Load
Shape, weather sensitivity, degree days, the duck curve, and the return of load growth after roughly two flat decades.
Words the industry uses are marked like this at the sentence that defines them, so you can tell a term you will hear on a desk from a phrase this book happens to be using.
Load is what the power industry calls demand. It is the only quantity in this book that nobody decides. Generation is offered, dispatched and instructed. Demand is the sum of several hundred million independent actions taken by people who have no idea what the wholesale price is and would not change their behaviour if they did.
The operator forecasts it, plans around it, and never controls it. Almost every design question in the preceding chapters is a question about how to cope with that.
Who uses it
Ask what electricity is for and the answers come back air conditioning, electric cars and data centres. Two of the three are small. Electric vehicles used about 180 TWh in 2024, roughly 0.7% of final electricity consumption, and data centres about 415 TWh, roughly 1.5%. Air conditioning is the exception, at close to a tenth of world electricity, and it carries a second property that matters more on a power system than its share of the energy: cooling accounts for close to 30% of peak demand where it is hot. The next section is about why that gap between energy and peak exists. The largest single use of electricity on earth is still industry, and it is not close.
- 42%Industry
- 27%Homes
- 21%Commercial
- 10%Other
- 25.7%Industry
- 37.3%Homes
- 36.8%Commercial
- 0.2%Other
- 35.1%Industry
- 29.7%Homes
- 30.2%Commercial
- 5%Other
- 64.8%Industry
- 15.2%Homes
- 18.6%Commercial
- 1.4%Other
Share of final electricity consumption. World from the IEA; United States from EIA retail sales, 2025; European Union from Eurostat, 2023; China from the China Electricity Council, 2024. Commercial covers shops, offices, schools, hospitals and public services. Other is transport plus agriculture and unallocated, which runs 0.2% to 5% everywhere except the world total, where unallocated consumption is larger. Two definitional notes, because they move the numbers: the US figures are retail sales and so exclude industrial self-generation, which understates industry there, and China's official classification puts transport inside the tertiary sector rather than reporting it separately.
The four pies also show that there is no single answer. Industry takes 42% of world electricity and almost two thirds of China’s, which is what an economy that makes things for everybody else looks like from the meter. The United States is the outlier in the other direction: homes and commercial buildings together take 74% of it, and American industry uses barely a quarter. Europe sits between the two. A market design that suits one of these does not automatically suit another, which is part of why Part Three treats the design choices as choices.
Where the popular answers do belong is in the growth rather than the level, and even there they are not first. Between 2024 and 2030 the IEA has electric vehicles adding about 838 TWh and air conditioning about 651 TWh, against data centres accounting for less than a tenth of the growth in world electricity demand. Chapter 25 is about data centres because of where they land and how fast they arrive, rather than because they are the biggest thing on the system.
The shape
Aggregate demand is remarkably predictable in shape and stubbornly variable in level. It rises through the morning, holds through the working day, peaks in the early evening as people return home, and falls overnight. Weekends run several per cent below weekdays. Holidays are distinctive enough that forecasters treat them separately.
On top of that sits weather, which is the largest single driver of variation. The industry measures it in degree days: for each day, the difference between the average temperature and a reference point, usually 65 degrees Fahrenheit or 18 Celsius, accumulated across the heating or cooling season. Degree days are the standard variable in any load model, and they appear again in Chapter 24 because weather derivatives are written on them.
The relationship is not linear, which matters more than it sounds. Moving from 30 to 32 degrees Celsius adds far more load than moving from 20 to 22, because at the higher temperature every air conditioner in the region is already running and the marginal degree brings in the last of them and pushes the rest toward full duty. Load curves bend upward at both extremes, and system operators worry about the bends rather than the averages.
The demand curve that is almost vertical
Chapter 8 draws demand as a vertical line through the supply stack, and that deserves justification, since a vertical demand curve would be a strange object in any other commodity.
In electricity it is close to accurate, because the overwhelming majority of consumers face a fixed retail tariff. A household paying a flat rate per kilowatt-hour sees no signal when the wholesale price moves from 30 dollars to 3,000, and would need a device to respond even if it wanted to. Demand therefore does not clear the market. Supply does all the work, which is why so much of this book is about how to make supply appear at the right moment.
Demand response is the attempt to fix this, by paying large consumers to reduce load on instruction. It works, it is genuinely valuable, and it remains small relative to total demand. The interesting development is that the data centres in Chapter 25 are the first new load in decades large enough, concentrated enough, and sophisticated enough to negotiate about when they consume.
The duck
Large-scale solar installations over the past two decades have meant that net load, meaning demand minus wind and solar output, resembles a duck shape when charted.
The mechanism is that solar does not reduce demand evenly. It removes a large quantity in the middle of the day and none at all after sunset. So the curve the dispatchable fleet, meaning the generators an operator can actually instruct, must serve develops a deep midday belly and a steep evening neck, and the deeper the belly the more violent the neck.
Two consequences run through the rest of the book. The belly is where Chapter 11's negative prices occur. The neck is a ramping problem, requiring the system to add many gigawatts within two or three hours, every evening, at a time when solar is falling and demand is rising together. California's evening ramp is the reason its storage fleet exists, and the reason flexibility became a product worth paying for.
Drag it far enough and net load goes negative in the shaded midday hours, which is the condition Chapter 11 describes. The dispatchable fleet still has to cover the evening, and has fewer hours in which to earn the revenue that keeps it available.
Twenty flat years, and then this
The most consequential fact about American electricity demand is that for about fifteen years it did not grow. Between 2005 and 2019 consumption rose by roughly 0.1% a year, which is indistinguishable from flat. Efficiency gains in lighting, appliances and industry offset population and economic growth almost exactly. Energy intensive industries were offshored to China and elsewhere.
Every market design described in this book was built or matured during that period. Capacity auctions, interconnection queues (the waiting list for permission to connect a new plant), planning reserve margins (the cushion of capacity held above expected peak demand) and transmission approval processes were all shaped by a world in which the central problem was allocating a fixed quantity of demand among competing suppliers, and in which a new power plant was replacing something rather than adding to a total.
That period has ended. US consumption reached an all-time high in 2024 and has continued rising, with data centres the largest single driver alongside manufacturing investment and the electrification of vehicles and heat. Data centres accounted for roughly 3.7% of American electricity in 2023, and credible projections run to something near 12%. Forecasts of total demand growth to 2040 reach as high as 50%.
A system built to allocate a fixed quantity is being asked to expand one. Those are different problems, and most of the institutional stress in Chapters 17 and 25, from PJM's capacity prices to ERCOT's interconnection queue, comes from applying machinery designed for the first to the second.
Forecasts of this kind have been wrong before, and the queue figures in Chapters 17 and 18 include a great deal that will never be built. What cannot be argued away is the direction. After fifteen years in which the hardest question was which plant to retire, the hardest question has become which to connect, and Part Two takes up how a market decides.