How much water does a power plant use? Nuclear, gas, coal, wind and solar, by cooling system
A power plant's water use is set by its cooling system first and by its fuel second. A nuclear plant on a cooling tower evaporates about 600 gallons for every megawatt-hour it generates (a megawatt-hour, MWh, is what a US home buys in about five weeks). A gas combined-cycle plant (a gas turbine whose exhaust heat raises steam for a second turbine) on the same kind of tower evaporates about 200, and either plant moved to dry cooling (cooling with outside air) evaporates none. Wind turbines and solar panels use almost none. The three sources everyone cites for these figures disagree with one another on one row, and the disagreement comes down to how each one defines consumption.
Why does the cooling system decide a power plant's water use before the fuel does?
Every steam plant (a plant that boils water into steam to drive a turbine) turns part of its fuel heat into electricity and has to get rid of all the rest. The cooling system decides whether that leftover heat leaves as evaporated water, as a warmer river, or as warmer air. The fuel decides how much leftover heat there is per megawatt-hour. That is why a nuclear plant uses more water than a gas plant on every cooling system in the table.
The 2024 operating heat rates published by EIA (the US Energy Information Administration, the federal energy statistics agency) record the fuel heat each plant burned for every kilowatt-hour it generated, in Btu (British thermal units, a unit of heat). Nuclear plants ran at 10,443 Btu for each kilowatt-hour they generated in 2024, and natural gas plants at 7,754 on average across all gas plant types. A kilowatt-hour of electricity contains 3,412 Btu, and the difference is heat that has to go somewhere.
A nuclear plant is therefore left with 7,031 Btu of heat to dispose of per kilowatt-hour, and nearly all of it passes through the condenser (the heat exchanger that turns the turbine's exhaust steam back into water). A gas combined-cycle plant is left with 4,342 Btu at that average, and part of that leaves through the stack and never reaches the condenser.
The cooling system then carries that condenser heat away in one of four ways, and the four differ from one another by orders of magnitude, meaning factors of ten, on withdrawal (the water taken in from the source, whether or not it goes back):
A recirculating tower (the tall concrete towers or the low banks of fans) evaporates water to carry the heat off. It withdraws little, under 2.5 gallons for every 1 it evaporates at nuclear, coal and combined-cycle plants, and evaporates the most.
A once-through system pumps water from a river, lake or bay through the condenser once and returns it a few degrees warmer. It withdraws 30 to 90 times what a tower does and evaporates almost nothing at the plant itself.
A cooling pond holds the warm water in a lake built for the purpose, where it cools by evaporation before being reused.
Dry cooling blows outside air across the condenser tubes (an air-cooled condenser) and uses no cooling water, at the cost of some output on hot afternoons.
Two nuclear plants can differ by a factor of 30 on withdrawal, one on a tower and one on once-through cooling. A nuclear plant and a gas plant on the same tower differ by a factor of 3 on consumption (the water evaporated and never returned to the source).
The 2024 plant reports show the same split across the whole reporting fleet: cooling towers generated 57 percent of the electricity from reporting thermoelectric plants (steam plants) and took 2.6 percent of the water withdrawn.
Once-through plants, which pump water through the condenser a single time, took 64 percent of the water the same fleet withdrew, and cooling ponds took another 26 percent.
How much water does each type of power plant use?
A nuclear plant on a cooling tower consumes 601 gallons of water per megawatt-hour in the 2024 plant reports. A coal plant on a tower consumes 580 and a gas combined-cycle plant on a tower 206. Once-through plants of any fuel withdraw tens of thousands of gallons per megawatt-hour and return nearly all of it warmer. Dry-cooled plants, wind and solar photovoltaics (panels) use little or none. Three sources supply the figures in common use. NREL's 2011 literature review (the National Renewable Energy Laboratory, a federal laboratory) is the first. USGS's 2015 heat-budget model (the US Geological Survey's estimate of water use from the heat each plant has to shed) is the second, and the plant reports EIA collects each year are the third.
