Skip to main content

Live estimate, not a meter reading

How much electricity and water AI data centers are using right now

There is no global meter for AI data centers. This counter uses IEA year anchors for all data centers, the OWID reading for 2025 AI use, derived AI values for 2024, and IEA water figures, then spreads them across the current second. The IEA itself treats today's totals as estimates.

Sources checked August 21, 2026IEA 2025 / 2026

Since January 1 this year

Calendar year in UTC

Share of the UTC calendar year: 65.4%

Electricity

126.3 billion

kWh

6,123 kWh per second

Annual rate in this calendar year: 193.1 TWh

Water

44.2 billion

liters

2,142 liters per second

Annual rate in this calendar year: 67.5 billion L

What is counted
Water boundary

Facilities with large numbers of accelerator chips for training and inference. That is the IEA’s definition. Water consumed in cooling, mostly through evaporation, and not returned to the same loop.

For scale

What these volumes mean

Households, using the IEA rule of thumb that 100 MW equals 100,000 homes
22,042,043
Olympic swimming pools in the year-to-date water total
17,672
Times the electricity use of all German data centers in 2025 (Bitkom: 21.3 TWh)
9.1×

Data map

Ten layers, one anchor ledger

The live counter comes first, then charts for each question. Tables stay collapsed and exist only for verification. They do not invent new measurements.

IEA curve

Electricity and water through 2030, year by year

The annual anchors come from the IEA reports. Between them sits the CAGR, not a meter reading. That is why 2026, and the live counter, already sit well above 2025.

Electricity for all data centers and AI-focused sites

The area between anchors is interpolated. All-DC 2024/2025/2030 and AI 2030 are IEA values. AI 2025 comes from the OWID reading, AI 2024 from +50 %. The dashed line marks the current calendar year.

All data centers
AI-focused
IEA Key Questions on Energy and AI (2026). In-between years interpolated.
|
CC BY 4.0
gradually.ai
Four years (2026 interpolated), AI at the bottom, the rest on top

2024

415 TWh

AI-focused: 103.3 TWh

2025

485 TWh

AI-focused: 155 TWh

2026

554.8 TWh

AI-focused: 193.1 TWh

2030

950 TWh

AI-focused: 465 TWh

Data centers without an AI focus
AI-focused

IEA Key Questions on Energy and AI (2026). In-between years interpolated.

From 415 TWh in 2024 to 950 TWh in 2030

Increase
Decrease
Total
IEA Key Questions on Energy and AI (2026). In-between years interpolated.
|
CC BY 4.0
gradually.ai

AI share of all data-center electricity

The share is about 32% in 2025 and just under 49% in the IEA 2030 base case.

202424.9%
202532%
202634.8%
203048.9%
IEA Key Questions on Energy and AI (2026). In-between years interpolated.
|
CC BY 4.0
gradually.ai

Electricity in TWh, water in billion liters

Two units, one timeline from 2024 to 2030. The water line reaches 1,200 billion liters in 2030.

All data centers
Water consumption
IEA 2025 and 2026. Water and electricity come from two reports.
|
CC BY 4.0
gradually.ai

Water in 2023 and 2030, same mix

The IEA publishes the split only for 2023. 2030 uses that same mix, as the counter does. That is an assumption, not a new IEA figure.

25%67%8%202356025%67%8%20301,200
Direct cooling
Power plants
Hardware
IEA Energy and AI (2025). Consumption (not withdrawals).
|
CC BY 4.0
gradually.ai

2025 versus 2030

20252030AI-focused155 TWh465 TWh+310.0 TWhAll data centers485 TWh950 TWh+465.0 TWh
IEA Key Questions on Energy and AI (2026). In-between years interpolated.
|
CC BY 4.0
gradually.ai

Watt-hours per text query

Median figures for short text prompts. Training is not included.

Gemini Apps, Google
0.24 Wh
GPT-4o, Epoch AI
0.3 Wh
ChatGPT, OpenAI
0.34 Wh
OpenAI 2025, Epoch AI, Google Gemini Apps (August 22, 2025).
|
CC BY 4.0
gradually.ai

Watt-hours per agent task, GPU only

IEA Figure 2.1. This is GPU electricity, not a full facility load.

Agent without reasoning
1.14 Wh
Agent with reasoning
50 Wh
IEA, Key Questions on Energy and AI, Figure 2.1.
|
CC BY 4.0
gradually.ai

Germany 2025 against the world totals

Bitkom counts every German data center. The world totals on the right are IEA figures, excluding crypto mining.

Germany 2025, all data centers
21.3 TWh
AI worldwide, 2026 interpolated
193.1 TWh
All data centers worldwide, 2026 interpolated
554.8 TWh
All data centers worldwide, 2030
950 TWh
Bitkom/Borderstep 2025 and IEA 2026.
|
CC BY 4.0
gradually.ai

More anchors

What else sits in the same IEA figures

New charts are not new measurements. Everything here comes from the IEA, Our World in Data, and Bitkom. Years in between, and the water mix after 2023, are interpolation or an assumption.

AI electricity from 2024 to 465 TWh

2024 follows from 155 TWh in 2025 and the IEA growth of 50%. The counter uses that same derivation.

Increase
Decrease
Total
IEA Key Questions on Energy and AI (2026). In-between years interpolated.
|
CC BY 4.0
gradually.ai

AI and the rest, stacked over time

2025 is 155 TWh of AI load and 330 TWh without an AI focus. The CAGR sits between the anchors.

AI-focused
Data centers without an AI focus
IEA Key Questions on Energy and AI (2026). In-between years interpolated.
|
CC BY 4.0
gradually.ai

Electricity against water, year by year

Each point is a year. Water values from 2024 through 2029 sit on the CAGR between 560 and 1,200.

IEA 2025 and 2026. Water and electricity come from two reports.
|
CC BY 4.0
gradually.ai

2025: 155 versus 330 TWh

The IEA total of 485 TWh splits into the OWID AI reading and the remainder.

IEA Key Questions on Energy and AI (2026). In-between years interpolated.
|
CC BY 4.0
gradually.ai

Watt-hours from a text query to an agent

The first three values are text prompts. 1.14 Wh and 50 Wh are GPU electricity from IEA Figure 2.1.

Gemini Apps, Google0.24 Wh
GPT-4o, Epoch AI0.3 Wh
ChatGPT, OpenAI0.34 Wh
Agent without reasoning1.14 Wh
Agent with reasoning50 Wh

OpenAI, Epoch AI, Google Gemini Apps, and IEA Figure 2.1.

2023 water consumption by source

The published IEA split. This is not withdrawals.

560billion L
Direct cooling14025%
Power plants37366.6%
Hardware478.4%
IEA Energy and AI (2025). Consumption (not withdrawals).
|
CC BY 4.0
gradually.ai

Consumption versus withdrawals, 2023

The IEA puts withdrawals at about 5,500 billion liters. Consumption is 560. The live ticker uses consumption only.

Consumption 2023
560 billion L
Withdrawals 2023, about
5,500 billion L
IEA Energy and AI (2025). Consumption (not withdrawals).
|
CC BY 4.0
gradually.ai

Water through 2030, 2023 mix held constant

The IEA publishes the mix only for 2023. The areas after that scale the same shares. That is an assumption.

Direct cooling
Power plants
Hardware
IEA Energy and AI (2025). Consumption (not withdrawals).
|
CC BY 4.0
gradually.ai

Total consumption and direct cooling

The direct line stays at 25% of the annual total because the mix is held fixed.

Water consumption
Direct cooling
IEA Energy and AI (2025). Consumption (not withdrawals).
|
CC BY 4.0
gradually.ai

What the ticker takes from 2023

Direct 140, operational 513, including hardware 560. Hardware stays out of the ticker.

Direct cooling only
140 billion L
Cooling plus power plants
513 billion L
Consumption including hardware
560 billion L
IEA Energy and AI (2025). Consumption (not withdrawals).
|
CC BY 4.0
gradually.ai

Milliliters of water per text query

Gemini publishes 0.26 ml. ChatGPT converts OpenAI’s 0.000085 gallons into milliliters.

