The price hits the ceiling and supply still falls short
Across thirteen states and millions of homes, the infrastructure that has long kept American lights on is encountering a demand it was never designed to meet. This summer, the PJM Interconnection—the nation's largest grid operator—held a capacity auction that hit its regulatory price ceiling and still could not secure enough reserves, a quiet but consequential signal that the electrical grid is straining beneath the weight of artificial intelligence. Data centers, running without pause and multiplying without precedent, are drawing power at a pace that outstrips the slow, deliberate rhythms by which power plants are built and grids are expanded. What is unfolding is not merely a technical shortfall but a collision between two very different speeds: the urgency of technological ambition and the patience required to sustain the commons.
- PJM's capacity auction hit its regulatory price ceiling and still failed to secure adequate power reserves—a rare and alarming outcome for the grid serving over 65 million people.
- AI data centers, running continuously and proliferating rapidly, are adding billions in electricity costs across thirteen states, with Ohio prices climbing to near-record highs.
- The reliability margin—the buffer of spare capacity that prevents blackouts—has thinned dangerously, raising the real possibility of rolling outages during peak demand or equipment failures.
- Grid operators are racing to finalize new regulations that would require data centers to curtail usage on short notice, install on-site storage, and fund transmission upgrades.
- The structural mismatch remains unresolved: data centers come online in months, while new power plants and transmission lines take years or decades to build and permit.
The PJM Interconnection, which manages electricity for more than 65 million Americans across the Mid-Atlantic and beyond, failed this summer to secure enough power reserves at auction. Bidding hit the regulatory price ceiling and still came up short—a signal that the mechanisms designed to keep the grid stable are beginning to buckle under a new and unfamiliar pressure.
That pressure is artificial intelligence. Data centers built to train and run large language models are not like ordinary industrial facilities. They operate around the clock, drawing enormous and unrelenting quantities of electricity, and they are being built at speed. Across thirteen states, AI infrastructure is adding billions of dollars in power costs. In Ohio, electricity prices have reached levels not seen in years, and analysts see no near-term relief.
The deeper problem is structural. Tech companies can break ground on a massive data center and have it operational within months. A new power plant takes a decade or more to permit and build. Even renewable projects face lengthy timelines. The grid's reliability margin—the cushion of spare capacity that prevents blackouts—has shrunk, and when it thins too far, the cost of preventing outages rises sharply for consumers and businesses far removed from any data center.
Grid operators are responding. PJM is moving to finalize regulations that would require new data centers to demonstrate they can reduce consumption on short notice, include on-site generation or battery storage, and fund upgrades to the transmission infrastructure their load demands. Whether regulation can close the gap remains uncertain. Until the pace of power generation catches up with the pace of construction, auctions where the price hits the ceiling and supply still falls short may become a recurring feature of American energy life.
The largest power grid operator in the United States failed to secure enough electricity reserves at auction this summer, a shortfall that underscores a widening gap between what the nation's electrical infrastructure can deliver and what the market now demands. The PJM Interconnection, which manages power across the Mid-Atlantic and into neighboring regions including Maryland, held an auction designed to lock in sufficient supply for future reliability. The bidding hit the regulatory price ceiling and still came up short—a signal that the traditional mechanisms for ensuring grid stability are beginning to strain under new pressure.
That pressure has a name: artificial intelligence. Data centers built to train and run large language models consume electricity at scales that have caught grid operators off guard. These facilities are not like factories or office parks. They run continuously, drawing power around the clock, and they are proliferating. Across thirteen states, the buildout of AI infrastructure is adding billions of dollars in power costs to the grid. In Ohio, electricity prices have climbed to levels not seen in years, and analysts expect them to remain elevated as demand continues to climb.
The PJM auction failure is not merely a technical problem. It reflects a structural mismatch between the speed at which data centers are being built and the speed at which power plants can be constructed or existing capacity can be expanded. Traditional generation—coal, nuclear, natural gas—takes years to permit and build. Even renewable energy projects face lengthy development timelines. Meanwhile, tech companies are signing long-term power purchase agreements and breaking ground on massive facilities with urgency that leaves grid planners scrambling.
The consequences ripple outward. Electricity prices spike not just in the immediate region where a data center is built, but across interconnected grids as operators compete for available supply. Consumers and businesses in states far from any data center find their power bills climbing. The grid's reliability margin—the cushion of extra capacity that keeps the lights on during peak demand or equipment failures—has shrunk. When that margin gets too thin, rolling blackouts become possible, and the cost of preventing them rises.
Grid operators are not passive observers. The PJM operator, which manages power for over 65 million people across multiple states, is moving to finalize new regulations specifically designed to govern how data centers connect to and draw from the grid. These rules will likely require data centers to demonstrate they can reduce their power consumption on short notice, to help stabilize the grid during emergencies. Some proposals would mandate that new facilities include on-site power generation or battery storage. Others would require developers to fund upgrades to transmission lines and substations that will be needed to handle the load.
What remains unclear is whether regulation alone can bridge the gap. The fundamental problem is one of scale and speed. The electricity demand from AI data centers is growing faster than the grid's capacity to supply it. Building new power plants takes a decade or more. Upgrading transmission infrastructure takes years. Data centers can be operational in months. Until the pace of power generation catches up with the pace of data center construction, grid operators will continue to face auctions where the price hits the ceiling and supply still falls short. The Mid-Atlantic grid's failure this summer may be the first of many such moments.
Notable Quotes
Grid operators are finalizing new regulations to govern how data centers connect to and draw from the grid, potentially requiring on-site power generation or battery storage— PJM Interconnection regulatory framework
The Hearth Conversation Another angle on the story
Why does it matter that one auction hit a price limit? Isn't that just how markets work?
Because hitting the price ceiling while still failing to secure enough supply means the market is broken. The price is supposed to rise until supply and demand balance. If the price maxes out and you still don't have enough power, it means there literally isn't enough to buy at any price.
So what happens then?
The grid becomes fragile. You lose your reliability margin. In a heat wave or cold snap, when demand spikes, there's nowhere to go. That's when blackouts happen.
And the data centers are the new demand?
They're the new demand arriving faster than anything the grid has seen before. A single facility can draw as much power as a city. And they're being built everywhere at once.
Can't they just build more power plants?
They're trying. But a natural gas plant takes five to seven years to build. A data center takes months. The math doesn't work.
So the regulations—do they actually solve this?
They buy time. They force data centers to be smarter about when and how much they use. But they don't create new power. Until someone builds generation capacity faster, we're just managing scarcity.