AI's Massive Power Grab: The PJM Grid Crisis Explained

Is it now AI's world, and we just live—and pay for electricity—in it? The AI revolution requires a staggering amount of electricity, and our power grid is struggling to keep up with the astonishingly rapid growth of data centers.

In this video, we dive into the unprecedented crisis facing grid operators like PJM Interconnection. With some regions seeing data-related energy loads jump from just 600 megawatts to 11,000 megawatts, grid operators are facing challenges never seen before in power markets.

Key topics covered in this video:

The Data Center Explosion: How utilities at ground zero, like Dominion in Virginia, are fielding up to 70,000 megawatts of large load interconnection requests—nearly equal to or exceeding the load they currently serve.

PJM’s "Hail Mary" Solution: A breakdown of PJM’s proposed parallel auction, a one-off bilateral contracting process aiming to secure 14,900 megawatts of new capacity by matching large loads with supply owners.

Supply Chain Roadblocks: Why finding enough equipment is a major hurdle, with grid-scale gas turbines essentially sold out globally through 2029.

A Broken Market: How the massive wealth of tech giants is warping energy economics. With data centers valuing early grid connection at up to $7 billion per gigawatt, capital is practically guaranteed to abandon the standard, price-capped public power markets in favor of lucrative, uncapped bilateral tech contracts.

As energy developers chase higher returns to power the AI boom, existing resources are being locked out, and everyday consumers might face the collateral impacts. Watch to understand why fixing the grid for AI is a nearly impossible puzzle.

Transcript

The Data Center Load Explosion

PJM’s Parallel Capacity Procurement

Turbines Are Sold Out

Price Caps And Capital Flight

Collateral Damage And Closing Thought

SPEAKER_00 0:03

Some would say this is now AI's world, and we only live in it. The implications of that development for our lives, jobs, our relationships to each other, what it means to be human, well, those remain to be revealed. In the meantime, this next phase of our technological development needs to be fed with electricity, a challenge which our grid is having trouble meeting. On Friday, April 11th, PGM took yet another step to address the challenge of meeting the astonishingly rapid growth of large, principally data-related loads. It'd be easy to call out PGM, as some have, and say they've mismanaged the process of dealing with these new data centers, but a prudent review of the situation on the ground might suggest that they're dealing the best they can with a situation that nobody has ever seen in power markets anywhere before. Why do I say that? Well, let's just look at the magnitude of the loads that have mushroomed up recently. In May of 2024, for example, AP Ohio served 600 megawatts of data-related load. Today, it has 11,000 megawatts either under energy service agreements or committed to pay for the required infrastructure to serve that load. Dominion, the utility at Ground Zero in Virginia, recently filed for approval of its large load interconnection queue process standards and noted in its filing that about 25,000 megawatts of large load requests have already been assigned projected connection dates between now and December 31st of 2031. That's about equal to the load they serve today, and that's pretty big. Oh yeah, the Utility also noted that as of December 31st of last year, the company's looking at an additional approximately 45,000 megawatts of large load interconnection requests under its Q process. So not surprisingly, PGM has a huge and nearly impossible challenge in front of it. That was demonstrated in its most recent three capacity auctions, the last one of which came up short in securing the capacity required to meet its desired reserve margins. You know, the extra supply that needs to be available in case we lose a critical power plant or transmission line. In response to the dilemma, PGM is proposing to hold a one-off procurement process for 14,900 megawatts of new capacity to serve these large loads. With the assistance of consulting firm Charles River Associates, CRA, PGM would facilitate a bilateral contracting process between large loads and supply owners, including demand response providers. The planned process will run from September through next March. If the 14,900 megawatt goal is not met, then PGM would run a second phase in March. The winning suppliers would then be paid through a contract for differences mechanism. Under that approach, they'd get revenue from PGM's normal base capacity auctions, and they'd either be paid more or have to pay back monies to reach their contracted strike price. Simple, right? Resources that are allowed to play include new projects, capacity upgrades on existing plants, repower generators that have been reactivated, as well as new demand response and distributed energy resources. The resource must demonstrate an ability to operate by June 1st of 2031, and delayed power plant retirements will be unable to participate. PGM and CRA will serve as confidential intermediaries and provide the matchmaking services for the load and new supplies. The contracting parties would then set terms and conditions and contract out of PGM's purview, and parties could consume bilateral transactions on their own. Sounds a little like Frankenstein to me, or at least a new kind of Hail Mary project. The genesis of the parallel auction concept came out of the January White House meeting with the 13 PGM states, a meeting to which the grid operator PGM and data center companies were not invited. Here's my concern with this approach. We all know that turbine supplies are limited, as is other energy supply infrastructure such as transformers and switchgear. While the big three, G or Vernova, Siemens, and Mitsubishi have all announced manufacturing expansion plans, the grid-scale turbines are basically sold out through 2029. They can't build those new factories quickly. Global Energy Monitor recently reported in its global oil and gas plant tracker that U.S. planned turbine capacity in development, that's announced in pre-construction or construction phase, jumped from 85 gigawatts to 252 gigawatts from 2024 to 2025. Of those numbers, on-site capacity, so co-located or never connected to the grid, that number catapulted from next to nothing in 2024 to just under 100 gigawatts in 2025. So where then will these new turbines come from? They're wanted in Texas, Arizona, Ohio, and all over the world for that matter. So we need to assume some finite amount of equipment supply. Then there's the interconnection queue that has to be overhauled for the parallel auction, even as it's cleaned up for the capacity bidding that occurs in the base residual auction, the normal BRA. Then there's the price cap for the next two PGM BRAs at around$325 per megawatt day. If I'm a developer of a new resource here and I'm only going to build on PGM, let's assume, I have two choices. First, I could bid into the BRA and accept my capped capacity revenue with price certainty for only a single year. Or I could bid that new resource into a bilateral auction and fix my uncapped price for two to as long as 15 years under this new approach. Hmm, which would I choose? Before we answer that question, though, let's add one more data point. A recent estimate from the CEO of Kamu Energy, whose job is to interconnect data centers to the grid, suggested that at the moment, the value of energizing one gigawatt of data central load a year earlier, a single year, is$7 billion. That's over$19 million a day. So let's go back to that question. I have one market that's capped, another with customers likely willing to pay enormous sums to get into the game. So if I'm a developer, where am I going to put my capital? It's kind of like the so-called billionaire's tax. If I mess too much with the incentive structure, global capital gravitates quickly and easily towards higher returns somewhere else. The likely outcome is that little to no capital gets devoted to serving the existing PGM capacity markets. And by the way, if I'm an existing resource, I'm not happy at being locked out of this new and lucrative bilateral opportunity within the same ISO. I don't fault PGM, it's kind of an impossible situation. However, I'm not sure this approach is going to be pulled off very easily without collateral impacts to other parties in PGM. As I said at the outset, though, it might be that it's now AI's world. We only live in and consume and pay for electricity in it. Well, thanks for watching, and we'll see you again soon.