Gemini Apps, Google
0.26 ml
ChatGPT, OpenAI
0.32 ml
Google Gemini Apps (August 22, 2025) and OpenAI 2025.
|
CC BY 4.0
gradually.ai

From withdrawals to direct cooling, 2023

One year, one source. No 2025 AI share is laid over these water totals.

IEA Energy and AI (2025). Consumption (not withdrawals).
|
CC BY 4.0
gradually.ai

IEA reference sites of 100 MW

Average annual power divided by 100 MW. That is an order of magnitude, not a site census.

AI 2025
176.9 sites
All 2025
553.7 sites
AI 2030
530.8 sites
All 2030
1,084.5 sites
IEA Key Questions 2026 and Energy and AI 2025. 100 MW is the rule of thumb.
|
CC BY 4.0
gradually.ai

Germany 2025 inside the world total

21.3 TWh is about 4.4% of the 485 TWh worldwide. Bitkom counts every German data center. The IEA counts the world without crypto mining.

Germany 2025, all data centers
21.3 TWh
World 2025, all data centers
485 TWh
Bitkom/Borderstep 2025 and IEA 2026.
|
CC BY 4.0
gradually.ai

What the IEA means by one site

Rules of thumb from Energy and AI, not the live ticker.

Reference capacity
100 MW
Household equivalent at 100 MW
100,000
Water per day in the IEA illustration
2,000,000 L
Households for that daily water volume
6,500

Arithmetic from the anchors

CAGR, per-second rates, and unit conversions

New tables are not new measurements. CAGRs, per-second rates, and query conversions are division. AI-attributed water is the same assumption the ticker uses.

Remainder without an AI focus, 2024 to 2030

Total minus AI. In 2025 that is the 330 TWh Our World in Data reads beside 155 TWh.

Increase
Decrease
Total
IEA Key Questions on Energy and AI (2026). In-between years interpolated.
|
CC BY 4.0
gradually.ai

Water consumption from 560 to 1,200

Only the two IEA anchors. The CAGR between them is what the ticker uses.

Increase
Decrease
Total
IEA Energy and AI (2025). Consumption (not withdrawals).
|
CC BY 4.0
gradually.ai

Water in 2023 versus 2030

The IEA publishes the mix only for 2023. The 2030 direct line keeps that same 25%. That is an assumption.

20232030Water consumption560 billion L1,200 billion L+640.0 billion LDirect cooling140 billion L300 billion L+160.0 billion L
IEA Energy and AI (2025). Consumption (not withdrawals).
|
CC BY 4.0
gradually.ai

2025 electricity and 2023 water, in one view

The 485 TWh total on the left. The published water split on the right.

AI-focused
Data centers without an AI focus
IEA Key Questions on Energy and AI (2026). In-between years interpolated.
|
CC BY 4.0
gradually.ai
Direct cooling
Power plants
Hardware
IEA Energy and AI (2025). Consumption (not withdrawals).
|
CC BY 4.0
gradually.ai

Four years, five measures

All-DC 2024/2025/2030 and AI 2030 are IEA anchors. AI 2025 is the OWID reading, AI 2024 derived (+50 %). 2026 sits on the CAGR. Water in 2024 and 2026 sits between 560 and 1,200.

All, TWh950415AI, TWh465103.3AI share, %49%25%Water, billion L1,200624.4Power, MW108,447.547,245
2024
2025
2026
2030
IEA 2025 and 2026. Water and electricity come from two reports.
|
CC BY 4.0
gradually.ai

945 TWh versus 950 TWh

Our World in Data still captions 2030 as 945 TWh on one chart. That is the April 2025 IEA total. The 2026 update uses 950 TWh.

OWID caption, IEA April 2025
945 TWh
IEA Key Questions, April 2026
950 TWh
Gap
5 TWh

Withdrawals divided by consumption, 2023

About 5,500 divided by 560. The ticker uses consumption only.

9.8×

IEA site water on a 365-day year

2 million liters a day, times 365. Leap days are not in this illustration. Then the 2023 global consumption total divided by that annual volume.

One site, liters in a 365-day year
730,000,000 L
Such sites inside 560 billion liters
767.1

CAGR, per-second rates, and unit conversions · The waterfalls and sunburst above visualize the same anchors. Here are CAGR, year series, and query conversions as tables.

Show tables (5)

Growth between the published years

A one-year span is a simple increase. Multi-year spans are CAGRs. The 50% AI rise in 2025 is an IEA figure.

SpanStartEndYearsRate
All data centers, 2024-2025415 TWh485 TWh116.9%
All data centers, 2025-2030485 TWh950 TWh514.4%
AI-focused, 2024-2025103.3 TWh155 TWh150%
AI-focused, 2025-2030155 TWh465 TWh524.6%
Water consumption, 2023-2030560 billion L1,200 billion L711.5%

Electricity, every year from 2024 through 2030

All-DC 2024/2025/2030 and AI 2030 are IEA anchors. AI 2025 is the OWID reading, AI 2024 derived (+50 %). The years in between sit on the CAGR. Delta is the change from the previous year.

YearAll data centersAI-focusedRemainder without an AI focusAI shareDelta, all, TWhDelta, AI, TWhkWh/s, allkWh/s, AITimes Germany 2025, allTimes Germany 2025, AI
2024415 TWh103.3 TWh311.7 TWh24.9%--13,123.6 kWh/s3,267.7 kWh/s19.5×4.9×
2025485 TWh155 TWh330 TWh32%70 TWh51.7 TWh15,379.2 kWh/s4,915 kWh/s22.8×7.3×
2026554.8 TWh193.1 TWh361.7 TWh34.8%69.8 TWh38.1 TWh17,592.7 kWh/s6,122.8 kWh/s26×9.1×
2027634.7 TWh240.5 TWh394.1 TWh37.9%79.8 TWh47.4 TWh20,124.6 kWh/s7,627.3 kWh/s29.8×11.3×
2028726 TWh299.6 TWh426.3 TWh41.3%91.3 TWh59.1 TWh22,958.1 kWh/s9,475.7 kWh/s34.1×14.1×
2029830.5 TWh373.3 TWh457.2 TWh44.9%104.5 TWh73.6 TWh26,334.2 kWh/s11,836.5 kWh/s39×17.5×
2030950 TWh465 TWh485 TWh48.9%119.5 TWh91.7 TWh30,124.3 kWh/s14,745.1 kWh/s44.6×21.8×

Water, every year from 2023 through 2030

The mix after 2023 is the 2023 split. AI columns start in 2024, when the electricity trajectory begins. AI 2024 is derived (+50 %), not an anchor. 2023 gets no AI share.

YearTotal water consumptionDirect coolingPower plantsHardwareLiters/s, totalAI direct, assumedAI operational, assumed
2023560 billion L140 billion L373 billion L47 billion L17,757.5 L/sno electricity anchorno electricity anchor
2024624.4 billion L156.1 billion L415.9 billion L52.4 billion L19,745.9 L/s38.9 billion L142.4 billion L
2025696.2 billion L174.1 billion L463.7 billion L58.4 billion L22,077.6 L/s55.6 billion L203.8 billion L
2026776.3 billion L194.1 billion L517.1 billion L65.2 billion L24,617 L/s67.5 billion L247.5 billion L
2027865.6 billion L216.4 billion L576.6 billion L72.7 billion L27,448.6 L/s82 billion L300.5 billion L
2028965.2 billion L241.3 billion L642.9 billion L81 billion L30,522.3 L/s99.6 billion L364.9 billion L
20291,076.2 billion L269.1 billion L716.8 billion L90.3 billion L34,126.4 L/s120.9 billion L443.1 billion L
20301,200 billion L300 billion L799.3 billion L100.7 billion L38,051.8 L/s146.8 billion L538.1 billion L

Queries in units of the published figures

1 kWh divided by the providers’ watt-hours. That is a unit conversion, not a world-fleet load profile.

ChatGPT queries per kWh2,941.2
GPT-4o queries per kWh, Epoch3,333.3
Gemini queries per kWh4,166.7
Agent without reasoning, in ChatGPT queries3.4
Agent with reasoning, in ChatGPT queries147.1

One UTC second of 2025 AI electricity in ChatGPT queries

The 155 TWh annual total, spread evenly across the seconds, divided by 0.34 Wh. Scale only.

