The Grid

A generator in Ohio signs a contract to deliver a hundred megawatts to a customer in Virginia. The money moves from the customer to the generator. The electricity does not travel from Ohio to Virginia.

What happens instead is that the generator injects a hundred megawatts at its own connection point, the customer withdraws a hundred megawatts at theirs, and the power flows across every conducting path between the two, splitting itself among them according to the electrical properties of the network. Some of it goes through Pennsylvania. Some through West Virginia. A measurable quantity may travel through Ontario. None of that was chosen, and changing it means physically altering the network.

Almost everything peculiar about electricity markets traces back to this one fact, so it gets a chapter before the market chapters begin.

Two networks with different jobs

The grid is two systems that get discussed as one: transmission and distribution.

Transmission moves bulk power over distance at high voltage. The American industry counts it in circuit miles, and there are roughly 600,000 of them, of which about 240,000 miles run at the highest voltages. Distribution takes power from substations to premises at low voltage, and it is a vastly larger network in length, more than 5.5 million miles of it, hung on something over 180 million poles.

Voltage is what separates them, and the reason is loss. Power delivered is voltage multiplied by current, while losses in a conductor (power line) rise with the square of the current. Sending the same power at a higher voltage means lower current, and lower current means dramatically lower loss. A line at 765 kilovolts carries power with roughly half the losses of one at a lower transmission voltage, which is why the highest voltages exist at all. Total losses across American transmission and distribution together run near 5.9% of the electricity generated.

Changing the voltage is what a transformer does, and every one of them is Faraday’s iron ring from Chapter 2 grown up: two coils sharing a magnetic core, where a changing current in one induces a current in the other, and the ratio of turns between the two coils sets the ratio of the voltages. More turns on the output side and the voltage rises.

A generator step-up unit lifts a plant’s output from the tens of kilovolts it is made at to transmission voltage. Substation transformers bring it back down for distribution, and a small unit on a pole or a concrete pad drops it again to what arrives at a wall socket. Power passes through four or so of them between the turbine and the kettle, each giving up a fraction of a percent on the way.

This is also what settled the current war described in Chapter 2. A transformer works only on a changing current, so direct current at the voltages of the 1880s could not be stepped up for transmission and back down for use. Alternating current could, and that single property decided which system got built.

A transformer changes voltage and nothing else, and it works only on alternating current. Moving between the two kinds of current takes different equipment: a rectifier turns alternating into direct, and an inverter turns direct into alternating. Those three devices are most of what a grid does to electricity between the generator and the appliance. A rectifier is what turns the alternating current at a charging point into the direct current a car battery can take, and where that rectifier sits explains something every electric-car driver has noticed. On a home charger it is inside the car, sized for what a car can reasonably carry and cooled by whatever room is left, which caps most vehicles near 7 or 11 kilowatts. A motorway fast charger is a large cabinet because the rectifier is in the cabinet instead, liquid-cooled and carrying no weight penalty, which is how the same car can accept 150 or 350. An inverter is what connects every solar panel and every battery in this book. And the converter stations at the two ends of the HVDC link described below are a rectifier and an inverter facing each other across a very long wire.

The economic consequence of the losses is that a megawatt-hour delivered far from its source requires slightly more than a megawatt-hour to be produced, which is the loss component of the locational price in Chapter 10.

A three-phase grid

Look up at a transmission tower and the conductors come in threes. Electricity is generated, transmitted and mostly consumed as three separate circuits, meaning three separate wires running all the way back to three sets of coils spaced a third of the way around the generator, timed a third of a cycle apart, which is what three-phase means. Each of those conductors, each of those wires, is a phase. The arrangement exists because the three waves sum to a constant, so a motor fed from all three turns smoothly rather than in pulses, and because three-phase moves a given amount of power with less conductor than single-phase would.

