Skip to content
WattCurious

How Pumped Hydro Storage Works

Updated 8 October 20267 min readBatteries & Storage

Pumped hydro stores electricity as height. Surplus power drives water uphill into an upper reservoir; when the power is wanted back, the same water falls through the same machine and turns it the other way. There is no chemistry, nothing degrades with cycling, and the stored energy is a volume multiplied by a drop — which is both the reason it scales to enormous capacity and the reason it needs a very particular piece of landscape.

Key takeaways

  • Stored energy is density times gravity times volume times head, so height and volume trade against each other exactly.
  • Gravitational energy is thin: a swimming pool raised a hundred metres holds less than you would guess, which is why real schemes are enormous.
  • One reversible pump-turbine does both jobs, which halves the machinery and complicates the hydraulic design.
  • Round-trip losses are mechanical and hydraulic rather than chemical, and they do not grow as the machine is cycled.
  • A rotating mass already spinning can change its output in seconds, which is worth as much to a grid as the energy itself.
On this page
  1. Storage as a hill
  2. One machine, two directions
  3. Where the round trip goes
  4. Why the response time matters as much as the energy
  5. What a site has to have

Storage as a hill

The whole principle is one equation. Lift a mass against gravity and you have stored energy equal to the weight multiplied by the height; let it fall through a machine and you get some of it back.

E = ρ · g · V · h

Density of water, gravitational acceleration, the volume moved, and the height it was moved through. No chemistry, no state of charge to estimate, no ageing term.

A section through a pumped hydro scheme. An upper reservoir sits on high ground, connected by a penstock to a powerhouse and a lower reservoir far below. Two states are drawn: charging, in which surplus electricity drives the machine as a pump and water moves uphill; and discharging, in which the same water falls back through the same machine, now acting as a turbine, and the motor-generator produces electricity. The vertical separation between the two water surfaces is marked as head.A section through a pumped hydro scheme. An upper reservoir sits on high ground, connected by a penstock to a powerhouse and a lower reservoir far below. Two states are drawn: charging, in which surplus electricity drives the machine as a pump and water moves uphill; and discharging, in which the same water falls back through the same machine, now acting as a turbine, and the motor-generator produces electricity. The vertical separation between the two water surfaces is marked as head.
One machine, two directions. Surplus power raises the water; demand lets it fall back through the same runner.

The honest thing to say next is that gravitational storage is thin. A cubic metre of water raised a hundred metres stores roughly a quarter of a kilowatt-hour before any losses — about what a kettle uses. Everything impressive about pumped hydro comes from multiplying that small number by an enormous volume and a large drop.

A worked calculation. A volume of one million cubic metres and a head of three hundred metres are multiplied by water density and gravitational acceleration to give about 2.9 times ten to the twelve joules. Converted to kilowatt-hours this is about 818 megawatt-hours of stored energy. Applying a round-trip efficiency of about 0.8 gives roughly 650 megawatt-hours returned. A note marks the figures as an illustrative example rather than a real scheme.A worked calculation. A volume of one million cubic metres and a head of three hundred metres are multiplied by water density and gravitational acceleration to give about 2.9 times ten to the twelve joules. Converted to kilowatt-hours this is about 818 megawatt-hours of stored energy. Applying a round-trip efficiency of about 0.8 gives roughly 650 megawatt-hours returned. A note marks the figures as an illustrative example rather than a real scheme.
Illustrative arithmetic. A million cubic metres and three hundred metres of head — a genuinely large piece of landscape — for a few hundred megawatt-hours.

That is the trade in a sentence: the energy density is poor and the scalability is superb. Where a chemical store's capacity is set by how much material you buy, a hydro store's is set by how much valley you have.

One machine, two directions

The mechanical heart is a reversible pump-turbine coupled to a motor-generator. Run electricity into the motor and the runner acts as a pump; let water fall through it and the same runner acts as a turbine driving the generator.

A pump-turbine unit shown in both modes. In pumping mode the motor-generator drives the shaft, the runner turns in one direction and forces water up the penstock against the head. In generating mode water falls through the penstock, passes the wicket gates and drives the runner the other way, turning the motor-generator as a generator. Labels mark the wicket gates that control flow, the draft tube below, and the common shaft that both modes share.A pump-turbine unit shown in both modes. In pumping mode the motor-generator drives the shaft, the runner turns in one direction and forces water up the penstock against the head. In generating mode water falls through the penstock, passes the wicket gates and drives the runner the other way, turning the motor-generator as a generator. Labels mark the wicket gates that control flow, the draft tube below, and the common shaft that both modes share.
The same runner, the same shaft, the same machine hall. Halving the machinery is the reason it is done this way — and the reason the hydraulic design is a compromise.

The hydraulics are the same family as any other water machine — the head and flow relationship that governs micro-hydro governs this too, just at a scale where the reservoir is the variable rather than the stream. Because all of the water is confined in the penstock and forced through the runner, the machine is not bound by the open-flow limit that caps a wind rotor; nothing can spill around it.

Using one runner for both jobs halves the machinery and the excavation, and it costs something real: a runner optimised for pumping is not optimal as a turbine, and vice versa, so the design is a negotiated compromise. It also has to be a reaction machine, running full of water under pressure: an impulse turbine that catches a free jet in open air has no way to run backwards as a pump. Wicket gates around the runner control flow and are how output is modulated; below it the draft tube recovers some of the remaining velocity as useful head.

Some schemes go further and hold a unit spinning in air — runner de-watered, machine synchronised to the grid, consuming very little — so it can take up load almost immediately. That is a deliberate trade of a small standing loss for a very short response time.

Where the round trip goes

Nothing is free in either direction, and unlike a battery the losses are all mechanical and hydraulic.