14.5 million

Intensity and time

Multiples, liters per kWh, and hours

New tables are not new measurements. Intensities, multiples, and time equivalents are division. Liters per kWh divide interpolated water by electricity from a different IEA report.

Where 2025 and 2023 sit against 2030

Electricity compares 2025 with 2030. Water compares 2023 with 2030, because that is where the water anchor sits.

All data centers 2025 of 203051.1%
AI electricity 2025 of 203033.3%
Water 2023 of 203046.7%
IEA 2025 and 2026. Water and electricity come from two reports.
|
CC BY 4.0
gradually.ai

Anchor end points, divided

465 TWh (IEA 2030) divided by 155 TWh (OWID reading 2025) is exactly 3. All-DC and water use IEA anchors.

All data centers, 2030 over 2024
2.3×
All data centers, 2030 over 2025
2×
AI electricity, 2030 over 2025
3×
Water, 2030 over 2023
2.1×
IEA 2025 and 2026. Water and electricity come from two reports.
|
CC BY 4.0
gradually.ai

Remainder without an AI focus, 2025 versus 2030

330 TWh in 2025 and 485 TWh in 2030. Both are total minus AI.

20252030Remainder without an AI focus330 TWh485 TWh+155.0 TWh
IEA Key Questions on Energy and AI (2026). In-between years interpolated.
|
CC BY 4.0
gradually.ai

Liters per kWh, 2024 versus 2030

That year's water divided by that year's electricity. This is not an IEA intensity.

20242030Consumption per kWh, all1.5 L/kWh1.3 L/kWh-0.2 L/kWhDirect cooling per kWh0.4 L/kWh0.3 L/kWh-0.1 L/kWh
IEA 2025 and 2026. Water and electricity come from two reports.
|
CC BY 4.0
gradually.ai

Multiples, liters per kWh, and hours · The radial and slope charts above show progress and multiples. Here are liters per kWh, rest growth, and time equivalents as tables.

Show tables (9)

Growth in the remainder only

2024 remainder is derived. 2025 and 2030 remainder sit in the OWID reading and the IEA difference.

SpanStartEndYearsRate
Remainder without an AI focus, 2024-2025311.7 TWh330 TWh15.9%
Remainder without an AI focus, 2025-2030330 TWh485 TWh58%

Intensity and year-on-year rates, 2024 through 2030

No 2023, because the electricity anchor starts later. YoY for all and AI follows each series CAGR from 2026. Rest YoY is the year-on-year growth of (all minus AI), not all YoY minus AI YoY, and is not a constant CAGR.

YearLiters/kWh, totalLiters/kWh, directYoY, allYoY, AIYoY, rest
20241.5 L/kWh0.4 L/kWh---
20251.4 L/kWh0.4 L/kWh16.9%50%5.9%
20261.4 L/kWh0.3 L/kWh14.4%24.6%9.6%
20271.4 L/kWh0.3 L/kWh14.4%24.6%9%
20281.3 L/kWh0.3 L/kWh14.4%24.6%8.2%
20291.3 L/kWh0.3 L/kWh14.4%24.6%7.2%
20301.3 L/kWh0.3 L/kWh14.4%24.6%6.1%

Power, sites, and households, every year

Power and sites follow the 100 MW rule of thumb. The Germany column holds Bitkom 2025 fixed.

YearAverage power, allAverage power, AI100 MW equivalents, all100 MW equivalents, AIHousehold equivalents, allBitkom 2025 / world year
202447,245 MW11,764 MW472.4117.647,244,9915.1%
202555,365 MW17,694 MW553.7176.955,365,2974.4%
202663,334 MW22,042 MW633.3220.463,333,5753.8%
202772,449 MW27,458 MW724.5274.672,448,6643.4%
202882,649 MW34,112 MW826.5341.182,649,1772.9%
202994,803 MW42,611 MW948426.194,803,2282.6%
2030108,447 MW53,082 MW1,084.5530.8108,447,4892.2%

Water growth, sites, and pools

The rate after 2023 is the same CAGR every year. Site counts use the 365-day year. Pools per day use UTC days.

YearYoY, water365-day sitesOlympic poolsPools per day
2023-767.1224,000613.7
202411.5%855.4249,766682.4
202511.5%953.8278,495763
202611.5%1,063.5310,529850.8
202711.5%1,185.8346,248948.6
202811.5%1,322.2386,0751,054.8
202911.5%1,474.3430,4831,179.4
203011.5%1,643.8480,0001,315.1

Germany 2025 in hours of the world fleet

21.3 TWh from Bitkom 2025, placed on the 2025 world annual total. Two sources.

Hours of 2025 world AI1,203.8 h
Hours of 2025 world, all384.7 h
Days of 2025 world AI50.2
Days of 2025 world, all16

One query inside the 2025 UTC second

The 155 TWh AI annual total, spread evenly across the seconds, in query units. Scale only.

Microseconds per ChatGPT query0.1 µs
Microseconds per agent with reasoning10.2 µs
Gemini queries per ChatGPT query1.4
GPT-4o queries per ChatGPT query, Epoch1.1

One 100 MW site in a UTC year

100 MW across every second of the UTC year. 2024 and 2028 are leap years.

2025, not a leap year0.9 TWh
2024, leap year0.9 TWh

2023 withdrawals, rates only

2023 only. No division by a later electricity total.

Liters/s, withdrawals174,403.9 L/s
365-day sites in withdrawals7,534.2
Olympic pools, withdrawals2,200,000

5 TWh as a share

The same gap, once against 945 and once against 950.

5 TWh of 945 TWh0.5%
5 TWh of 950 TWh0.5%

Scale and remainder

Daily averages, shares, and query equivalents

Another layer of division. Daily values split year anchors by UTC days. Query counts split AI kWh by published Wh per request. No load profiles.

485 TWh in 2025 as 100 dots

Each dot is 4.85 TWh. Colors match the AI/rest split used in the electricity charts. 155 is the OWID reading, 330 is the remainder (485 minus 155).

485 TWh in 2025 as 100 dots

100 dots = 485 TWh. Each dot = 4.9 TWh.

Rest without AI focusRest without AI focusAI-focused
Rest without AI focus330 TWh(68%, 68 dots)
AI-focused155 TWh(32%, 32 dots)
IEA Key Questions on Energy and AI (2026). In-between years interpolated.
|
CC BY 4.0
gradually.ai

AI-to-rest ratio, 2024 vs 2030

2024: derived AI divided by rest. 2030: 465 AI divided by 485 rest (950 minus 465). A TWh ratio, not market share.

20242030AI / rest0.3 ×1.0 ×+0.6 ×
IEA Key Questions on Energy and AI (2026). In-between years interpolated.
|
CC BY 4.0
gradually.ai

155 TWh in 2025 as requests per year

Same annual total, different Wh per request. Agent and reasoning are IEA estimates; ChatGPT and Gemini from OpenAI and Google.

Gemini, 0.24 Wh
645,833.3 billion
Epoch, 0.3 Wh
516,666.7 billion
ChatGPT, 0.34 Wh
455,882.4 billion
Agent, 1.14 Wh
135,964.9 billion
Reasoning, 50 Wh
3,100 billion
OpenAI 2025, Epoch AI, Google Gemini Apps (August 22, 2025). Agent and reasoning: IEA, Key Questions on Energy and AI, Figure 2.1.
|
CC BY 4.0
gradually.ai

Daily averages, shares, and query equivalents · The waffle and query bars above show shares and query equivalents. Here are daily averages, gaps to 2030, and water mix as tables.

Show tables (7)

Electricity as daily average, UTC year

Year TWh divided by UTC calendar days. No seasonality.

YearAll, kWh/dayAI-focused, kWh/dayRest, kWh/day
20241,133,879,781.4 kWh/day282,331,511.8 kWh/day851,548,269.6 kWh/day
20251,328,767,123.3 kWh/day424,657,534.2 kWh/day904,109,589 kWh/day
20261,520,005,808.3 kWh/day529,009,029.2 kWh/day990,996,779.1 kWh/day
20271,738,767,927.6 kWh/day659,002,915 kWh/day1,079,765,012.6 kWh/day
20281,983,580,239.7 kWh/day818,697,314.1 kWh/day1,164,882,925.6 kWh/day
20292,275,277,475.6 kWh/day1,022,670,756.8 kWh/day1,252,606,718.9 kWh/day
20302,602,739,726 kWh/day1,273,972,602.7 kWh/day1,328,767,123.3 kWh/day

Water as daily average, 2023 to 2030

Consumption in billion L, interpolated between 2023 and 2030.

