How Capacity Is Paid For

Chapter 12 ended with a question left open on purpose. A peaking plant earns nothing for most of its life and has to recover its costs from a few hours a year, and offer caps, operator intervention and ordinary caution all conspire to make those hours pay less than the theory requires. The world has produced three answers, and the map of who chose which is the third design axis.

Answer one: pay for energy and nothing else

An energy-only market pays generators for megawatt-hours and for reserves, and pays nothing at all for existing. If the system is short, prices rise, and the prospect of those prices is what persuades somebody to build. The scarcity machinery in Chapter 12, including the operating reserve demand curve, exists to make sure the price rises far enough and early enough for the arithmetic to work.

Texas and the Australian National Electricity Market both run this way, and the argument for it is clean. Nobody has to forecast how much capacity the system will need in four years, because the price does that job continuously. Nobody has to decide what counts as a megawatt of reliability. Investors take the risk, and if they misjudge it, they lose their own money rather than a consumer's.

The design depends on prices being allowed to reach levels that make the arithmetic work, and on customers and politicians tolerating those prices when they arrive. Uri tested both conditions inside a single week, and the subsequent lowering of the offer cap from 9,000 dollars to 5,000 is the clearest available evidence of where the tolerance ran out.

Answer two: buy the capacity separately

A capacity market pays generators to be available, in a market held apart from energy. PJM runs the largest example, the Reliability Pricing Model, and its shape is worth following because every capacity market resembles it.

The operator forecasts peak demand several years ahead, adds a reserve margin, and constructs a downward-sloping demand curve for capacity. It then runs a forward auction, typically about three years before the delivery year, in which resources offer to commit. Clearing resources receive a payment per megawatt per day for the whole delivery year, and in return carry an obligation: they must be available when called, and they pay penalties when they are not. The delivery-year obligation, rather than the payment, is what distinguishes this from a subsidy.

Two properties follow from the design and cause most of the argument. The quantity bought is chosen by the operator rather than discovered, and the definition of a megawatt of capacity is written by the operator too.

What the price did

PJM's recent auctions show what a capacity market produces when supply tightens against an administratively set demand curve.

The 2026/2027 auction, announced in July 2025, cleared at the cap and cost around 16.4 billion dollars for a single delivery year, with estimated retail bill effects of roughly 1.5 to 5% depending on the state. Data centre load growth, thermal retirements and an interconnection queue too slow to replace them all pushed the same way at once.

Great Britain, running the same species of mechanism, moved the other way. Its T-4 auction for delivery in 2028/29 cleared at 60 pounds per kilowatt year. The following auction, for 2029/30, cleared at 27.10 pounds, a fall of about 55%, because more capacity prequalified than the target required. Across its history the British auction has cleared anywhere between 6.44 and 65 pounds.

Answer three: buy some of it, for some people

Europe mostly sits between the two, under the label capacity remuneration mechanism, which covers arrangements of quite different ambition. A strategic reserve holds a small number of plants outside the market entirely, to be called only in emergency, which keeps them from depressing the ordinary price. A market-wide mechanism, as in Britain, France and Poland, pays every qualifying resource.

These sit inside a framework the electricity market of 2010 would not recognise. Regulation (EU) 2019/943 sets an emissions limit of 550 grammes of fossil carbon dioxide per kilowatt-hour for participation. New plants above that line cannot take capacity payments at all, and existing plants above it, together with an annual average above 350 kilogrammes per kilowatt installed, were excluded from 1 July 2025.

A capacity mechanism therefore does more than buy reliability in Europe. It selects which technologies are permitted to be reliable, which makes it an instrument of climate policy operating through a reliability product. That dual purpose is a large part of why these mechanisms are argued over as fiercely as they are.

The question that has become the hard one

Buying capacity requires deciding how many megawatts a resource is worth, and that has stopped being obvious.

A gas turbine rated at 300 megawatts contributes close to 300 megawatts of reliability, discounted by how often it breaks. A 300 megawatt solar farm contributes nothing at all on a January evening. A four-hour battery contributes fully to a four-hour shortfall and very little to a thirty-hour one. Nameplate capacity stopped being a useful measure of what a resource does for reliability.

The industry's answer is effective load carrying capability: the amount of additional demand a system can serve, at unchanged reliability, once a resource is added. It is computed from probabilistic simulation rather than engineering ratings, and the resulting numbers vary enormously. Onshore wind is commonly credited at something like 15 to 21%, four-hour batteries at 83 to 100%, and solar anywhere from single digits to over 80% depending on the region and on when that region actually has its shortfalls.

PJM applies this to renewables, storage and hydro and has been extending it across all capacity resources. MISO is moving its accreditation to a direct loss-of-load approach later this decade. The methods keep changing because the answers keep changing.

The property that makes accreditation genuinely difficult is that these values depend on each other. Adding storage raises the capacity value of solar, because the battery can move midday output into the evening when the shortfall occurs, and adding solar raises the value of storage for the same reason. A resource's contribution also falls as more of its own kind is built, since the tenth solar farm defends the same midday hours the first nine already covered. Accreditation is a moving judgement about a portfolio, recomputed as the portfolio changes, and every revision moves real money between real companies.

Where the decision sits

Set the three answers beside each other and a pattern emerges that Chapter 12 half-stated already.

A capacity market contains two administrative decisions: how much to buy, and what counts. Both are made by a regulator, years ahead, and the auction discovers only the price at which the chosen quantity of the defined product can be met. An energy-only market contains an administrative decision too, in the value of lost load and the shape of the reserve demand curve, made once and embedded in a formula that then runs continuously.

So both designs place a judgement about the worth of reliability in the hands of an authority. They differ in where that judgement sits, how often it is revisited, and how visible it is when it turns out to be wrong. An energy-only market makes it wrong quietly for years and then expensively in a single week. A capacity market makes it wrong in public, once a year, with a number attached and a bill impact somebody can put in a press release.

That completes the three axes. Who dispatches, how location is priced, and how capacity is paid for. Every system in Part Four is a combination of those three choices, and the combinations are not random: they follow from what each place was trying to protect when it wrote the rules.