For most of my career, the smart thing to do with any electrical load you could move was to run it overnight, in the quiet hours when demand was low and power was cheap. When electric vehicles came along, the advice was no different.

It is advice almost everyone in this industry has given at one time or another. Plug in after dinner, let the car fill up while the world sleeps, and you are doing the grid a favour. You are drawing power when the big always-on plants are producing into a system that barely needs them, instead of piling onto the late-afternoon crunch. Ontario has built a whole rate around exactly this instinct. Its Ultra-Low Overnight price charges 3.9 cents a kilowatt-hour between the small hours and morning, against more than 39 cents at the afternoon peak. Ten times cheaper to wait until midnight.

And on a grid like Ontario’s, it is the right signal. But follow the logic onto a grid with a lot of solar, and something strange happens. The advice quietly turns upside down. If the cheapest, cleanest power on the system now arrives in the middle of the day, then charging your car at midnight is no longer the responsible choice. It is the expensive one. The question that opens up is bigger than EVs, and it runs underneath the whole transition: for a century we built power plants to chase demand around the clock, and now, for the first time, it may be demand that has to go chasing the power.

The hunter and the hunted

It helps to be clear about what the old arrangement actually was. From the very beginning, electricity was built on the assumption that demand is fixed and supply is flexible. You, the consumer, flick a switch whenever you please, and somewhere out of sight a generator ramps up to match you. Power balance was achieved entirely on the supply side, by adjusting generation to follow whatever load the world happened to demand. The grid spent every second of every day hunting demand, tracking its prey up the morning ramp, through the afternoon peak, and down into the overnight lull.

That made complete sense, because the generators doing the hunting were dispatchable. A gas turbine, a coal unit, a hydro dam, each can be told when to run, its fuel waiting in storage until you call for it. When demand is the thing you cannot control and supply is the thing you can, of course you build the system to let supply do the chasing.

Cheap solar breaks the symmetry. Sunshine is not dispatchable. It shows up when it shows up, in enormous quantity at noon and not at all at night, and no control room can summon it after dark. Once a large share of your generation arrives on its own schedule rather than yours, the old division of labour stops working. The supply side can no longer do all the chasing, because a growing part of it cannot be told when to run. So the hunt reverses. The cheap power becomes the fixed point, and demand becomes the thing that has to move. The hunter, after a hundred years, becomes the hunted.

How we learned to love the night

The overnight habit was not a mistake; it was a sensible response to a real problem. The tool most of us know it through is the time-of-use rate, the practice of charging more for electricity when the system is strained and less when it is slack. For decades that meant expensive afternoons and evenings and cheap nights. The logic was straightforward. Big thermal plants, especially nuclear and coal, are happiest running at a steady, constant output and dislike being ramped up and down or switched off. Through the night, when offices are dark and factories are quiet, all that steady baseload generation kept producing into a grid that barely needed it. Baseload here just means the always-on generation that runs around the clock under the daily demand. Filling that overnight trough with deferrable load, the dishwasher, the water heater, eventually the EV, was genuinely useful. It put cheap, otherwise-wasted power to work and softened the next day’s peak.

So the overnight signal earned its place. The mistake would not be having built it, but assuming it still points the right way once the generation mix beneath it has changed.

This is the test I keep coming back to, the one regular readers will recognise as the legacy-paradigm test. Take a rule everyone treats as permanent, and ask whether it is still true or merely still familiar. Is “shift your load to overnight” a law of physics, or is it a habit inherited from a grid that no longer exists? On Ontario’s nuclear-and-hydro system, with its genuine overnight surplus, the habit still holds, and the Ultra-Low Overnight rate is well designed for it. On a grid filling up with solar, the same habit sends you to charge precisely when power is scarcest and dirtiest, and to sit idle through the hours when it is nearly free. Same advice, opposite grid, opposite result. The rule did not change. The ground underneath it did.