YearTotal, L/dayDirect, L/day
20231,534,246,575.3 L/day383,561,643.8 L/day
20241,706,049,733.3 L/day426,512,433.3 L/day
20251,907,500,470.3 L/day476,875,117.6 L/day
20262,126,911,401.2 L/day531,727,850.3 L/day
20272,371,560,153.8 L/day592,890,038.5 L/day
20282,637,124,718.3 L/day659,281,179.6 L/day
20292,948,516,999.5 L/day737,129,249.9 L/day
20303,287,671,232.9 L/day821,917,808.2 L/day

AI share and ratio, every year

Share is AI divided by all. pp is the year-on-year change in percentage points.

YearAI shareΔ shareAI / rest
202424.9%-0.3×
202532%+7.1 pp0.5×
202634.8%+2.8 pp0.5×
202737.9%+3.1 pp0.6×
202841.3%+3.4 pp0.7×
202944.9%+3.7 pp0.8×
203048.9%+4 pp1×

AI year electricity in requests, 2024 to 2030

Five Wh assumptions applied to the same AI year total (anchor, derived, or CAGR).

YearChatGPT, 0.34 WhGemini, 0.24 WhEpoch, 0.3 WhAgent, 1.14 WhReasoning, 50 Wh
2024303,921.6 billion430,555.6 billion344,444.4 billion90,643.3 billion2,066.7 billion
2025455,882.4 billion645,833.3 billion516,666.7 billion135,964.9 billion3,100 billion
2026567,906.8 billion804,534.6 billion643,627.7 billion169,375.7 billion3,861.8 billion
2027707,459 billion1,002,233.6 billion801,786.9 billion210,996.5 billion4,810.7 billion
2028881,303.6 billion1,248,513.4 billion998,810.7 billion262,844.9 billion5,992.9 billion
20291,097,867.1 billion1,555,311.8 billion1,244,249.4 billion327,434.1 billion7,465.5 billion
20301,367,647.1 billion1,937,500 billion1,550,000 billion407,894.7 billion9,300 billion

Still missing by 2030

2030 anchor minus current year. Water only from 2023 where the anchor starts.

YearAll, TWh missingAI, TWh missingRest, TWh missingWater, bn L missing
2024535 TWh361.7 TWh173.3 TWh575.6 billion L
2025465 TWh310 TWh155 TWh503.8 billion L
2026395.2 TWh271.9 TWh123.3 TWh423.7 billion L
2027315.3 TWh224.5 TWh90.9 TWh334.4 billion L
2028224 TWh165.4 TWh58.7 TWh234.8 billion L
2029119.5 TWh91.7 TWh27.8 TWh123.8 billion L
20300 TWh0 TWh0 TWh0 billion L

Anchor snapshot 2024, 2025, 2030

One row per metric. Water in 2024 is interpolated, not a standalone anchor.

Metric202420252030
All data centers, TWh415 TWh485 TWh950 TWh
AI-focused, TWh103.3 TWh155 TWh465 TWh
Rest without AI, TWh311.7 TWh330 TWh485 TWh
AI share24.9%32%48.9%
Water, bn L624.4 billion L696.2 billion L1,200 billion L
Power, MW47,245 MW55,365 MW108,447 MW

Water mix shares, fixed

140 / 373 / 47 from the IEA water report. Every row is identical because the mix is fixed.

YearDirectPower/heatHardware
202325%66.6%8.4%
202425%66.6%8.4%
202525%66.6%8.4%
202625%66.6%8.4%
202725%66.6%8.4%
202825%66.6%8.4%
202925%66.6%8.4%
203025%66.6%8.4%

Index and pace

Month grids, progress, and query ranking

Layer six keeps the uniform spread. Month and week values divide year anchors. Heatmaps look flat because the model is flat, not because the market has no seasonality.

485 TWh in 2025, spread evenly across 12 months

Each month = annual total divided by 12. No season, just another grid.

JanFebMarAprMayJunJulAugSepOctNovDec202540.41666666666666440.41666666666666440.41666666666666440.41666666666666440.41666666666666440.41666666666666440.41666666666666440.41666666666666440.41666666666666440.41666666666666440.41666666666666440.416666666666664
TWh per month: lowhigh(max 40)
IEA Key Questions on Energy and AI (2026). In-between years interpolated.
|
CC BY 4.0
gradually.ai

155 TWh AI in 2025, spread evenly across 12 months

Same assumption as above, for the AI-focused share only.

JanFebMarAprMayJunJulAugSepOctNovDec202512.91666666666666612.91666666666666612.91666666666666612.91666666666666612.91666666666666612.91666666666666612.91666666666666612.91666666666666612.91666666666666612.91666666666666612.91666666666666612.916666666666666
AI TWh per month: lowhigh(max 13)
IEA 2025 and 2026. Water and electricity come from two reports.
|
CC BY 4.0
gradually.ai

2030 vs index base (power 2024, water 2023)

100 = base year. Water in 2030 is more than double 2023.

AI-focused
450 index
All data centers
228.9 index
Water
214.3 index
Rest without AI
155.6 index
IEA 2025 and 2026. Water and electricity come from two reports.
|
CC BY 4.0
gradually.ai

Share of 2030 anchors already reached

Power compares 2024 to 2030. Water compares 2023 to 2030. No new anchors.

20242030All data centers43.7%100%+56.3%AI-focused22.2%100%+77.8%
IEA Key Questions on Energy and AI (2026). In-between years interpolated.
|
CC BY 4.0
gradually.ai

Query ranking by Wh assumption, 2024 to 2030

Rank 1 = most requests per year at the same AI TWh. Gemini wins on 0.24 Wh, reasoning sits last on 50 Wh.

1.2.3.4.5.2024202520262030ChatGPTGeminiEpochAgentReasoning
OpenAI 2025, Epoch AI, Google Gemini Apps (August 22, 2025). Agent and reasoning: IEA, Key Questions on Energy and AI, Figure 2.1.
|
CC BY 4.0
gradually.ai

Water progress, 2023 vs 2030

Power compares 2024 to 2030. Water compares 2023 to 2030. No new anchors.

20232030Water46.7%100%+53.3%
IEA Energy and AI (2025). Consumption (not withdrawals).
|
CC BY 4.0
gradually.ai

Month grids, progress, and query ranking · The heatmaps and bump chart above show index and rank. Here are monthly, weekly, and progress values as tables.

Show tables (7)

Index 2024 = 100 (power), 2023 = 100 (water)

2023 only has the water index. No electricity anchor there.

YearAll, indexAI, indexRest, indexWater, index
2023---100 index
2024100 index100 index100 index111.5 index
2025116.9 index150 index105.9 index124.3 index
2026133.7 index186.9 index116.1 index138.6 index
2027152.9 index232.8 index126.5 index154.6 index
2028174.9 index290 index136.8 index172.4 index
2029200.1 index361.2 index146.7 index192.2 index
2030228.9 index450 index155.6 index214.3 index

Electricity as monthly average, uniform

TWh divided by 12. No calendar month profile.

YearAll, TWh/monthAI, TWh/monthRest, TWh/month
202434.6 TWh/month8.6 TWh/month26 TWh/month
202540.4 TWh/month12.9 TWh/month27.5 TWh/month
202646.2 TWh/month16.1 TWh/month30.1 TWh/month
202752.9 TWh/month20 TWh/month32.8 TWh/month
202860.5 TWh/month25 TWh/month35.5 TWh/month
202969.2 TWh/month31.1 TWh/month38.1 TWh/month
203079.2 TWh/month38.8 TWh/month40.4 TWh/month

Water as monthly average, 2023 to 2030

Billion liters divided by 12 at a fixed mix.