Three is not arbitrary, and it was not the only number anybody tried. A single phase works, and is what most houses get. The live wire is black or red in American practice and brown in European, and it needs a return conductor beside it, white or grey in America and blue in Europe. Its power falls to zero twice a cycle, so a single-phase motor will not start on its own. Tesla’s original polyphase patents were two-phase and Niagara opened that way in 1895, which takes four wires, or three with an oversized neutral carrying the imbalance. Three is the smallest number whose currents cancel: spaced a third of a cycle apart and balanced, they sum to zero at every instant, so no return conductor is needed at all and three wires carry three circuits.

Above three the saving flattens while each additional phase costs another conductor, another insulator string, another breaker pole and more transformer, so nothing higher is used on a tower. Higher orders do exist where the economics differ: six and twelve inside the converter transformers of an HVDC link, and five-phase and seven-phase motor drives that keep turning when one phase is lost.

The matter was settled in public in 1891, when a link built to Mikhail Dolivo-Dobrovolsky’s design at AEG carried power 175 kilometres from Lauffen to the electrical exhibition in Frankfurt at 15,000 volts. After that, new transmission was three-phase.

It also settles something people half-remember about birds. A bird on a wire is safe because both feet are on the same conductor, and along a few centimetres of good conductor there is almost no voltage difference, so nothing flows through the bird. The number of wires has nothing to do with it, and neither does being clear of the ground. What kills birds is bridging two phases, or a phase and an earthed structure, which is why large raptors and storks die on distribution lines whose conductors sit closer together than their wingspan. A wire swinging into another wire in a storm and an eagle bridging two of them are the same failure.

Why power cannot be routed

A pipeline moves nat gas from an inlet to an outlet along a defined path. A shipper books capacity on a specific route and the molecules follow it. Readers arriving from NatGas 101 will have that model in mind, and it does not transfer.

An alternating current network obeys Kirchhoff's laws, which means that when power is injected at one point and withdrawn at another, it distributes itself across all available parallel paths in inverse proportion to their impedance, which is the opposition a path presents to alternating current, combining ordinary resistance with effects that depend on frequency. A path with half the impedance carries twice the flow. This happens instantaneously, automatically, and with no regard for who contracted with whom.

The practical results are strange enough that European regulators have spent decades on them. Power scheduled from Germany to Austria physically flows through Poland and the Czech Republic, which is called loop flow or unscheduled flow, and it consumes capacity on networks belonging to countries that are not party to the transaction. Poland and the Czech Republic eventually installed phase-shifting transformers on their borders, devices whose purpose is to push back against flows they never agreed to carry.

Those devices are worth a sentence, because their name misleads. A phase-shifting transformer does not raise or lower voltage, which is what an ordinary transformer does, and it does not add impedance either. Real power on an alternating current line depends on the angle between the voltages at its two ends, so the device injects a voltage at right angles to the line's own and shifts that angle. That lets an operator dial a flow up, hold it down, or push it backwards, and it is the closest thing an alternating current network has to a valve.

It only redirects, though. Flow pushed off the Polish border does not disappear, it takes some other path through the same meshed network and becomes somebody else's problem, which is this section's argument arriving from the other side.

Limits, and what happens at them

In the power industry, people say “line” instead of “power line” or power cable. Every line has a rating, and the constraint is usually heat. Current warms the conductor, the conductor expands, and the line sags closer to whatever sits beneath it. Exceed the limit for long enough and the line either faults against something or is tripped out by its protection. The 2003 blackout across the American Northeast began with sagging lines in Ohio contacting trees. What turned three local faults into a cascade that reached fifty-five million people was that the control room did not know. A race condition had locked up the alarm process about an hour earlier and nothing announced it, so the operators sat watching a screen that was telling them everything was fine.

Ratings are not fixed numbers. A line carries more on a cold windy day than on a still hot one, because ambient conditions determine how fast the conductor sheds heat, which is the basis of dynamic line rating. Some limits are not thermal at all, and are set instead by voltage stability or by the need for the system to remain stable after a fault, which is where Chapter 6 picks up.

Which raises a fair question, since a cold windy day is also when ice forms. If current heats the conductor, why does an ice storm bring lines down?