Plant and cooling system | NREL literature medians (2011) | USGS heat-budget model (2015) | EIA plant reports (2024, our arithmetic) |
Nuclear, tower | 672 / 1,101 | 500 / 700 | 601 / 1,375 |
Nuclear, once-through | 269 / 44,350 | 400 / 39,000 | 103 / 41,699 |
Nuclear, cooling pond | 610 / 7,050 | 480 / 480 | 488 / 38,535 |
Coal, tower | 687 / 1,005 | 400 / 570 | 580 / 1,056 |
Coal, once-through | 250 / 36,350 | 330 / 32,000 | 114 / 54,990 |
Coal, cooling pond | 545 / 12,225 | 490 / 490 | 613 / 47,714 |
Gas combined cycle, tower | 198 / 253 | 180 / 250 | 206 / 286 |
Gas combined cycle, once-through | 100 / 11,380 | 150 / 14,000 | not reported / 24,658 |
Gas combined cycle, dry | 2 / 2 | not modeled | 0 / 0 |
Gas steam turbine, tower | 826 / 1,203 | 490 / 690 | 757 / 2,374 |
Gas combustion turbine (no steam cycle) | no cooling water | not modeled | not required to report |
Solar photovoltaic | 26 / 26 | not modeled | not required to report |
Wind | 0 / 0 | not modeled | not required to report |
Concentrating solar, tower cooling | 865 / 865 | not modeled | no per-MWh figure |
Each cell reads consumption first and withdrawal second, in gallons per megawatt-hour, from the source named in the column. NREL figures are the medians in Macknick and colleagues' review of published estimates, using the generic coal rows.
The review sets withdrawal equal to consumption for solar and wind. USGS figures are the 2015 coefficients in Harris and Diehl's heat-budget models, converted from gallons per kilowatt-hour. EIA figures are generation-weighted averages (each plant counted in proportion to its output) we computed from the 2024 cooling water file. The two concentrating solar plants (plants that run a steam turbine on mirror-focused heat) on towers in the file report water but no generation.
The nuclear tower figure differs little across the three sources, at 672, 500 and 601 gallons per megawatt-hour. The gas combined-cycle tower figure falls within a few dozen gallons of the same value in all three. Where the sources measure the same thing, they agree within the spread of the plants themselves.
Figure 1. Annual water withdrawal and consumption for a plant of any size, fuel and cooling system, computed from the per-megawatt-hour factors above. A 1,000 MW nuclear plant on towers evaporates 4.8 billion gallons a year; on once-through cooling it withdraws 332 billion. Sources, formula and vintages: methods note. Analysis: Gasilov Group, our arithmetic and cross-reference, as of September 2026.
Why do the definitive sources disagree on once-through plants?
The three sources disagree on once-through consumption because they define consumption differently, and the difference is a choice of accounting. EIA counts what a plant evaporates on its own site. USGS counts the evaporation the plant's warm discharge forces from the river downstream, and NREL's review mixed studies that did each. The current Form EIA-923 instructions (the annual survey power plants file with EIA) define water consumption as "water consumed through evaporative losses in cooling towers or cooling ponds." The same instructions tell once-through plants that "evaporative losses are not expected with once-through systems as withdrawal should equal discharge." In the 2024 file, 122 of the 149 plants on once-through cooling reported zero or left the consumption field blank.
EIA's definition of consumption is what puts the once-through figures at 103 gallons per megawatt-hour for nuclear and 114 for coal. Both figures are the evaporation reported by the few plants that filled the field in, spread over the generation of all of them.
Five of the 25 nuclear plants on once-through cooling reported any consumption in 2024, and 11 of the 43 coal plants did. The national average divides that reported evaporation over every once-through plant's generation, the zero reporters included.
USGS counts something the plant never measures, the water its heat takes out of the river after the discharge. Its models estimate "the forced evaporation from the surface-water body that receives the heated discharge water," in the report's words. On that accounting, a once-through nuclear plant consumes 400 gallons per megawatt-hour and a coal plant 330. Those figures are four fifths of what the same plants would evaporate on a tower.
Macknick's literature medians, 269 for nuclear and 250 for coal, fall between the two because the studies he compiled used both definitions. The review, by its own account, "did not alter (except for unit conversion) or audit for accuracy" any estimate it used.
I would take the USGS figure for any question about a basin and the EIA figure for any question about a permit. The river loses the water either way, in the plant's tower or two miles downstream, and a state water plan or a drought curtailment order (an order cutting water use during a shortage) counts the loss wherever it happens. A discharge permit under the Clean Water Act (the federal water pollution law) counts what passes through the intake and the outfall. A table that prints one of them without saying which has hidden a factor of 3 or 4. In OGEL Energy Law Journal in April 2026, I argued that water scarcity in the Colorado River Basin puts a physical limit on steam plants that utility resource plans in the West still leave out.