YearTotal, bn L/monthDirect, bn L/monthPower/heat, bn L/monthHardware, bn L/month
202346.7 bn L/month11.7 bn L/month31.1 bn L/month3.9 bn L/month
202452 bn L/month13 bn L/month34.7 bn L/month4.4 bn L/month
202558 bn L/month14.5 bn L/month38.6 bn L/month4.9 bn L/month
202664.7 bn L/month16.2 bn L/month43.1 bn L/month5.4 bn L/month
202772.1 bn L/month18 bn L/month48 bn L/month6.1 bn L/month
202880.4 bn L/month20.1 bn L/month53.6 bn L/month6.8 bn L/month
202989.7 bn L/month22.4 bn L/month59.7 bn L/month7.5 bn L/month
2030100 bn L/month25 bn L/month66.6 bn L/month8.4 bn L/month

Electricity as weekly average, UTC year

TWh divided by UTC days/7. Leap years change the week count slightly, not the 7-day week.

YearAll, TWh/weekAI, TWh/weekRest, TWh/week
20247.9 TWh/week2 TWh/week6 TWh/week
20259.3 TWh/week3 TWh/week6.3 TWh/week
202610.6 TWh/week3.7 TWh/week6.9 TWh/week
202712.2 TWh/week4.6 TWh/week7.6 TWh/week
202813.9 TWh/week5.7 TWh/week8.2 TWh/week
202915.9 TWh/week7.2 TWh/week8.8 TWh/week
203018.2 TWh/week8.9 TWh/week9.3 TWh/week

Progress to 2030, every year

Percent = year value divided by 2030 anchor. Water from 2023.

YearAll, % of 2030AI, % of 2030Water, % of 2030
2023--46.7%
202443.7%22.2%52%
202551.1%33.3%58%
202658.4%41.5%64.7%
202766.8%51.7%72.1%
202876.4%64.4%80.4%
202987.4%80.3%89.7%
2030100%100%100%

Rest power and 100 MW sites

Rest MW = all MW minus AI MW. Sites follow the 100 MW rule of thumb.

YearRest, MWRest, 100 MW sites
202435,481 MW354.8
202537,671 MW376.7
202641,292 MW412.9
202744,990 MW449.9
202848,537 MW485.4
202952,192 MW521.9
203055,365 MW553.7

AI-attributed water, 2024 to 2030

AI share of year water times mix. 2023 omitted because no AI power anchor.

YearAI direct, bn LAI operational, bn L
202438.9 billion L142.4 billion L
202555.6 billion L203.8 billion L
202667.5 billion L247.5 billion L
202782 billion L300.5 billion L
202899.6 billion L364.9 billion L
2029120.9 billion L443.1 billion L
2030146.8 billion L538.1 billion L

Benchmarks and ratios

Ratios, hours, Germany, and remainder to 2030

Layer seven keeps dividing. Germany stays at Bitkom 21.3 TWh in 2025 everywhere. No German time series, only world year divided.

Electricity ratios, 2024 vs 2030

All/AI and rest/AI as pure TWh quotients. Not market share.

20242030All / AI4.0 ×2.0 ×-2.0 ×Rest / AI3.0 ×1.0 ×-2.0 ×
IEA Key Questions on Energy and AI (2026). In-between years interpolated.
|
CC BY 4.0
gradually.ai

World year divided by Germany 2025

21.3 TWh Bitkom 2025 in the denominator, each world year in the numerator. Germany does not grow here.

2030
44.6×
2029
39×
2028
34.1×
2027
29.8×
2026
26×
2025
22.8×
2024
19.5×
Bitkom/Borderstep 2025 and IEA 2026.
|
CC BY 4.0
gradually.ai

Requests per AI TWh in 2025

Annual query equivalent divided by 155 TWh of AI in 2025 (OWID reading of IEA Fig. 1.5).

Gemini
4,166.7 billion
Epoch
3,333.3 billion
ChatGPT
2,941.2 billion
Agent
877.2 billion
Reasoning
20 billion
OpenAI 2025, Epoch AI, Google Gemini Apps (August 22, 2025). Agent and reasoning: IEA, Key Questions on Energy and AI, Figure 2.1.
|
CC BY 4.0
gradually.ai

Ratios, hours, Germany, and remainder to 2030 · The slope and bar charts above show ratios and Germany multiples. Here are hourly values, UTC calendar facts, and query density as tables.

Show tables (10)

Ratios per calendar year

All/AI, rest/AI, and all/rest from published, derived, and CAGR-interpolated TWh.

YearAll / AIRest / AIAll / rest
20244×3×1.3×
20253.1×2.1×1.5×
20262.9×1.9×1.5×
20272.6×1.6×1.6×
20282.4×1.4×1.7×
20292.2×1.2×1.8×
20302×1×2×

Electricity as hourly average, UTC year

TWh divided by UTC hours. No intraday profile.

YearAll, kWh/hAI, kWh/hRest, kWh/h
202447,244,990.9 kWh/h11,763,813 kWh/h35,481,177.9 kWh/h
202555,365,296.8 kWh/h17,694,063.9 kWh/h37,671,232.9 kWh/h
202663,333,575.3 kWh/h22,042,042.9 kWh/h41,291,532.5 kWh/h
202772,448,663.6 kWh/h27,458,454.8 kWh/h44,990,208.9 kWh/h
202882,649,176.7 kWh/h34,112,388.1 kWh/h48,536,788.6 kWh/h
202994,803,228.2 kWh/h42,611,281.5 kWh/h52,191,946.6 kWh/h
2030108,447,488.6 kWh/h53,082,191.8 kWh/h55,365,296.8 kWh/h

Water as hourly average, 2023 to 2030

Billion liters divided by UTC hours.

YearTotal, L/h
202363,926,940.6 L/h
202471,085,405.6 L/h
202579,479,186.3 L/h
202688,621,308.4 L/h
202798,815,006.4 L/h
2028109,880,196.6 L/h
2029122,854,875 L/h
2030136,986,301.4 L/h

Absolute year-on-year gains

TWh or billion liters vs prior year. Power deltas start in 2025 because the anchor begins in 2024.

YearAll, Δ TWhAI, Δ TWhRest, Δ TWhWater, Δ bn L
2023----
2024---64.4 billion L
202570 TWh51.7 TWh18.3 TWh71.8 billion L
202669.8 TWh38.1 TWh31.7 TWh80.1 billion L
202779.8 TWh47.4 TWh32.4 TWh89.3 billion L
202891.3 TWh59.1 TWh32.2 TWh99.6 billion L
2029104.5 TWh73.6 TWh30.9 TWh111 billion L
2030119.5 TWh91.7 TWh27.8 TWh123.8 billion L

Germany 2025 as fixed divisor

World/AI/rest divided by 21.3 TWh. Only the denominator is country-specific.

YearWorld / DEAI / DERest / DE
202419.5×4.9×14.6×
202522.8×7.3×15.5×
202626×9.1×17×
202729.8×11.3×18.5×
202834.1×14.1×20×
202939×17.5×21.5×
203044.6×21.8×22.8×

Olympic pools and pools per day

2.5 million liters per pool. Growth follows the water CAGR.

YearPools totalPools / day
2023224,000613.7
2024249,766682.4
2025278,495763
2026310,529850.8
2027346,248948.6
2028386,0751,054.8
2029430,4831,179.4
2030480,0001,315.1

Requests per AI TWh, 2024 to 2030

Scales with AI TWh, not a load profile.

YearChatGPTGeminiEpochAgentReasoning
20242,941.2 billion4,166.7 billion3,333.3 billion877.2 billion20 billion
20252,941.2 billion4,166.7 billion3,333.3 billion877.2 billion20 billion
20262,941.2 billion4,166.7 billion3,333.3 billion877.2 billion20 billion
20272,941.2 billion4,166.7 billion3,333.3 billion877.2 billion20 billion
20282,941.2 billion4,166.7 billion3,333.3 billion877.2 billion20 billion
20292,941.2 billion4,166.7 billion3,333.3 billion877.2 billion20 billion
20302,941.2 billion4,166.7 billion3,333.3 billion877.2 billion20 billion

Remainder to 2030 in percent

100 % minus layer-six progress. Water from 2023.

YearAll, % missingAI, % missingWater, % missing
2023--53.3%
202456.3%77.8%48%
202548.9%66.7%42%
202641.6%58.5%35.3%
202733.2%48.3%27.9%
202823.6%35.6%19.6%
202912.6%19.7%10.3%
20300%0%0%

UTC calendar per year

Transparency for days, hours, and weeks behind every rate.