The belly of the duck is the bargain

California has already crossed into the new territory, though it took a while to get the price signals to admit it. When I lived in Berkeley from 2019 to 2022, the rates I was on pointed almost exactly the wrong way. The cheapest hours were overnight, when the grid leaned most heavily on gas, and the daytime carried the premium, just when California’s solar was most abundant. The signal quietly rewarded the opposite of what the physical grid needed. It pulled consumption into the night, onto gas-fired power, and away from the middle of the day, when surplus solar was often so plentiful the system was curtailing it for lack of takers. It was the cleanest case I have lived through of a price signal lagging the grid it was meant to govern.

Since then the rate designers have caught up. California’s big utilities have pushed their peak pricing window back to 4 to 9 in the evening, the stretch after the sun drops but before demand does, with summer peak rates approaching 49 cents a kilowatt-hour. Midday, when solar is flooding the system, has become the cheap super-off-peak window. So after a hundred years of holding steady, the expensive hours and the cheap hours have swapped places on the clock. And it happened in the space of a few years.

In the last piece I wrote, I described the duck curve, the deep midday dip in net demand that heavy solar carves out as it pushes other generation off the system. I framed that dip as a problem to be managed, and from the supply side it is. From the demand side it is the opposite, the best deal on the grid, a predictable daily window of power that is abundant, clean, and frequently priced near zero.

A century of grid design trained us to see that surplus as something to dispose of. The more useful question is who gets to consume it. Any load steered into the middle of the day drinks from the cheapest water on the system and does the grid a second favour, filling the belly, lifting the depressed midday price toward something a solar developer can earn on, and shrinking the surplus that would otherwise be curtailed. In the two-by-two I used last time, this is the transition success case made real. Flexible demand is the missing half of a grid we have so far tried to balance using supply alone.

What can actually move, and what cannot

The catch is that not all demand can chase the sun. Flexibility exists on a spectrum, not as a binary choice, and understanding where a given load falls on that spectrum is central to the entire discussion.

At the easy end are loads with storage built into them. An electric vehicle is the clearest case. A commuter’s car sits parked far longer than it takes to charge, so the timing is almost free to move, and it does move when the price says so. A field experiment with 390 vehicles found that incentives to charge during solar hours lifted midday charging by 34 per cent, much of it shifting to workplace chargers while the car sat through the working day. Hot water tanks, building cooling, the thermal mass of a commercial space, industrial processes that can run a shift early or bank product when power is cheap, all store energy in one form or another and can absorb a midday surplus without anyone noticing.

At the hard end sit the loads chained to human schedules and physical necessity. Lights come on when it is dark. Much of a home’s demand answers to when people are actually home. A hospital, a transit system, a process that cannot be interrupted without spoiling a batch, none of these will wait for the sun. That portion of demand stays stubbornly fixed, and pretending otherwise is how flexibility gets oversold.

Then there is a kind of flexibility that is not really shifting at all. The loads at the easy end store energy and move when they draw it, so the work still gets done, just later. Another kind stores nothing. It simply does less when power is scarce and more when it is cheap, and the work it skips is gone for good. Cryptocurrency mining is the purest example. A Bitcoin miner can power down within minutes, with no process to spoil and nothing to restart gently, which makes it about the most interruptible large load on any grid. In Texas, mining and loads like it are on track for roughly a tenth of all the electricity the grid uses this year, the bulk of what the system operator classes as large flexible load, and miners are paid well to switch off when supply is tight. One operator collected around 31 million US dollars in a single hot month for curtailing. The honest caveat is that this only helps the grid if the miner genuinely flexes, rather than running around the clock on the cheapest power it can find, fossil included, and whether those payments are a grid service or a subsidy is the subject of a live political fight. But as proof that even an enormous load can be built to chase cheap power instead of demanding it on a fixed schedule, mining is the clearest case going.

Artificial intelligence raises the same question at a much larger scale, though a more constrained one, since a good deal of AI computing cannot be paused as freely as a Bitcoin mining rig. The data centres being built for it are enormous, fast-growing, and widely treated as a pure problem for the grid, a wall of new demand with no give in it. But more of that load turns out to be movable than the wall image suggests. A 2025 study from Duke University put a striking number on it. If large new loads agree to ease off for a sliver of the year, between a quarter and one per cent of annual hours, the existing grid could absorb 76 to 126 gigawatts of them with no new capacity built at all. The researchers call this curtailment-enabled headroom, room that already exists the moment a load is willing to be interrupted occasionally. The same idea works in reverse, steering that demand toward the hours when power is abundant rather than scarce. We have spent a century treating large loads as fixed and building supply to serve them. Some of the largest we are about to add can instead be shaped to fit the supply we already have.

