Storage and Firming

In 2026 the world's installed battery capacity passed 250 gigawatts and overtook pumped hydro for the first time. Pumped storage had crossed 200 gigawatts only the year before, after a century of construction.

Pumped hydro is exactly what the name says. Two reservoirs sit at different heights, cheap power pumps water from the lower one up to the upper one, and when the system wants it back the water runs down through machines usually built to work in both directions, pumping one way and generating the other. The pumping was traditionally done overnight, absorbing output from coal and nuclear stations that could not easily be turned down while demand was low. In systems with a lot of solar it increasingly happens in the middle of the day instead, when there is more generation than anybody can use, which is the curve Chapter 7 draws.

The battery headline is true and it is the less useful of the two numbers available, because pumped hydro still holds the overwhelming majority of the world's stored electrical energy. Both statements describe the same fleet. Understanding why they coexist is the whole of this chapter.

Firming, the other half of this chapter’s title, is the job storage is bought to do. A firm resource is one the system can count on at a stated hour, and a variable one is not. Firming is whatever gets you from the second to the first, and storage is only one of the ways to do it: a nat gas plant with fuel behind it does the same job, and so does a contract obliging somebody else to deliver. Chapter 28 comes back to the word, because it began as a commercial promise and has turned into a question about physics.

Power and energy are different products

A generator is described by one number, its capacity in megawatts. Storage requires two.

Power, in megawatts, is how fast the device can charge or discharge. Energy, in megawatt-hours, is how much it holds. Dividing the second by the first gives duration, the number of hours the device can sustain full output before it is empty.

A 100 megawatt battery with four hours of duration and a 100 megawatt pumped hydro scheme with sixteen hours are identical on a capacity table and do entirely different jobs. Batteries have overtaken pumped hydro on power while remaining far behind on energy, because their duration is typically four hours against the eight to sixteen or more common in a reservoir scheme. Any statistic about storage that quotes one number without the other is telling half the story.

The efficiency row, which measures how much of the electricity put in comes back out again, deserves attention because it is often quoted wrongly. Lithium cells are highly efficient in isolation, and a complete installation loses energy at the inverter, in thermal management and in auxiliary loads, which brings the delivered figure close to that of pumped hydro at around 80%. Storage of any kind is a net consumer of electricity. It buys hours, and pays for them in megawatt-hours.

Why four hours

Four hours has become the default battery configuration for three reasons.

The daily shape is the first. In a system with substantial solar, the shortfall runs from late afternoon into the evening, and that gap is a few hours long rather than a few days. A device sized to cover it is sized for the problem the system has most nights.

Capacity accreditation is the second. Chapter 16 describes how a resource is credited for reliability usingeffective load carrying capability, a measure of how much extra demand a system can serve once that resource is added, and a four-hour battery scores very highly, in the range of 83 to 100% in several regions, because it covers the shortfall the system actually experiences. Extending to eight hours costs roughly twice as much in cells and adds comparatively little accredited capacity, so the return per dollar falls away sharply.

Arbitrage economics is the third. The revenue from moving energy comes from the spread between the cheapest and dearest hours of a day, and most of that spread is captured within the first few hours of duration.

Costs have collapsed underneath all of this. The global benchmark cost of a four-hour battery project fell around 27% in a single year to about 78 dollars per megawatt-hour in 2025, the lowest since tracking began in 2009, and pack prices have declined roughly 93% over the past decade.

How a battery actually earns

Chapter 13 tells this story from the market's side, and it looks different from the asset's side.

A battery does not have a product. It has a set of them, and it chooses between them every interval: frequency regulation, contingency reserves, energy arbitrage, and capacity payments where a capacity market, which pays generators simply to be available, exists. The practice of moving between these is revenue stacking, and it is why a storage business plan is a forecast about several markets at once.

Texas showed what happens when everybody stacks the same way. Batteries went first to ancillary services, the products a system buys to keep itself stable rather than to supply energy, which paid best, saturated those markets within about three years, and moved to arbitrage as the returns compressed. Ancillary revenue for ERCOT batteries fell close to 90%. The assets are fine. The market they entered was small, and they filled it.

What storage does not solve

Here the chapter has to be blunt, because storage is routinely described as the answer to intermittency and it answers one version of the problem.

A four-hour battery covers a four-hour shortfall. Winter Storm Uri lasted days. A still, cold week in northern Europe lasts a week. The seasonal mismatch between summer solar and winter heating demand lasts months.

The reason duration does not simply scale is that the cost structure works against it. Adding hours means adding cells, in proportion, while the inverter and connection are already paid for. So the cost per hour of duration stays roughly linear while the value of each additional hour falls quickly, since long shortfalls are rare and the device sits idle waiting for them. Storing energy from June to December in a chemical battery would mean an asset that cycles once a year, which no cost curve rescues.

This is why long-duration storage attracts so much attention and so little deployment. Iron-air chemistries, compressed air in caverns, thermal stores and hydrogen are all attempts at the hundred-hour problem, and all of them trade efficiency for cheapness of the energy component, which is the correct trade for something that cycles rarely. None is deployed at scale yet.

Meanwhile the firming job at seasonal timescales continues to be done the way it always was, by fuel. A nat gas plant with a pipeline behind it, or a nuclear station with a fuel load, is a storage device whose energy component costs almost nothing to enlarge. That is the comparison every long-duration proposal has to beat, and it is the argument running underneath Chapters 12, 16 and 25.