YearDaysHoursSecondsWeeks
20233658,76031,536,00052.1
20243668,78431,622,40052.3
20253658,76031,536,00052.1
20263658,76031,536,00052.1
20273658,76031,536,00052.1
20283668,78431,622,40052.3
20293658,76031,536,00052.1
20303658,76031,536,00052.1

100 MW-years and site equivalents

One 100 MW-year from layer four, split into world/AI/rest.

Year100 MW-year, TWhAll, 100 MW-yearsAI, 100 MW-yearsRest, 100 MW-yearsAll, 100 MW sites
20240.9 TWh472.4117.6354.8472.4
20250.9 TWh553.7176.9376.7553.7
20260.9 TWh633.3220.4412.9633.3
20270.9 TWh724.5274.6449.9724.5
20280.9 TWh826.5341.1485.4826.5
20290.9 TWh948426.1521.9948
20300.9 TWh1,084.5530.8553.71,084.5

Composition and micro-time

Growth shares, minutes, seconds, and AI water

Layer eight keeps decomposing only. No intraday profile, no German time series. Germany stays the Bitkom 2025 anchor as share or divisor.

Growth 2024 to 2030: AI plus rest

415 TWh start, plus AI and rest delta to 950 TWh. Pure TWh bridge.

Increase
Decrease
Total
IEA Key Questions on Energy and AI (2026). In-between years interpolated.
|
CC BY 4.0
gradually.ai

AI share and AI/rest ratio

Bar: AI share of total power in %. Line: AI divided by rest, scaled to 100.

AI share, %
AI/rest × 100
IEA Key Questions on Energy and AI (2026). In-between years interpolated.
|
CC BY 4.0
gradually.ai

Seconds per TWh in 2025

UTC year seconds divided by 2025 TWh (485 all-DC IEA, 155 AI OWID reading, rest derived). Uniform annual pace.

AI
203,458.1 s/TWh
Rest
95,563.6 s/TWh
All
65,022.7 s/TWh
IEA Key Questions on Energy and AI (2026). In-between years interpolated.
|
CC BY 4.0
gradually.ai

One request vs one AI second in 2025

Query Wh divided by global AI Wh per UTC second, in %.

Reasoning
0.00101729%
Agent
0.00002319%
ChatGPT
0.00000692%
Epoch
0.0000061%
Gemini
0.00000488%
OpenAI 2025, Epoch AI, Google Gemini Apps (August 22, 2025). Agent and reasoning: IEA, Key Questions on Energy and AI, Figure 2.1.
|
CC BY 4.0
gradually.ai

Growth shares, minutes, seconds, and AI water · The waterfall and dual-axis charts above show growth bridges and shares. Here are minute values, seconds per TWh, and growth contributions as tables.

Show tables (8)

Electricity as minute average, UTC year

TWh divided by UTC minutes. No intraday profile.

YearAll, kWh/minAI, kWh/minRest, kWh/min
2024787,416.5 kWh/min196,063.5 kWh/min591,353 kWh/min
2025922,754.9 kWh/min294,901.1 kWh/min627,853.9 kWh/min
20261,055,559.6 kWh/min367,367.4 kWh/min688,192.2 kWh/min
20271,207,477.7 kWh/min457,640.9 kWh/min749,836.8 kWh/min
20281,377,486.3 kWh/min568,539.8 kWh/min808,946.5 kWh/min
20291,580,053.8 kWh/min710,188 kWh/min869,865.8 kWh/min
20301,807,458.1 kWh/min884,703.2 kWh/min922,754.9 kWh/min

Water as minute average, 2023 to 2030

Billion liters divided by UTC minutes.

YearTotal, L/minDirect, L/min
20231,065,449 L/min266,362.3 L/min
20241,184,756.8 L/min296,189.2 L/min
20251,324,653.1 L/min331,163.3 L/min
20261,477,021.8 L/min369,255.5 L/min
20271,646,916.8 L/min411,729.2 L/min
20281,831,336.6 L/min457,834.2 L/min
20292,047,581.2 L/min511,895.3 L/min
20302,283,105 L/min570,776.3 L/min

Seconds per TWh, 2024 to 2030

How long one TWh takes at an even annual pace.

YearAllAIRest
202476,198.6 s/TWh306,023.2 s/TWh101,462.2 s/TWh
202565,022.7 s/TWh203,458.1 s/TWh95,563.6 s/TWh
202656,841.9 s/TWh163,324.2 s/TWh87,184.9 s/TWh
202749,690.4 s/TWh131,107.2 s/TWh80,017.4 s/TWh
202843,557.6 s/TWh105,533.5 s/TWh74,170.5 s/TWh
202937,973.4 s/TWh84,484.7 s/TWh68,976.2 s/TWh
203033,195.8 s/TWh67,819.4 s/TWh65,022.7 s/TWh

Who delivers the year-on-year growth?

Δ TWh vs prior year. AI share = AI Δ divided by total Δ.

YearAI, Δ TWhRest, Δ TWhAll, Δ TWhAI share of Δ
2024----
202551.7 TWh18.3 TWh70 TWh73.8%
202638.1 TWh31.7 TWh69.8 TWh54.6%
202747.4 TWh32.4 TWh79.8 TWh59.4%
202859.1 TWh32.2 TWh91.3 TWh64.7%
202973.6 TWh30.9 TWh104.5 TWh70.5%
203091.7 TWh27.8 TWh119.5 TWh76.7%

Total span 2024 to 2030

Sum of year deltas as a check on the waterfall bridge.

All, Δ TWh totalAI, Δ TWh totalRest, Δ TWh totalAI share of total ΔRest share of total Δ
535 TWh361.7 TWh173.3 TWh67.6%32.4%

Germany 2025 as a share of world

21.3 TWh divided by that year's world/AI/rest TWh. No DE growth.

YearDE / world, %DE / AI, %DE / rest, %
20245.1%20.6%6.8%
20254.4%13.7%6.5%
20263.8%11%5.9%
20273.4%8.9%5.4%
20282.9%7.1%5%
20292.6%5.7%4.7%
20302.2%4.6%4.4%

AI-attributed water per AI TWh

2023 water mix scaled by AI share. No 2023 AI water as fact.

YearDirect, bn L AI/yearOperational, bn L AI/yearDirect, L/kWh AIOperational, L/kWh AI
202438.9 billion L142.4 billion L0.4 L/kWh1.4 L/kWh
202555.6 billion L203.8 billion L0.4 L/kWh1.3 L/kWh
202667.5 billion L247.5 billion L0.3 L/kWh1.3 L/kWh
202782 billion L300.5 billion L0.3 L/kWh1.2 L/kWh
202899.6 billion L364.9 billion L0.3 L/kWh1.2 L/kWh
2029120.9 billion L443.1 billion L0.3 L/kWh1.2 L/kWh
2030146.8 billion L538.1 billion L0.3 L/kWh1.2 L/kWh

One request vs one AI second in 2025

Query Wh divided by global AI Wh per UTC second, in %.

ChatGPTGeminiEpochAgentReasoning
0.00000692%0.00000488%0.0000061%0.00002319%0.00101729%

Cadence and coupling

Quarters, households, withdrawal proxy, and query multiples

Layer nine keeps flat UTC rates. Withdrawals only as the 2023 ratio applied to consumption, no withdrawal time series.

Household equivalents, 2024 vs 2030

IEA: 100 MW AI DC = 100,000 households. Scales with interpolated MW.

20242030All DC47,244,990.9108,447,488.6+61,202,497.7AI DC11,763,813.053,082,191.8+41,318,378.8
IEA Key Questions on Energy and AI (2026). In-between years interpolated.
|
CC BY 4.0
gradually.ai

MW per TWh and liters per kWh

Left axis: average MW divided by TWh. Right axis: water in bn L divided by power TWh (i.e. L/kWh).

MW / TWh
L/kWh
IEA 2025 and 2026. Water and electricity come from two reports.
|
CC BY 4.0
gradually.ai

Query Wh relative to ChatGPT

Published Wh values divided by OpenAI 0.34 Wh. No load profile.

