentLora, updated Nov 7, 2026. https://contentlora.com/explain/ai-data-centers-and-power-explained

    Spotted an error? Suggest a correction or emailcorrections@contentlora.com.