Because the two constraints are different. The heating is modest, tens of degrees at heavy load, and an ice storm supplies freezing rain and wind to carry that heat away, on distribution lines that were never carrying much current to begin with. A line that has already tripped carries none at all and ices faster for it. In the Quebec ice storm of January 1998 the radial ice reached 70 millimetres and added close to 1,400 kilograms to every hundred metres of conductor, against towers designed for 30, and 1,500 of them were damaged. Most outages in a smaller storm are not the line failing at all but ice-laden branches coming down on it, and ice also reshapes a round conductor into a crude aerofoil, so wind sets the span galloping in oscillations metres deep until the phases touch and short the line.

So a line in an ice storm is at its most thermally capable and its most mechanically vulnerable at the same moment. The intuition behind the question is sound enough that it has been built on purpose: after 1998 Hydro-Québec installed the Lévis de-icer, a 250 megawatt converter station whose job is to force enough current through an alternating current line to melt the ice off it.

The rest of the year the same power electronics run as a voltage support device, and that has nothing to do with storing anything the way a battery does. Voltage on an alternating current network is set by the balance ofreactive power, the component of the flow that does no useful work, so equipment that pushes reactive power onto the network raises local voltage and equipment that absorbs it lowers local voltage. The de-icer does that continuously and fast, which is the service Chapter 6 explains and Chapter 13 puts a price on.

When a limit binds, the operator can no longer dispatch the cheapest available generation, and has to back down something cheap on one side of the constraint while starting something expensive on the other. That substitution is congestion. It is the largest term in the locational prices of Chapter 10, and the reason the design argument in Chapter 15 exists.

Four machines, not one

North America runs as four separate synchronous areas: the Eastern Interconnection, the Western Interconnection, ERCOT covering most of Texas, and Quebec. Within each, every generator turns in step at the same frequency. Between them, there is no synchronous connection at all.

Synchronous is meant literally. Every one of those thousands of generators turns in exact step with all the others, permanently, and the coupling that holds them there is magnetic rather than mechanical. The combined field of the network rotates at the system frequency, and each rotor is locked to it the way two magnets are locked across a gap.

Which produces the fact that surprises everybody. Open the steam valve on a turbine and it does not speed up. The rotor advances a fraction of a turn against the rotating field, the magnetic pull opposing it grows in proportion, and the machine delivers more power at exactly the same speed. That angle, and not the speed, is what the operator is really moving when the plant is told to produce more. The feedback does run through the wires, and it arrives as torque rather than as an instruction.

Two consequences follow. Frequency only moves when generation and demand across the whole area disagree, because a surplus of mechanical power accelerates every rotor at once and a deficit slows them all together, which is why one number tells an operator whether the balance in Chapter 1 is holding. And the magnetic grip can be broken: push the angle past about ninety degrees and the machine slips a pole and falls out of step, with the violence that implies. That is what a weak tie between two large systems risks, and it is why the coast-to-coast experiment did not survive.

Power still crosses those boundaries, through direct current ties that convert alternating current to DC, move it, and convert it back. The conversion is what allows two systems with no common frequency reference to exchange energy. It also means the transfers are small relative to the size of the systems, and that a generator in one interconnection is, for most commercial purposes, unavailable to a customer in another.

Why four, and not one

The obvious question is why a continent that shares a currency, a railway gauge and a set of plug sockets did not end up sharing a grid. The boundaries have three different causes.

The Eastern and Western split is geography and physics, in that order. Each grew by accretion, as neighbouring utilities tied their systems together for mutual support, and the accretion stopped where the load stopped, load being the industry’s word for electricity demand. Between the Mississippi and the coastal west lies a great deal of mountain and a small number of customers, so there was never enough demand in between to justify the heavy transmission that synchronising two continental systems requires. A thin tie is worse than no tie, because two large systems joined by a weak link do not behave as one machine. They swing against each other.