Reasoning
147.1×
Agent
3.4×
Epoch
0.9×
Gemini
0.7×
OpenAI 2025, Epoch AI, Google Gemini Apps (August 22, 2025). Agent and reasoning: IEA, Key Questions on Energy and AI, Figure 2.1.
|
CC BY 4.0
gradually.ai

Stack shares in 2030

AI and rest share of total power in percent, IEA 2030 anchor.

Rest
51.1%
AI
48.9%
IEA Key Questions on Energy and AI (2026). In-between years interpolated.
|
CC BY 4.0
gradually.ai

Quarters, households, withdrawal proxy, and query multiples · The household slope and dual-axis charts above show couplings. Here are quarterly, weekly, and rest-share values as tables.

Show tables (9)

Electricity as quarterly average, UTC year

Annual TWh divided by 4. No quarterly profile.

YearAll, TWh/quarterAI, TWh/quarterRest, TWh/quarter
2024103.8 TWh/quarter25.8 TWh/quarter77.9 TWh/quarter
2025121.3 TWh/quarter38.8 TWh/quarter82.5 TWh/quarter
2026138.7 TWh/quarter48.3 TWh/quarter90.4 TWh/quarter
2027158.7 TWh/quarter60.1 TWh/quarter98.5 TWh/quarter
2028181.5 TWh/quarter74.9 TWh/quarter106.6 TWh/quarter
2029207.6 TWh/quarter93.3 TWh/quarter114.3 TWh/quarter
2030237.5 TWh/quarter116.3 TWh/quarter121.3 TWh/quarter

Water as quarterly average, 2023 to 2030

Billion liters divided by 4.

YearTotal, bn L/quarter
2023140 bn L/quarter
2024156.1 bn L/quarter
2025174.1 bn L/quarter
2026194.1 bn L/quarter
2027216.4 bn L/quarter
2028241.3 bn L/quarter
2029269.1 bn L/quarter
2030300 bn L/quarter

Water as weekly average, UTC year

Billion liters divided by UTC weeks.

YearTotal, bn L/week
202310.7 bn L/week
202411.9 bn L/week
202513.4 bn L/week
202614.9 bn L/week
202716.6 bn L/week
202818.5 bn L/week
202920.6 bn L/week
203023 bn L/week

Household equivalents per calendar year

All and AI from MW path and IEA reference.

YearAll, householdsAI, households
202447,244,99111,763,813
202555,365,29717,694,064
202663,333,57522,042,043
202772,448,66427,458,455
202882,649,17734,112,388
202994,803,22842,611,282
2030108,447,48953,082,192

TWh per UTC second

Year TWh (anchor, derived, or CAGR) divided by UTC year seconds.

YearAll, TWh/sAI, TWh/sRest, TWh/s
20240.00001312 TWh/s0.000003268 TWh/s0.000009856 TWh/s
20250.00001538 TWh/s0.000004915 TWh/s0.00001046 TWh/s
20260.00001759 TWh/s0.000006123 TWh/s0.00001147 TWh/s
20270.00002012 TWh/s0.000007627 TWh/s0.0000125 TWh/s
20280.00002296 TWh/s0.000009476 TWh/s0.00001348 TWh/s
20290.00002633 TWh/s0.00001184 TWh/s0.0000145 TWh/s
20300.00003012 TWh/s0.00001475 TWh/s0.00001538 TWh/s

Rest share of total power

100 % minus AI share, pure TWh quote.

YearAI, %Rest, %
202424.9%75.1%
202532%68%
202634.8%65.2%
202737.9%62.1%
202841.3%58.7%
202944.9%55.1%
203048.9%51.1%

YoY growth AI vs rest

Ratio of year-on-year growth rates. Null when rest does not grow.

YearAI, YoYRest, YoYAI/rest YoY
2024---
202550%5.9%8.5×
202624.6%9.6%2.6×
202724.6%9%2.7×
202824.6%8.2%3×
202924.6%7.2%3.4×
203024.6%6.1%4×

Water Δ to power Δ

Billion liters Δ divided by TWh Δ vs prior year. From 2025 only.

YearWater, Δ bn LAll, Δ TWhΔwater/Δpower
2024---
202571.8 billion L70 TWh1
202680.1 billion L69.8 TWh1.1
202789.3 billion L79.8 TWh1.1
202899.6 billion L91.3 TWh1.1
2029111 billion L104.5 TWh1.1
2030123.8 billion L119.5 TWh1

Withdrawal proxy from 2023 ratio

5,500/560 only as scalar on consumption CAGR, not measured per year.

YearConsumption, bn LWithdrawal proxy, bn L2023 ratio
2023560 billion L5,500 billion L9.8×
2024624.4 billion L6,132.6 billion L9.8×
2025696.2 billion L6,838 billion L9.8×
2026776.3 billion L7,624.6 billion L9.8×
2027865.6 billion L8,501.6 billion L9.8×
2028965.2 billion L9,479.5 billion L9.8×
20291,076.2 billion L10,569.9 billion L9.8×
20301,200 billion L11,785.7 billion L9.8×

Anchor ledger and references

Anchor and derived constants, half-years, and site benchmarks

Layer ten keeps the anchor ledger open. Heatmaps and chord diagrams stay flat. Withdrawals only indirect via site liters/day.

MW stack, 2024 vs 2030

AI MW and rest MW from UTC annual pace. Rest = all minus AI.

20242030AI MW11,763.8 MW53,082.2 MW+41,318.4 MWRest MW35,481.2 MW55,365.3 MW+19,884.1 MW
IEA Key Questions on Energy and AI (2026). In-between years interpolated.
|
CC BY 4.0
gradually.ai

2025 TWh comparison (symmetric)

AI, rest, and Bitkom 21.3 TWh as nodes. Ribbon width = minimum of the two TWh.

AIRestDE 2025
AI
Rest
DE 2025
Bitkom/Borderstep 2025 and IEA 2026.
|
CC BY 4.0
gradually.ai

2023 water mix in billion liters

140/373/47 from IEA Energy and AI. No CAGR on mix shares.

Power
373 billion L
Direct
140 billion L
Hardware
47 billion L
IEA Energy and AI (2025). Consumption (not withdrawals).
|
CC BY 4.0
gradually.ai

950 TWh in 2030, spread evenly per month

950 TWh IEA 2030 anchor, divided by 12. No seasonality.

JanFebMarAprMayJunJulAugSepOctNovDec203079.1666666666666779.1666666666666779.1666666666666779.1666666666666779.1666666666666779.1666666666666779.1666666666666779.1666666666666779.1666666666666779.1666666666666779.1666666666666779.16666666666667
TWh/month: lowhigh(max 79)
IEA Key Questions on Energy and AI (2026). In-between years interpolated.
|
CC BY 4.0
gradually.ai

465 TWh AI in 2030, spread evenly per month

465 TWh IEA 2030 anchor, divided by 12.

JanFebMarAprMayJunJulAugSepOctNovDec203038.7538.7538.7538.7538.7538.7538.7538.7538.7538.7538.7538.75
TWh/month: lowhigh(max 39)
IEA 2025 and 2026. Water and electricity come from two reports.
|
CC BY 4.0
gradually.ai

Anchor and derived constants, half-years, and site benchmarks · The heatmaps and chord chart above show anchors and the Germany comparison. Here are constants, half-years, and the MW stack as tables.

Show tables (10)

Anchor values and derived constants

IEA, OWID, Bitkom, and query cards. AI 2025 is the OWID reading of IEA Fig. 1.5. AI 2024 is back-calculated from +50%, not a typed IEA numeral.

MetricValue
All 2024, TWh415 TWh
All 2025, TWh485 TWh
All 2030, TWh950 TWh
AI 2024, TWh103.3 TWh
AI 2025, TWh155 TWh
AI 2030, TWh465 TWh
Water 2023, bn L560 billion L
Water 2030, bn L1,200 billion L
OWID 2030, TWh945 TWh
IEA 2030, TWh950 TWh
DE 2025, TWh21.3 TWh
ChatGPT, Wh0.3
Gemini, ml0.3

Electricity as weekly average, UTC year

Complements layer six (power) and nine (water/week).

