The power industry is ordering gas turbines for 2031 — chasing demand nobody can yet see

Climate & Energy 5 min 12 sources

A record turbine order book, a data-center forecast that keeps swinging, and a bill that lands on ordinary ratepayers — plus Europe's heat toll and the EV slowdown.

Key takeaways

Turbine makers have a record order book for gas plants that won't arrive until 2031, all built for an AI-data-center demand surge nobody can yet measure.
The demand forecast is inflated because the same projects get counted in several places at once — and ordinary ratepayers already pay billions for capacity built to it.
Europe's June heat killed thousands and deepened a drought, while the EV market cooled unevenly — a demand curve arriving on its own schedule, not the industry's.

The clearest picture of the energy world this week isn’t a storm or a policy. It’s an order book. GE Vernova, one of the three big makers of heavy power-plant turbines, said its gas-turbine backlog has climbed to 116 gigawatts, up from 100 GW just one quarter earlier [1]. Its total order backlog across all its lines now stands at $176 billion, up from $129 billion a year ago [1]. The company expects combined orders and reservations to reach 125 GW by year-end [1].

That is a wall of demand. But look at the delivery dates. GE Vernova, like rivals Siemens and Mitsubishi Heavy Industries, is now booking reservations four to five years out — taking orders today for turbines that arrive in 2031 [1]. In the last quarter it shipped 3 GW of turbines while signing 20 GW of new orders and reservations [1]. The queue is growing far faster than the metal is being delivered.

What everyone is building for

The thing pulling all these orders is the same thing: the expected surge in electricity demand from AI data centers. About 20% of GE Vernova’s turbine customers are now data-center buyers, up from almost none a few years ago [1]. Southern Company’s Georgia utility just signed a 25-year deal to power OpenAI facilities [4].

Here is the catch. The demand everyone is racing to build for is a forecast, and the forecast keeps moving. Pacific Gas and Electric, California’s biggest utility, says it has 12.7 GW of data centers “in the pipeline” [2]. But only 490 megawatts of that — under 4% — have signed firm interconnection agreements [2]. And the pipeline number keeps lurching. It went from 7.3 GW at the end of last year, down to 5.4 GW, then a revised 5.1 GW, and now back up to 12.7 GW [2]. The company’s chief executive expects to actually serve just 1.8 GW of new data-center load by 2030 [2].

That gap — between the “pipeline” and the load that will really show up — is the whole story. A data center can reserve grid capacity in several places at once while it shops around. The same project gets counted more than once. So the demand signal the whole industry is building to is inflated, and nobody knows by how much.

Who pays for a forecast that overshoots

The cost is not abstract, and it doesn’t wait for 2031. A capacity auction is the yearly market that pays power plants to promise they’ll be available. In the mid-Atlantic grid run by PJM, data centers were responsible for $6.3 billion — 38% — of the charges from the most recent one [3]. Over the last four auctions, data-center-driven charges came to $29.4 billion, nearly half the total [3]. Those charges land on every customer’s bill.

PJM’s independent market monitor, Joseph Bowring, put it plainly: “PJM is continuing to act like it’s business as usual. You have to open your eyes and recognize that it is really a paradigm shift” [3]. His proposed fix is to run a separate auction for data centers, so their costs don’t spread to everyone else [3]. Until something like that happens, ordinary households are underwriting capacity built for demand that may or may not arrive.

The EV market cools, unevenly

The other half of the transition — cars — sent mixed signals this week. Tesla reported profit of 31 cents a share against the 51 cents Wall Street expected [5]. The stock had its worst day in years as heavy discounts ate into earnings [6]. General Motors is quietly reversing course. Cadillac, once set to go all-electric by 2030, will launch new gas models next spring, and the Chevy Bolt will be replaced by a gas crossover [7]. Volkswagen scrapped its hope for any 2026 revenue growth and posted a 9.5% profit slump [8].

But the picture isn’t a collapse. In Europe, electrified cars kept driving overall auto-market growth in June, offsetting a sharp fall in petrol and diesel sales — with Chinese brands gaining ground fast [9]. The lesson buried in both stories is the same as the turbines. Companies placed big bets on a demand curve, and the curve is arriving on its own schedule, not theirs.

Europe counts the heat

While the industry argues about future megawatts, the present delivered a grim number. France’s national health agency recorded 5,764 excess deaths between 17 June and 2 July — a 36% excess mortality rate, with two-thirds of the dead aged 75 or older [10]. A separate scientific analysis found that fossil-fuel warming made this summer’s western European drought about 80 times more likely [11]. It turned what would have been a “moderate” drought into an “extreme” one [11]. In Spain, a wildfire near Guadalajara burned 32,000 hectares and forced 1,200 people from their homes [10].

