Ancillary Services

Two things happened to the services that keep a power system standing up, and they happened at the same time in opposite directions.

In Texas, batteries discovered frequency regulation and reserves, poured into them, and drove the returns close to nothing. Average annual revenue for a battery in ERCOT fell from around 149 dollars per kilowatt-hour of installed capacity in 2023 to a projected 17 dollars for 2025, and the share of that revenue coming from ancillary services dropped from about 84% to roughly 48% in two years. The market did not shrink because the service stopped mattering. It shrank because it was small to begin with and something arrived that was very good at supplying it.

Meanwhile, in October 2024, the US federal regulator concluded that generators should not be paid at all for supplying reactive power within the normal operating range. Order No. 904 took effect on 27 January 2025. Three months later, a failure of reactive power control blacked out the Iberian Peninsula.

Those two facts are not connected. Different continent, different regulator, and nothing about a US tariff rule bears on what happened in Spain. But putting them beside each other exposes the thing this chapter is about, which is that the services holding the system together fall into two categories, bought in completely different ways, and which category a service falls into has remarkably little to do with how badly the system needs it.

What the system is actually buying

Ancillary services is an unhelpful name for a real distinction. Energy is the product. Everything below is a capability that has to be present so the product can be delivered at all, and each of them defends a different variable on a different timescale.

Read down the right-hand column and the pattern that emerges tracks measurability and tradability rather than importance. Where a service can be expressed as a quantity of megawatts held in reserve for an hour, a market forms. Where it is a property of being connected, or a capability nobody exercises for decades, no market forms and something else has to do the job.

The half that became a market, and then a commodity

Regulation and reserves have all the properties a market needs. The quantity required can be calculated in advance, the product is substitutable between suppliers, delivery can be measured after the fact, and it clears alongside energy, so a generator can be paid to stand ready instead of to produce. Most organised markets co-optimise the two, which simply means the operator solves for the cheapest combined way of covering both the energy and the reserve, rather than buying them separately and discovering they wanted the same machine.

Then batteries arrived, and they turned out to be almost perfectly suited to this. A battery responds in milliseconds rather than the tens of seconds a thermal machine needs, it can move up and down with equal ease, which regulation requires and a turbine dislikes, and it earns nothing sitting idle, so the opportunity cost of committing to a reserve product is low. Texas is the clearest case anywhere. By mid-2025 ERCOT had roughly 11 gigawatts of installed battery capacity, and it went where the money was.

The money then left. Ancillary service revenues for ERCOT batteries fell by close to 90%. When ERCOT introduced a new contingency reserve product, it cleared at high prices for a short period and then normalised hard as batteries qualified into it, at one point averaging under five dollars per megawatt per hour. Operators that had built business cases on ancillary revenue moved to energy arbitrage and the day-ahead market.

This matters beyond storage economics. It is the reason ancillary services cannot be the answer to the missing money problem in Chapter 12. These markets are worth a small fraction of the energy market, they saturate, and a peaking plant cannot finance itself on them.

The half nobody buys

Now the other column, and the service Chapter 6 left on the table.

Reactive power is what holds voltage where it belongs. Unlike frequency it is local, so it has to be supplied near the problem, and unlike energy it cannot usefully be shipped. Generators supply it by adjusting excitation, and the equipment to do so is part of any synchronous machine and can be built into an inverter.

In the United States it has been treated as a discrete ancillary service since the open-access rules of the 1990s, but priced on a cost basis rather than through any market, and the treatment varied by region. New York paid a flat rate. California paid nothing at all. Then in Order No. 904, issued on 17 October 2024 and effective on 27 January 2025, FERC concluded it was unjust and unreasonable to charge transmission customers for reactive power supplied within the standard power factor range, being 0.95 leading to 0.95 lagging, and prohibited such charges.

The reasoning is better than the outcome sounds, and it deserves stating properly rather than being made into a scandal. Operating within that band is a basic condition of being connected to the network. A generator earning in an energy market is already being paid to be there and to be capable of it, so a separate payment for the same capability risks paying twice for one thing. Compensation outside the standard range, and individually negotiated bilateral arrangements, both survive.

But it does place the service firmly on the obligation side of the ledger. Voltage support in much of the US is now something a generator must do, not something it is paid to do. And an obligation has a different failure mode from a purchase.

Two ways to fail, and they are not the same

This is the distinction Chapter 6 promised and could not finish, because it is an economic point rather than a physical one.

A procurement failure means the capability was never bought or never built. Not enough reserve was contracted, or nobody installed the equipment that would have held voltage in a particular corner of the network. It shows up in studies and planning documents, and the remedy is to buy more.

A compliance failure means the capability existed, somebody was obliged to provide it, and on the day they did not. Nothing was under-procured. The thing was there and it did not do its job.

Iberia was the second kind. The expert panel found non-compliant reactive power control by conventional thermal plants: machines that were required to hold voltage down and did not. No amount of additional procurement would have prevented that, because the problem was not a shortage. And this is exactly the weakness of the obligation model. A market has a settlement process that checks, every interval, whether what was bought was delivered, and it pays or penalises accordingly. An obligation is checked by whoever is enforcing the grid code, at whatever interval they choose, which in practice is far less often than every five minutes.

So the summary of the two halves is uncomfortable. The services that are bought in markets are verified continuously and have been competed down to almost no margin. The service that failed catastrophically in 2025 is, across a large part of the world, neither bought nor continuously verified.

Black start, and the limit of the idea

Black start is the reductio. If the entire system goes down, most power stations cannot restart, because starting a large thermal plant requires electricity. A small number of units, typically hydro or units with their own diesel generators, can start unaided and energise a path so that others can follow.

There is no market for this anywhere, and there never will be. The service is used perhaps once in a generation, the required quantity is a handful of specific units in specific locations rather than a fungible megawatt, and the value of having it is either zero or effectively unbounded depending on a day that has not happened. It is procured through bilateral contracts, negotiated, with the operator specifying who and where. It is the clearest case in the book of a capability the system cannot function without and a market cannot price.

Buying the thing Chapter 6 was about

Which leaves inertia and grid strength, where the interesting work is now happening, because some systems decided to stop treating them as free byproducts of having thermal plant and started buying them on purpose.

Ireland went first, and had to, since a small system with very high wind penetration hit these limits a decade before anyone else. The DS3 programme, begun in 2011, built out a set of system services well beyond the standard list, including fast frequency response, ramping margin products and a synchronous inertial response service that pays explicitly for the stored rotating energy a machine contributes. Great Britain followed with its Stability Pathfinder, which contracted for inertia and short-circuit level as products in their own right, including from synchronous condensers built for no other purpose, which are generators that burn no fuel and produce no energy and exist purely to spin.

A machine paid to rotate and generate nothing is a strange object in a commodity book, and it is the cleanest possible statement of the argument. The system was never buying only energy. It was buying energy and getting stability thrown in, because the machines that made one happened to make the other. Once that stopped being true, somebody had to write a contract for the part that used to be free.

That is why a market-design book has to spend a chapter on equipment. Chapter 12 asked whether the energy market pays enough to keep a plant available. This chapter asks the harder version: whether it pays for the properties that plant was quietly providing while everyone was arguing about the energy.