YearAll, TWh/weekAI, TWh/weekRest, TWh/week
20247.9 TWh/week2 TWh/week6 TWh/week
20259.3 TWh/week3 TWh/week6.3 TWh/week
202610.6 TWh/week3.7 TWh/week6.9 TWh/week
202712.2 TWh/week4.6 TWh/week7.6 TWh/week
202813.9 TWh/week5.7 TWh/week8.2 TWh/week
202915.9 TWh/week7.2 TWh/week8.8 TWh/week
203018.2 TWh/week8.9 TWh/week9.3 TWh/week

Electricity as half-year average

Annual TWh divided by 2.

YearAll, TWh/half-yearAI, TWh/half-yearRest, TWh/half-year
2024207.5 TWh/half-year51.7 TWh/half-year155.8 TWh/half-year
2025242.5 TWh/half-year77.5 TWh/half-year165 TWh/half-year
2026277.4 TWh/half-year96.5 TWh/half-year180.9 TWh/half-year
2027317.3 TWh/half-year120.3 TWh/half-year197.1 TWh/half-year
2028363 TWh/half-year149.8 TWh/half-year213.2 TWh/half-year
2029415.2 TWh/half-year186.6 TWh/half-year228.6 TWh/half-year
2030475 TWh/half-year232.5 TWh/half-year242.5 TWh/half-year

Water as half-year average

Billion liters divided by 2.

YearTotal, bn L/half-year
2023280 bn L/half-year
2024312.2 bn L/half-year
2025348.1 bn L/half-year
2026388.2 bn L/half-year
2027432.8 bn L/half-year
2028482.6 bn L/half-year
2029538.1 bn L/half-year
2030600 bn L/half-year

Absolute water mix per year

2023 shares on interpolated consumption.

YearDirect, bn LPower, bn LHardware, bn LTotal, bn L
2023140 billion L373 billion L47 billion L560 billion L
2024156.1 billion L415.9 billion L52.4 billion L624.4 billion L
2025174.1 billion L463.7 billion L58.4 billion L696.2 billion L
2026194.1 billion L517.1 billion L65.2 billion L776.3 billion L
2027216.4 billion L576.6 billion L72.7 billion L865.6 billion L
2028241.3 billion L642.9 billion L81 billion L965.2 billion L
2029269.1 billion L716.8 billion L90.3 billion L1,076.2 billion L
2030300 billion L799.3 billion L100.7 billion L1,200 billion L

MW stack per calendar year

Average MW from TWh and UTC seconds.

YearAll, MWAI, MWRest, MW
202447,245 MW11,764 MW35,481 MW
202555,365 MW17,694 MW37,671 MW
202663,334 MW22,042 MW41,292 MW
202772,449 MW27,458 MW44,990 MW
202882,649 MW34,112 MW48,537 MW
202994,803 MW42,611 MW52,192 MW
2030108,447 MW53,082 MW55,365 MW

Leap years and UTC seconds

2024 has 366 days. TWh/s shifts slightly at the same CAGR.

YearDaysSecondsΔ vs 365 daysAll, TWh/s
202436631,622,40086,4000.00001312 TWh/s
202536531,536,00000.00001538 TWh/s
202636531,536,00000.00001759 TWh/s

World vs IEA reference site, liters/day

2 million L/day illustration vs world daily mean (UTC days, interpolated between anchors).

YearWorld, L/daySite, L/daySites at 2M L
20241,706,049,733.3 L/day2,000,000 L/day853
20251,907,500,470.3 L/day2,000,000 L/day953.8
20262,126,911,401.2 L/day2,000,000 L/day1,063.5
20272,371,560,153.8 L/day2,000,000 L/day1,185.8
20282,637,124,718.3 L/day2,000,000 L/day1,318.6
20292,948,516,999.5 L/day2,000,000 L/day1,474.3
20303,287,671,232.9 L/day2,000,000 L/day1,643.8

Query water per AI TWh

ml/query from cards × requests/AI TWh from layer seven.

YearChatGPT, ml/AI TWhGemini, ml/AI TWh
2024946.4 billion1,083.3 billion
2025946.4 billion1,083.3 billion
2026946.4 billion1,083.3 billion
2027946.4 billion1,083.3 billion
2028946.4 billion1,083.3 billion
2029946.4 billion1,083.3 billion
2030946.4 billion1,083.3 billion

OWID 945 vs IEA 950 (2030)

Caption gap from layer three, as a ledger row.

MetricValue
OWID, TWh945 TWh
IEA, TWh950 TWh
Δ TWh5 TWh

Your use

What a single query costs

The world counters above are annual totals. The figures here are marginal costs per text query, as providers and researchers published them in 2025. Training is not included. The IEA agent figures are GPU electricity only, not a full facility load.

ChatGPT, OpenAI (0.34 Wh)
6.8 Wh
Gemini Apps, Google (0.24 Wh)
4.8 Wh
ChatGPT water (0.000085 gallons)
6.4 ml
Gemini Apps water (0.26 ml)
5.2 ml
IEA agent without reasoning (1.14 Wh)
22.8 Wh
IEA agent with reasoning (50 Wh)
1,000 Wh

today

IEA anchors

The annual figures behind the counter

Compact overview of the IEA year anchors for power and water. The wide detail table below is optional and shows derived scaling from the same numbers.

YearAll data centersAI-focusedTotal water consumption
2024415 TWhpublished103.3 TWhderived from +50% in 2025624.4 billion Linterpolated
2025485 TWhpublished155 TWhOWID reading of IEA figure696.2 billion Linterpolated
2026554.8 TWhinterpolated193.1 TWhinterpolated776.3 billion Linterpolated
2030950 TWhpublished465 TWhpublished1,200 billion Lpublished
Detail table: MW, sites, households

Derived figures from the same anchors

Power, 100 MW equivalents, and households follow the IEA rule of thumb. Pool counts use the annual water total, not the live ticker.

YearRemainder without an AI focusAI shareAverage power, allAverage power, AI100 MW equivalents, all100 MW equivalents, AIHousehold equivalents, allOlympic pools, annual total
2024311.7 TWh24.9%47,245 MW11,764 MW472.4117.647,244,991249,766
2025330 TWh32%55,365 MW17,694 MW553.7176.955,365,297278,495
2026361.7 TWh34.8%63,334 MW22,042 MW633.3220.463,333,575310,529
2030485 TWh48.9%108,447 MW53,082 MW1,084.5530.8108,447,489480,000

Methodology

How the counter calculates

  1. 01

    IEA annual anchors

    Electricity comes from “Key Questions on Energy and AI” (April 16, 2026): 415 TWh in 2024, 485 TWh in 2025, and 950 TWh in 2030 for all data centers. Our World in Data reads IEA Figure 1.5 as 155 TWh for AI-focused sites in 2025. 155 plus 330 TWh of non-AI load equals the published 485 TWh. Those AI sites reach about 465 TWh by 2030. Our World in Data’s chart caption still prints 945 TWh for 2030, the total from the April 2025 IEA report. The 2026 update uses 950 TWh. Water comes from “Energy and AI” (April 10, 2025): 560 billion liters consumed in 2023 and 1,200 billion liters in 2030.

  2. 02

    Growth between anchors

    Between two IEA years the annual total grows exponentially at the CAGR. That is why 2026 AI electricity sits near 193 TWh rather than a linear midpoint. After 2030 the counter holds the last IEA value. It does not invent its own long-range forecast.

  3. 03

    Even spread across the calendar year

    The IEA publishes annual totals, not hourly loads. Inside a UTC calendar year the counter spreads that total evenly across every second, including leap years. There is no seasonality, no outage, and no peak-shaving.

  4. 04

    Water follows the electricity share

    The IEA does not split water into AI and the rest. The counter applies the current AI share of data-center electricity to the global water total. The 2023 mix stays fixed: 25% direct cooling, about 67% power plants, about 8% hardware. Hardware stays out of the live ticker because it is not an on-site operating flow. The IEA also puts 2023 withdrawals at about 5,500 billion liters. The ticker uses consumption only.

Formula

annualTotal(year) × (seconds since January 1 UTC / seconds in the year)

What this is not

Not a live meter, not a satellite wattmeter, not a real-time operator feed. The IEA figures are modeled too. Other institutes such as S&P or the Energy Institute sit higher, in part because they include crypto mining. Newer AI halls often evaporate less water. The direct water line may therefore overstate the future.

Sources

Primary sources, not press copy

Full write-up with tables and forecasts

Frequently asked questions