Regulators move to clear a 438-gigawatt traffic jam of data centers waiting for power

Climate & Energy 4 min 78 sources

A flood of speculative AI data center requests has gummed up the power grid; FERC and Texas are now changing the rules to sort real projects from phantom ones.

Key takeaways

Texas's grid is sitting on 438 gigawatts of data center connection requests — five times the size of the actual grid — because developers file speculatively to hold a spot in line.
FERC and Texas are changing the rules to prioritize "real, financeable, flexible" projects, partly to stop ordinary ratepayers from paying for infrastructure built for projects that never get built.
China faces the same crunch: AI data centers can't easily flex their power use, which makes them a poor match for both renewables and a strained grid.

The thing slowing the electricity grid right now is not a shortage of power plants. It is a queue.

In Texas, the grid operator is sitting on requests for 438 gigawatts of new large customers — mostly AI data centers — waiting to plug in [24]. For scale, the entire Texas grid peaks at roughly 85 gigawatts on the hottest summer afternoon. The waiting list is five times the size of the grid it wants to join. Almost none of it will get built. But nobody can easily tell which projects are real and which are placeholders.

On Monday, the Texas grid operator, ERCOT, approved a new process to start sorting through the pile [24]. The first batch of projects to run the gauntlet has a blunt name: “Batch Zero.” And last week the Federal Energy Regulatory Commission — FERC, the agency that referees the interstate power system — issued six orders telling regional grid operators across the country their rules for connecting these giant new customers are not good enough [30].

Why the queue is so swollen

For years, U.S. electricity demand barely grew [30]. Then, about two years ago, the AI building boom hit, and the picture flipped. Data centers — warehouses of computer chips that run AI — need enormous, steady power.

That set off a race. A developer who gets to the front of the interconnection queue gets power sooner, so the rational move is to file early, file big, and file in several places at once to hold a spot [30]. The result is a queue stuffed with requests that may never become real buildings — the 438-gigawatt figure in Texas is what that looks like. FERC’s own response drew more than 3,500 pages of comments and took eight months to write [30].

The cost of that traffic jam does not stay with the developers. Building out transmission lines and substations is expensive, and FERC said one of its main goals is “prevention of cost shifts” so that ordinary ratepayers do not foot the bill for infrastructure built to serve speculative projects [30]. Rising electricity bills have already sparked political backlash in several states [30].

The rule change is about telling real from phantom

FERC’s signal to developers is plain. The agency will prioritize projects that can prove they are, in its words, “real, financeable, operationally flexible, and capable of integrating with the grid without imposing unjustified costs on other customers” [30].

“Operationally flexible” is the interesting part. ERCOT’s new Texas process opens a path for large customers who agree to let the grid operator curtail their power use when the local grid is strained [24]. Curtailment here means the data center agrees, in advance, to power down for a stretch when the grid is tight — in exchange for getting connected faster. A flexible customer is far easier to plug in than one that demands full power every second.

The deeper move is to change what each developer is rewarded for. Right now, the incentive is to hoard a place in line. The new rules try to make hoarding cost something — real commitments, financial responsibility for the infrastructure, a willingness to flex — so that the queue starts to reflect actual demand instead of everyone’s worst-case land grab.

The same jam, a continent away

China is hitting the same wall from a different direction. Beijing wants renewables to supply four-fifths of its data center power by 2030, up from just 11% in 2023 [32]. But power demand from Chinese data centers is set to rise by 300 to 500 billion kilowatt-hours by 2030 — the low end alone roughly equals the United Kingdom’s entire yearly electricity use [32].

The problem is the same one Texas faces: these customers can’t easily flex. “GPUs are very expensive, so once they are purchased, operators want to use them as quickly and as intensively as possible,” one Chinese power executive said [32]. Grid operators there are wary too, worried that direct green-power links to data centers would cut into the electricity sales that pay off their grid investments [32]. Everyone’s individual incentive points away from the flexible, shared solution.

What’s actually getting built

Underneath the queue, real deals are landing. Microsoft said Monday it will build a 2-gigawatt data center in Pecos, Texas, and signed a 20-year deal with Chevron for a new gas plant — “Project Kilby” — that will ramp to 2.67 gigawatts by 2028 [55][24]. That is one company, two projects, nearly 5 gigawatts — a meaningful slice of a real grid, and a sign that some of the queue is genuine.

For anyone with an electricity bill, this is the quiet story behind rising costs. Whether your rates climb depends a lot on whether regulators can sort the real projects from the phantom ones — and on whether the cost of all that new wire and steel lands on the data centers that need it, or gets spread across everyone else [30].