A loss chain running from grid electricity through pumping and back to grid electricity. On the pumping side: transformer and converter losses, motor losses, pump hydraulic losses and friction in the penstock. Energy is then stored with only evaporation and seepage acting on it. On the generating side: friction in the penstock again, turbine hydraulic losses, generator losses, and transformer losses. The bar arriving back at the grid is noticeably shorter than the one that left it, and a note says the stages multiply.A loss chain running from grid electricity through pumping and back to grid electricity. On the pumping side: transformer and converter losses, motor losses, pump hydraulic losses and friction in the penstock. Energy is then stored with only evaporation and seepage acting on it. On the generating side: friction in the penstock again, turbine hydraulic losses, generator losses, and transformer losses. The bar arriving back at the grid is noticeably shorter than the one that left it, and a note says the stages multiply.
Every stage is paid twice, once uphill and once down. What survives is the round-trip efficiency, and none of it is chemical.

Each stage is paid on the way up and again on the way down: conversion, machine, and friction in the waterway. Round-trip efficiency for pumped hydro is typically around 70 to 80 per cent, with some modern plants higher — the range published by the US Department of Energy. The exact figure still belongs to a particular scheme rather than to the technology.

The important property is not the number but its stability. Battery round-trip efficiency and capacity both drift as the cells age, which is what makes chemistry and cycle life such a central question for electrochemical storage. A pump-turbine's efficiency is set by its hydraulics, and hydraulics do not care how many times they have run.

Why the response time matters as much as the energy

A grid does not only need energy moved from one hour to another. It needs something that can change its output quickly when demand or generation moves unexpectedly.

A timeline of how quickly different resources can change output. A unit already spinning in air responds within seconds. A unit starting from standstill takes a few minutes. Thermal plant takes considerably longer, from many minutes to hours depending on its state. A note explains that a pumped hydro scheme can also switch from pumping to generating, so its usable swing is larger than its generating capacity alone.A timeline of how quickly different resources can change output. A unit already spinning in air responds within seconds. A unit starting from standstill takes a few minutes. Thermal plant takes considerably longer, from many minutes to hours depending on its state. A note explains that a pumped hydro scheme can also switch from pumping to generating, so its usable swing is larger than its generating capacity alone.
A mass already turning can be useful almost at once. And a scheme that is pumping can stop pumping and start generating, which doubles the swing available.

Two things make pumped hydro unusually valuable here. A synchronised machine can pick up load in seconds. And a scheme that is currently pumping can stop and start generating, so the swing available to the system is the sum of both — considerably more than its generating rating alone.

This is the same argument that makes surplus a control problem rather than an accounting one, as what happens to excess solar energy sets out: what a system needs is something that can absorb and release on demand, quickly, not merely something large.

What a site has to have

The physics is simple and the siting is not. The requirements are severe, simultaneous and geographic.

What a pumped hydro site requires
RequirementWhyWhat happens without it
Two reservoirs, close togetherHead is the energy; distance is frictionLong waterways, high losses, high construction
Large vertical separationEnergy scales directly with itEnormous volumes needed for modest energy
Geology that holds waterSeepage is a permanent lossLining works, or the scheme leaks
Somewhere to put the spoilReservoirs mean excavationAn unsolved problem before work begins
A water source to top upEvaporation and seepage continueThe store slowly empties
A grid connection worth havingStorage is useless where nothing needs itStranded capacity

Physical requirements rather than a ranked list. All of them have to be true at once, which is why suitable sites are scarce and why none of them can be engineered into existence.

The fifth row is where closed-loop schemes differ. A closed-loop pair is not connected to a river: the same water circulates and only evaporation and seepage need making up. That removes the dependence on a watercourse, and with it a large part of the environmental objection, which is why closed-loop siting has become the more interesting question.

And this is finally where pumped hydro sits among the alternatives. It is the largest and longest-lived store available, it does not degrade with use, and it exists only where the landscape allows. Storing surplus as hydrogen accepts a much worse round trip in exchange for working anywhere — which is the trade that keeps both on the table.

Frequently asked questions

How much energy does pumped hydro actually store?

Per unit of water, very little — gravitational storage is thin. A cubic metre raised a hundred metres holds roughly a quarter of a kilowatt-hour before losses. That is why the reservoirs are measured in millions of cubic metres and the heads in hundreds of metres: the scale is doing the work, not the density.

Does it wear out from cycling?

Not in the way a battery does. There is no chemistry to degrade, so cycling costs wear on bearings, seals and runners rather than capacity. A scheme can be cycled daily for decades and still store what it did when it was built, which is the property that most distinguishes it from electrochemical storage.

Why is it so fast to respond?

Because the mass is already moving. A unit spinning in air, with the runner de-watered and the machine synchronised, can take up load in seconds — far faster than anything that has to start from rest. Some schemes hold units in that state deliberately, ready to be useful within moments.

Does the water get used up?

No, it circulates between the two reservoirs. Losses are evaporation and seepage rather than consumption, and a closed-loop scheme with no river connection only needs topping up for those. That is one of the reasons closed-loop sites are of increasing interest: they do not depend on a watercourse.

Why isn't there more of it?

Because the site requirements are severe and non-negotiable: two reservoirs with a large vertical separation, close together, on geology that will hold water, with somewhere to put the excavated material and a grid connection worth having. Those conditions exist in a limited number of places, and no amount of engineering creates them.

Sources

Named organisations whose published material underpins this article. Where no link is given, the source is named rather than linked.

Editorial Team

Research, drafting and review

Articles are drafted from primary engineering and physics references with AI-assisted tools, then reviewed and fact-checked line by line by a human editor before publication. We publish explanations, not recommendations: no products, no pricing, no country-specific rules, and no invented author personas.

Last reviewed 8 October 2026. How we research and review