An Exelon executive stood in front of investors and explained why her pipeline shrank. “Hey, our pipeline went down from 18 to 11,” Jeanne Jones said. Investors wanted to know why. She told them it was the smart thing to do. Jennifer Hiller's story in Wednesday's Journal explains why a utility would be glad to see seven gigawatts of requests drop off its list.
The AI boom is driving a once-in-a-generation surge in electricity demand. That should be a great time to be a utility. The problem is telling real projects from wishes. A developer shopping for a site asks several utilities for power at once. So one project can be counted more than once, sometimes several times over. Wood Mackenzie puts only 23% of the disclosed requests in its high-confidence bucket. Everything else is a maybe. The industry's nickname for a maybe is a phantom data center.
The requests got huge fast. Jones says a normal request used to be around 50 megawatts, which would run about 50 Walmart stores. Now they run to 1,000 megawatts or more, roughly what a large nuclear reactor puts out. At Exelon, large-customer requests tripled to 18,000 megawatts. Then the utility changed the rules. Before it would do detailed engineering, a developer had to sign a binding commitment and put up collateral. The high-probability list shrank to 11,000 megawatts. Those projects have posted about $1 billion with the company. Jones put the stakes plainly. If the grid gets built and the data center walks away, “someone's got to pay for it and that's not fair to them.”
Who pays for the phantom
That someone is you. The cost of new wires and substations gets spread across every customer on the system. If the data center never opens, the rest of the customers carry what was built for it. Rising power bills from the AI build-out are already a political fight. Utilities are answering with higher rates for data centers, multi-year minimum contracts, early-exit fees and paid studies. Minimum contracts for big customers now run about 10.8 years, up from about 4.6, per Halcyon. Texas, the busiest market, has about 12,800 megawatts under way and a queue of more than 400,000. It now charges a $50,000-a-megawatt study deposit. A 500-megawatt center puts up $25 million before the study starts.
Vistra (VST) CEO Jim Burke told investors he expects about 12,000 to 15,000 megawatts to connect in Texas by 2030. Requests, he said, are “overstated by more than 20x,” and he supports thinning the queues. We wrote about the backlash on Oct. 1 and Oct. 7, the town letters and the moratorium. What's new is that the utilities are now the skeptics, and they're asking for cash up front.
Our read
Two domains here. Cash Flow · M5, because this is your electric bill. And Investments/Risk · IN04, because the power names in the books are priced partly on that queue. Vistra is held at weight in 15 books, Constellation (CEG) in 9 and GE Vernova (GEV) in 30; nothing added. The collateral rules are good news on both sides of that. Ratepayers stop funding ghosts. Holders get a demand number backed by deposits instead of a wish list. Burke's 12,000 to 15,000 megawatts is a smaller number than the headlines, and a far more solid one to plan on.
Here's your check this week, and it's on last month's electric bill. Find the delivery or transmission charges, the part that isn't the power itself. That's generally the line that grows when a grid gets built out, and it's the line a phantom data center would've landed on. Know what it is now, so you can see it move. If you're a retiree on a fixed draw, plan for that line as one that can climb, not a flat one. Then see whether a utility sits in your dividend fund. The same rules that protect your bill also keep the utility from building for a customer that never shows, which is part of what keeps its dividend steady. If that line has jumped since last year, bring the bill to your next review. Clear the gutters before the storm. Nobody enjoys doing it during one.