02 · Lesson · why it matters

When everyone grabs a spot in line, the line stops meaning anything

A shared resource breaks not when people are greedy, but when each person does the sensible thing — and the sensible things add up wrong.

A waiting list five times too big

Texas has a power grid that peaks around 85 gigawatts on its hottest day. It also has a waiting list of 438 gigawatts of new customers — mostly AI data centers — asking to plug in. The list is five times the size of the grid.

Read that again, because the strangeness is the point. The waiting list isn’t reporting how much power people will use. It’s reporting something else entirely. Most of those projects will never get built. So what is the queue actually measuring?

It’s measuring fear of being last.

The sensible move that wrecks the line

Put yourself in a data center developer’s chair. You need power, and power takes years to connect. Whoever gets to the front of the line gets it first. So what do you do?

You file early. You file for more than you might need. You file in three places at once, so you’re covered no matter which one comes through. None of this is cheating. Every step is the careful, responsible thing for you to do.

Now imagine everyone in the room making the same careful calculation. The line fills with placeholders. The grid operator stares at 438 gigawatts and can’t tell which requests are real buildings and which are someone hedging their bets. The queue, which exists to organize who gets power when, has stopped being able to do that one job.

Nobody set out to break it. Each person just took out a little insurance. The insurance, added up across everyone, is the disaster.

This is the oldest trap in the book

A shared pasture where everyone grazes one more cow. A fishery where every boat takes one more haul. A road everyone drives onto at once. In each case the individual choice is reasonable and the collective outcome is ruin — the pasture stripped, the fish gone, the road jammed. The thing that’s shared gets used up by people who were each only doing what made sense.

The grid queue is the same shape. The queue is the commons. A spot in line is the grass. And the reason it’s so hard to fix is that you can’t blame anyone. There’s no villain over-grazing. There’s just a crowd of sensible people, and a structure that turns a crowd of sensible choices into one bad result.

You’ve stood in this exact spot. You’ve held a restaurant reservation you weren’t sure you’d use, kept a meeting on the calendar as a backup, applied to the school you’d probably skip. Holding the option felt free to you. It usually is — to you. The cost lands on the system, spread thin across everyone who couldn’t get the table, the slot, the place.

Why “just ask them to be reasonable” never works

Here’s the part that catches people. The obvious fix — telling developers to only request what they truly need — does nothing. As long as hoarding a spot is free and being last is costly, the smart play is still to hoard. Asking people to act against their own interest, for the good of a system they’re competing in, is asking them to volunteer to lose.

That’s why the regulators didn’t issue a plea. They changed the price of the spot. FERC said it will favor projects that are “real, financeable, operationally flexible.” Texas opened a path that connects you faster if you agree to let the grid switch your power down when it’s tight. The point isn’t the specific rule. It’s the move underneath: make holding a spot cost something — a real commitment, money on the line, a willingness to flex — so the line goes back to meaning what it’s supposed to mean.

You don’t fix this kind of problem by improving the people. You fix it by changing what the structure rewards.

The reader is in the queue too

It’s tempting to read this as a story about distant tech giants and grid bureaucrats. But the cost of that swollen line doesn’t stay with them. Building the wires and substations to serve all those requests is expensive, and FERC’s stated worry is “cost shifts” — the bill landing on ordinary ratepayers for infrastructure built to serve projects that evaporate. The reason electricity rates are climbing in several states traces partly back to a waiting list full of ghosts.

And the same shape sits inside your own life, smaller but identical. Every backup option you hold open, every just-in-case you keep alive, is nearly free to you and quietly costly to whatever pool it draws from. Most of the time that’s fine. The trouble starts when everyone reaches for the same insurance at once, and the shared thing — a grid, a road, a fishery, a calendar everyone’s trying to book — stops being able to do its job.

Seeing that doesn’t tell you what to do. It just makes it harder to believe that a broken shared system is the fault of bad actors. Usually it’s the opposite: a crowd of good ones, each protecting themselves, in a structure that quietly turns self-protection into collective harm.

03 · Lab · your turn

The Queue

Rehearse filing for a shared grid spot under free vs priced rules, and feel why a queue jams from sensible choices — and why pricing the spot, not scolding the people, is what fixes it.

04 · Hope · carry this

When a shared thing breaks under the weight of everyone protecting themselves, it isn't proof that people are bad — only that the rules need redrawing. And rules are the one part we have always been able to change.