The rivalry nobody mentions

Here is where it gets uncomfortable, because the obvious answer to the midday surplus and the one I have just spent several paragraphs on are quietly at war with each other.

When I wrote last time about why anyone still builds solar into a glutted market, the answer was storage. Charge a battery on cheap midday power, discharge it into the expensive evening, and you have rescued the economics. That arbitrage is the foundation of the entire merchant battery business, and it depends completely on the gap between the midday price and the evening price, the spread between the belly of the duck and its neck.

Now look at what flexible demand does. Every EV that charges at noon, every data centre that leans into the cheap hours, every water heater pulled into the middle of the day adds load exactly when the battery wants to buy low. That extra demand lifts the midday price. It fills the belly. And the very flatness that flexible demand creates, the thing that is so good for solar economics and for the grid as a whole, is precisely what destroys the spread the battery lives on. Flexible demand and storage are both reaching for the same cheap midday hours, and there is only so much cheapness to go around.

This is not a thought experiment. In California, four-hour price spreads compressed by 22 per cent in a single year as midday battery charging grew by half. Batteries competing with other batteries for the same arbitrage have begun to erode their own margin, and flexible demand pushes the same way. It is the cannibalisation dynamic I described for solar projects last time, where each new entrant degrades the value of the hours it depends on, now operating across two technologies fighting over one finite spread.

The honest way to hold this is that storage and flexible demand are complements for the grid and rivals for the margin. The system wants both. Each one absorbs surplus, firms up solar, and softens the evening ramp, and a serious grid will need them working together. But they earn their living from the same scarcity, and as they succeed they compete that scarcity away. Which of them captures the value is not settled by physics. It is settled by what each one costs to build and, just as much, by how the market is designed to pay them. Show me the incentive, as Charlie Munger liked to say, and I will show you the outcome. Whether storage or flexible demand comes out ahead is a choice encoded in market design, not a law of nature.

Step back and you can see what is really being repriced. For a hundred years the valuable thing on the grid was energy itself, and the system was built to produce more of it on command. In a world of abundant midday solar, raw energy is the cheap part. What becomes scarce is the ability to consume on the grid’s schedule rather than your own. Flexibility is the new product, rigidity the new cost, and the battery and the steerable load are two ways of selling the same precious thing, the willingness to show up when the sun does.

The components exist; the design is the hard part

So where does this leave the advice I started with? On a solar-rich grid, charging at midnight out of old habit is the modern equivalent of hauling water uphill. The cheap, clean power is sitting there in the middle of the day, and the technologies to use it already exist, from smart chargers and connected water heaters to interruptible industrial load and data centres that can throttle for a few dozen hours a year. As with most things in this business, the hardware is the easy part. The hard part is the design.

It comes down to a question we are only beginning to ask seriously. Who gets to be flexible, and who ends up paying for staying rigid? A homeowner with an EV and a smart tariff can ride the midday trough. A renter on a flat rate, or a business with a process that cannot move, cannot, and may end up quietly subsidising those who can. The same instrument that rewards flexibility punishes those who have none to give. Designing it fairly, so flexibility is broadly shared rather than hoarded, is one of the harder problems of the next decade.

And there is a further question behind it, one that deserves its own piece. Even once we agree flexibility is the valuable new product, it is not obvious who should capture the money it earns. The consumer who moves their load, the utility that runs the wires, the aggregator that orchestrates a thousand water heaters as a single virtual battery, each has a claim, and the rules that divide the spoils are mostly unwritten. That is where I want to go next.

For now the takeaway is narrower, and I think more useful. We spent a century perfecting a grid that chased demand, and we got very good at it. The coming grid asks demand to do some of the chasing instead, and almost none of our habits, rates, or instincts are built for that yet. The sun is not going to wait for us. The interesting work, and the real money, is in helping everything else learn to follow it.