This is not hypothetical. It was tried. After the November 1965 blackout that darkened much of the northeast, a federal task force activated four AC ties between the two systems, and from 1967 to 1975 something close to a single coast-to-coast machine actually existed, carrying roughly 94% of US generating capacity. It did not hold. The two halves oscillated against each other, power flowed across the ties that nobody had scheduled and nobody wanted, lines overloaded, and the system broke apart repeatedly. The ties were cut in 1975 and the continent has run as separate machines since. High voltage direct current, described below, arrived as the better answer to the same problem: it moves the energy without demanding that the two systems agree on a rhythm.

Quebec is a variation on the same reasoning. Its system was built around remote hydro sending power a very long way, it already ran on direct current for its own backbone, and it exports heavily into a neighbour it has no need to be synchronous with. It trades with the Eastern Interconnection through DC ties and variable frequency transformers instead.

ERCOT is the one boundary that is not about physics at all. Texas stayed out on purpose, because a utility that does not send power across a state line is not engaged in interstate commerce and therefore falls outside federal jurisdiction. That is a legal position rather than an engineering one, and it has been defended as such, which Chapter 18 takes up in detail. The physics is the same as anywhere; the reason for the seam is the Federal Power Act.

The distinction carries forward. Two of these seams exist because a machine that large is hard to hold in step, and one exists because somebody preferred a regulator they could reach.

Chapter 1 made the argument that electricity cannot leave the grid it was born on. The four interconnections are what that means concretely.

What direct current changes

High voltage direct current, or HVDC, removes two of the constraints described above.

Flow on an HVDC link is controlled rather than determined by impedance. The AC/DC converters set how much power moves and in which direction, so an operator can route power on an HVDC line in the way a pipeline shipper would expect. And losses over long distances are lower: a line at plus or minus 800 kilovolts loses roughly 3% per thousand kilometres against something nearer 7% for alternating current, at the cost of converter stations, the equipment at each end that performs the conversion, which consume around 0.6% of the energy passing through them.

Those properties make HVDC the technology for three jobs: very long distances where the loss saving outweighs the converter cost, undersea crossings where alternating current is impractical, and connections between systems that are not synchronised with each other. The DC ties out of Texas exist for the third reason.

The reason HVDC has not replaced the alternating current network is cost and topology. Converter stations are expensive, and a DC link connects two points, whereas a meshed AC network connects everything to everything with no equipment at the junctions. The grid is meshed because meshing is cheap, and it produces loop flow because meshing is cheap.

Cables under the sea, and the case for the Sahara

The undersea job is the busiest of the three in European waters, and the reasoning behind it generalises.

Viking Link, a 1,400 megawatt link between Lincolnshire (UK) and Jutland (Denmark), began commercial operation at the end of 2023 at 765 kilometres, the longest combined land and subsea HVDC connection built. The Celtic Interconnector, 700 megawatts over 575 kilometres between east Cork (Ireland) and Brittany (France) at a cost near 1.6 billion euros, is in commissioning as this is written. More are under construction across the North Sea and the Mediterranean.

An interconnector earns the difference between the prices at its two ends in the same instant, so its value rises with how often those two prices disagree. Two systems running mostly thermal plant on similar fuel tend to agree, and there is little to collect. Two systems running large amounts of weather-driven generation disagree constantly, because the wind is not blowing in both places at once, and every hour of disagreement is revenue. The build-out of subsea links is therefore a consequence of the renewables boom rather than a separate story. The Celtic Interconnector carries a second motive: it is Ireland’s first direct electrical connection to the European Union grid, which until now it could reach only through Britain.

Which raises the obvious question. The Sahara receives more sunlight than any populated part of Europe, land there is close to free, and Europe pays some of the highest power prices in the world. Why is there not a cable?

The losses are not the answer, and the arithmetic says so. At the 3% per thousand kilometres quoted above, a 3,800 kilometre run from southern Morocco to southern England arrives with roughly nine tenths of what it started with, plus the converter stations at each end. Nine tenths of very cheap electricity is still cheap electricity.

