AI data centers are hitting a wall, and the grid may not have the power to feed them

The computing boom behind generative artificial intelligence runs on electricity, and the sprawling facilities that house it are beginning to strain the power systems they depend on. Across the United States and much of the world, the pace at which new data centers are being proposed now outstrips the pace at which utilities can build the generation and transmission lines to serve them. The result is a widening gap between what the technology industry wants to plug in and what the grid can actually deliver.

That gap is not a distant worry. Grid planners, national energy agencies and utilities keep pointing to the same near-term squeeze: demand forecasts have jumped sharply, the queues to connect new projects are clogged, and the heavy equipment needed to expand the network is back-ordered for years. The mismatch is already reshaping where AI computing can be built and how quickly it can come online.

How fast data-center demand is climbing

The scale of the increase is what makes it hard for the grid to absorb. The International Energy Agency estimates that data centers consumed roughly 415 terawatt hours of electricity in 2024, about 1.5 percent of the global total, and projects that figure to roughly double to around 945 terawatt hours by 2030. In that base case, data-center consumption grows about 15 percent a year, more than four times faster than electricity demand from every other sector combined, with servers running AI workloads accounting for close to half of the net increase.

Those numbers could climb much higher if AI adoption accelerates. The agency’s more aggressive scenario puts global data-center demand above 1,700 terawatt hours by 2035, or roughly 4.4 percent of world electricity use. The absolute share still looks modest, but the problem is concentration: unlike electric vehicles or air conditioning, data centers cluster in a handful of locations, and a single campus can request as much power as a small city.

Where the grid is falling behind

In North America, the reliability picture has darkened as forecasts have been revised upward. The North American Electric Reliability Corporation, the body that sets and enforces grid standards, warned in its most recent long-term assessment that reliability conditions are deteriorating because demand is now growing faster than new firm generation and transmission can be added. Data centers, alongside new manufacturing and broader electrification, are the primary reason the outlook shifted so quickly.

The revision was steep. NERC’s 2025 assessment shows summer peak demand across the continent forecast to grow by 224 gigawatts over the next decade, more than 69 percent higher than the prior year’s forecast, with data centers cited as the largest single driver. Several regions, including the Midcontinent system operator, PJM, the Texas grid and parts of the Pacific Northwest, face an elevated risk of falling short of reserve margins within five years if planned resources arrive late or not at all.

The physical bottlenecks slowing everything down

Even where power exists on paper, moving it to a new data center is the hard part. Connecting a large load requires new substations, high-voltage lines and large transformers, and the wait times for that equipment have stretched into multiple years as manufacturers struggle to keep up. Interconnection queues, the lists of projects waiting for permission to plug in, have swollen with both generation and demand requests, leaving developers stuck in line even after a site is chosen.

Geography compounds the strain. Federal energy data show that data-center growth is heavily concentrated in a few corridors, with northern Virginia hosting one of the densest clusters on the planet and other hot spots forming around Texas, the Midwest and the Pacific Northwest, according to figures published by the U.S. Energy Information Administration in its ongoing energy analysis. When dozens of gigawatt-scale requests land on the same regional grid at once, local transmission becomes the binding constraint long before national generation does.

What the power crunch is forcing operators to do

The squeeze is changing how the largest technology companies approach electricity. Rather than wait years for a grid connection, some operators are building generation on site or nearby, striking long-term deals for gas turbines, and signing agreements to underwrite nuclear plants and next-generation small modular reactors that do not yet exist at commercial scale. Others are exploring behind-the-meter arrangements and batteries so a campus can run partly independent of the public grid.

Grid operators, in turn, are pressing data centers to become more flexible, curtailing or shifting their draw during the tightest hours in exchange for faster connection. Whether those measures close the gap or merely slow its growth remains an open question, and it is one that will shape both the cost of electricity for ordinary customers and the geography of where the next wave of AI infrastructure can actually be built.

The costs that land on everyone else

The scramble to power AI does not stay confined to the technology sector. Large new loads can push up the cost of maintaining and expanding the grid, and in many regions those costs are spread across the broader base of ratepayers, raising the prospect that ordinary households and small businesses help underwrite the infrastructure a data center requires. Regulators in several states have begun crafting special large-load tariffs and interconnection rules meant to ensure that the biggest customers cover more of the expense they create, rather than shifting it onto everyone else.

There is an environmental dimension as well. Because firm, around-the-clock power is easiest to secure from natural gas, the rush to serve data centers has extended the life of fossil generation and, in some cases, prompted new gas plants, complicating utilities’ pledges to cut emissions. Cooling the servers also consumes large volumes of water in some designs, adding strain in drought-prone areas. The combination has produced local pushback, with some communities imposing moratoriums or tighter siting rules on new campuses while they weigh the trade-offs between economic investment and pressure on power, water and land.

This article was produced with AI assistance and reviewed by Morning Overview editors.

More from Morning Overview