The under-covered one: China’s clean giants still lean on coal

Finally, a useful corrective to a familiar headline. China is building more large-scale wind and solar than the rest of the world combined. But a report this week found two things are limiting how much of that clean power actually gets used [12]. One is inadequate transmission lines — the high-voltage wires that carry power from remote deserts to cities. The other is continued reliance on coal-fired backup [12]. Building the panels is the easy part. Moving the electricity, and retiring the coal behind it, is the slow one — a reminder that capacity on paper is not the same as clean power delivered.

02 · Lesson · why it matters

Why the cure for a shortage often arrives as a glut

When a shortage takes years to fix, the fix lands late — and a late correction sails past the target, not onto it.

A shower with a slow tap

Anyone who has fought a shower with a slow tap knows this feeling. The water is cold, so you turn it hot. Nothing changes, so you turn it further. Then the heat finally arrives — all of it at once — and you leap back and crank it cold. You spend the next minute swinging between scalding and freezing, never landing on warm.

The problem was never the tap. It was the delay. Every correction you made was aimed at how cold the water was, not how hot it would become once your last move caught up. When the response to a signal lags, the fix keeps overshooting the target.

The power industry is standing in exactly that shower. Turbine makers have a record order book — 116 gigawatts of gas turbines on backlog at one company alone. But the turbines take four to five years to deliver. Orders placed today arrive in 2031. Everyone feels the water is cold now, and everyone is cranking the tap.

Why the fix lands late

A fast system corrects itself gently. You feel a chill, you nudge the dial, the water warms, you stop. The feedback is quick, so small moves do the job.

A slow system can’t do that. The gap between deciding to build a power plant and switching it on is years. During those years the shortage is still visible, still pressing, still on the news. So the natural human move is to keep ordering — more turbines, more plants, more reservations. Each order is sensible on its own. Together they pile up into a wave of supply that will all land around the same time, years from now.

When that wave arrives, it may arrive into a world that no longer needs it. The shortage that justified it has, by then, been over-corrected. A steadying force — build more when you’re short — becomes a swinging one, because the steadying happens too late to steady anything.

The target keeps moving

Overshoot is bad enough when you at least know the target. Here, nobody does.

The whole build-out is aimed at one forecast: the surge in electricity that AI data centers are expected to need. But that forecast keeps lurching. One large California utility reported its data-center “pipeline” at 7.3 gigawatts, then 5.4, then a revised 5.1, and then 12.7 — inside a single year. Of that latest figure, under four percent have actually signed firm agreements. The company’s own estimate of what it will really serve by 2030 is a fraction of the headline number.

The number moves because a data center can reserve grid capacity in several places at once while it decides where to land. The same project gets counted twice, three times, four. So the demand signal the whole industry is building toward is not just delayed in its answer — it is blurry in its question. They are aiming a five-year correction at a target that won’t hold still.

The arrangement that hides the risk

Here is the part that poses as normal. Why can a data center inflate the forecast by reserving capacity it may never use? Because the rules let it. Reserving a slot in the interconnection queue is cheap and carries little commitment. Promising future load in a capacity market costs the promiser almost nothing if the load never shows.

That arrangement isn’t an accident, and it isn’t only villainy. It serves the data-center firms, who get to keep their options open. It serves the utilities, whose investment base grows with every plant approved. And it does keep the lights reliably on, which everyone wants. But it quietly moves the cost of overshoot onto the people who never placed an order. In one large grid, data centers already account for billions in capacity charges — 38% of the latest auction. Those charges are spread across every household bill.

So the risk of building too much, for demand that may never arrive, sits with the ratepayer, while the option to inflate the demand sits with the buyer. That split is a choice someone made in the rules. It reads like plain arithmetic on your bill.

Nobody is above the fog

It is tempting to end by pointing at whoever is getting it wrong. But the honest picture is that no one in this system can see the whole of it. The turbine maker sees its order book, not whether the orders are real. The utility sees its pipeline, not how many times the same project was counted elsewhere. The regulator sees the capacity charges, not the true demand behind them. The market monitor watching all of it can only say the current rules aren’t built for the moment.

And the reader is not outside this, watching. The reader is a node in it — paying now, in capacity charges, for turbines that arrive in 2031, for demand that is still a rumour. A distant boom in a technology you may never touch is already priced into your electricity.

The lesson isn’t that someone is foolish. It’s that a delay turns even careful, sensible corrections into a swing. When the answer to a question takes five years, the question has usually changed by the time the answer lands. The humble move, standing in that shower, is to turn the dial in smaller amounts than the cold seems to demand. And to remember that the person cranking the tap next to you can’t see the water any better than you can.

03 · Lab · your turn

Build to the Forecast

Order power capacity against a rising-then-flat demand with a four-year build delay, and feel how a strong reaction overshoots into a glut.

04 · Hope · carry this

A system that swings is also a system that can be taught its rhythm, and the utilities now tightening their forecasts and the watchdogs naming the overshoot out loud are quietly doing exactly that. We have learned to steady slower machines before; there is no reason we cannot learn to steady this one.