Once-through coal plants in the 2024 file withdrew 54,990 gallons per megawatt-hour on average, a figure well above the older sources. The USGS 2015 coefficients put once-through coal at 32,000 and Macknick's medians at 36,350. The difference comes from how much of the year the plants ran.
The coal fleet ran at a 42.6 percent capacity factor (the share of the year a plant runs at full output) in 2024, down from 52.8 percent in 2016. A once-through plant's circulating pumps run whenever the unit is online, at full load or not. We sorted the 43 once-through coal plants in the file by capacity factor to test that explanation.
The plants running under 30 percent of the year withdrew a median 114,000 gallons per megawatt-hour of generation. Plants running more than half the year withdrew a median 38,000, close to the older figures.
A coal plant kept online as a peaking unit (one run on the highest-demand hours) withdraws the water of a baseload unit (one run around the clock) and delivers a fraction of the electricity. Any per-megawatt-hour factor from 2011 or 2015 understates such a plant's withdrawal, and a plant-level factor should be recomputed from the plant's own capacity factor.
A team writing a water clause into a siting study, a basin study or a power contract uses a different figure for each. A siting study uses the withdrawal figure to test the intake against the source. EPA's 2014 cooling water intake rule (EPA is the US Environmental Protection Agency) applies to any existing facility designed to withdraw more than 2 million gallons a day with a quarter or more of it for cooling. The rule sets entrainment requirements (limits on fish eggs and larvae drawn into the intake) site by site. A basin study uses the USGS consumption figure for once-through plants, including the downstream evaporation. A power purchase agreement (a contract to buy electricity from a named plant) that promises low water use names the cooling system, because two plants burning the same fuel can differ by a factor of 30 on water.
What does this mean for a large electricity buyer?
For a data center or any other large load (a single customer whose demand is measured in megawatts), the water evaporated at the power plant supplying it is set by the same two choices, cooling system and fuel, and both are made in the power contract. A buyer signing that contract fixes the plant's water use, whether or not the contract names it. I told Developing Telecoms in August 2026 that every data center should have to report both water withdrawn and water consumed, and the same two figures apply to the plant behind it.
A buyer contracting for a gas combined-cycle plant on towers is contracting for 206 gallons per megawatt-hour of evaporation at that plant. The same plant on dry cooling evaporates none, and the cost comes out of output. The DOE and Argonne National Laboratory analysis (the US Department of Energy and one of its laboratories) of coal plants converted to indirect dry cooling (a dry system in which air cools the circulating water loop) put the penalty at 4.2 to 5.2 percent of annual output.
A buyer contracting for nuclear on towers is contracting for about 600, three times the gas figure, and a buyer contracting for wind or solar for close to nothing. Which of those a corporate water target should count is an open question. The target is usually written for the buyer's own sites, and the power plant's water is usually left out. For every closed-loop facility (a data center that recirculates its cooling water in place of evaporating it) we have analyzed, the plant's water has been larger than the site's.
If you are not sure where to start, the two-week water exposure screen is the first step. If you already know what you need, get in touch and we can scope it.
Arif Gasilov is partner for Natural Resources & Built Environment at Gasilov Group, where he works on energy regulatory analysis, water governance, utility rate cases, and the resource footprint of large development. Meet Our Partners →
Frequently Asked Questions (FAQ)
Which power plants have to report their cooling water use?
Thermoelectric plants with at least 100 megawatts of steam-electric (steam turbine) capacity, including nuclear and combined-cycle plants, file the cooling schedule of Form EIA-923. Combustion turbines, wind farms and photovoltaic plants do not. Plants below the threshold and plants outside the United States appear in no comparable public dataset, and their figures have to be estimated from the factors in the table.
Do wind and solar farms use any water at all?
Utility-scale solar photovoltaic plants use water to wash panels, a median of 26 gallons per megawatt-hour in the NREL review, and wind turbines use almost none. Concentrating solar plants, which run a steam turbine on mirror-focused heat, are the exception. On wet cooling towers they evaporate 865 gallons per megawatt-hour, more than a nuclear plant, and a small fraction of that on dry cooling.