It has been tried twice. Desertec, launched in 2009, proposed roughly 400 billion euros of North African solar and transmission, of which the transmission was somewhere between a tenth and a quarter, so the great majority of it was the solar plant itself. It was, at one point, the most discussed energy project in Europe, and its industrial consortium fell from seventeen partners to three by the end of 2014. Xlinks proposed something narrower and more buildable: 3.6 gigawatts from Moroccan solar and wind with storage, over 3,800 kilometres of subsea cable, landing in Devon. The company put that at seven million British homes and 8% of the country’s electricity, for around 24 billion pounds, which is roughly 3,400 pounds a household. In June 2025 the UK government declined to support it, concluding that it was not in the national interest at that time, and the developer has continued to look for private backing since.

None of the reasons either project stalled is electrical. Three of them apply to every long link anybody proposes.

The destination got cheaper while the cable did not. Between the launch of Desertec and the rejection of Xlinks, the cost of solar and of offshore wind inside Europe collapsed. A cable competes against whatever the importing country could build for the same money at home, and that alternative has been improving faster than subsea cable has.

A cable concentrates a supply that was diversified. A country buying nat gas buys it from several suppliers on several routes and can substitute one for another. A country buying through a single 3.6 gigawatt cable from a single counterparty in a single jurisdiction has bought a supply with one point of failure and one government behind it. After 2022 that argument became much harder to answer than it had been in 2009.

And the money has to be committed against a price nobody can quote that far out. The link costs most of its total up front, earns over thirty years or more, and there is no forward market at that horizon in which to fix the spread it lives on, which is the same problem Chapter 12 describes for generation and Chapter 16 answers with an administrative auction.

The general form: the question for a long link is never whether the electricity arrives. It is whether it arrives cheaper than the same money spent at the far end, against a destination that keeps getting cheaper.

The people who build it and put it back

Everything in this chapter is installed and repaired by hand, at height, at voltages that do not forgive, frequently in the weather that caused the problem. The trade is line work, and a lineworker reaches journeyman status through an apprenticeship of about four years and several thousand supervised hours. It is among the more dangerous occupations the US Bureau of Labor Statistics tracks, and it cannot be done remotely, quickly trained, or automated to any meaningful degree.

The arrangement that keeps the lights coming back on after a hurricane is unusual enough to state plainly, because it is the one part of this industry that is not a market at all. Under standing regional mutual assistance agreements, utilities lend each other crews. When a storm is forecast, the utility in its path issues a request, and hundreds or thousands of trucks convoy in from states that are not affected, often arriving before landfall. The requesting utility pays the costs. Nobody bids and nobody clears, and scarcity plays no part in what is charged.

It works because of the structure Chapter 14 describes. A utility with an exclusive service territory is not lending crews to a competitor, because it has no competitor; it is buying insurance against the week it is the one underwater. Restoration capacity is held as a shared pool by an industry that auctions almost everything else.

What stops a company running thin and living off everybody else’s crews is the regulator of Chapter 14, who reviews its own staffing and its restoration record in a rate case. That, and the fact that a company which has never sent crews finds it harder to get them.

The pool is now under pressure from two directions at once. The workforce that built most of the existing network is retiring, and an apprenticeship cannot be compressed below four years by paying more for it. At the same time the same crews are wanted for the largest transmission and connection buildout in decades, which Chapter 25 describes from the demand side. Skilled labour behaves like every other item in Chapter 26: it has a lead time, that lead time does not respond to price in the short run, and it is increasingly the thing that sets when a project can actually be energised.

Why this chapter comes before the market chapters

Three properties established here determine most of what follows.

Power flows along every path rather than a chosen one, so a market cannot sell transmission as a route. Lines have limits, so the cheapest generation is frequently unusable and something dearer runs instead. And the whole arrangement is divided into synchronous islands, so a surplus in one place cannot reach a shortage in another.

A market design that ignores all three still produces prices. It produces prices that describe a system nobody has, which is the subject of Chapter 15.