Unless you’ve had your head in the sand, you’re likely aware that AI has a major energy problem. And as AI companies scramble to source power for their ever-expanding fleet of data centers, the technology is reshaping the US grid.

After more than a decade of flat growth, nationwide electricity demand has been climbing at 1.7 percent annually since 2020, according to the US Energy Information Administration. The agency primarily attributes this increase to the rapid expansion in data centers over that period.

This trend is only likely to accelerate based on an analysis by S&P Global, which estimated that grid demand from these facilities would rise by 22 percent by the end of 2025 and nearly triple by 2030.

Data centers have always been large electricity consumers, but the scale and pace of the AI build-out puts them in a different league. And utility companies bearing the brunt of this shift are being forced to rewire their long-term planning in response to the surge in demand.

Dominion Energy, which services the world’s largest data center market in Virginia, reported that by the end of last year it had signed deals to supply nearly 48.5 gigawatts of power to data centers. This prompted it to raise its five-year capital spending plan nearly 30 percent to $64.7 billion.

CenterPoint Energy, another major utility serving the Houston area, boosted its 10-year capital plan to $65.5 billion in response to the jump in demand. It now expects to hit a 50 percent increase in peak load by 2029, two years ahead of schedule.

The pace of change promises to significantly reshape the US energy mix. In a March forecast, the Energy Information Administration projected that natural gas generation could jump 7.3 percent between 2025 and 2027 if data center demand is on the higher side of estimates. It also predicted that the steady decline in coal generation over recent decades would slow in this scenario.

But in perhaps the most striking shift, tech companies are now bankrolling new capacity themselves. Nuclear power is experiencing a major resurgence as AI providers and data center operators invest in new reactor development and sign long-term deals with existing plants. The activity could grow nuclear capacity 63 percent by 2050.

Meta also recently took the unusual step of privately funding a major expansion of the Louisiana grid to power its new $27 billion Hyperion data center. The facility, due to come online in 2028, could eventually consume over 7 gigawatts—enough to supply several million homes.

To account for its impact on the grid, Meta has agreed to pay for the construction of seven new natural gas power plants by utility Entergy—in addition to three already-approved plants—as well as 240 miles of new transmission lines to connect South Louisiana to North Louisiana and Arkansas and three new battery storage facilities.

The deal is likely a reaction to growing public discontent about the impact data centers are having on energy prices. People are also worried about how the surge in demand will affect long-term grid stability.

PJM Interconnection, the largest power grid operator in the US, warned in February that the country could face supply shortfalls of up to 60 gigawatts in coming decades and strained capacity could lead to blackouts as soon as 2027.

One potential workaround is the possibility of throttling data center workloads, and therefore energy use, when the grid is under stress. Major utilities including AES, Constellation, NextEra Energy, and Vistra are reportedly working on these so-called “flexible AI factories.”

But the idea is still largely experimental, and it’s uncertain whether big tech would willingly commit to regularly downing tools. IT consultant Heunets told Reuters it can cost companies about $9,000 a minute when their data centers go offline.

Given the complexities of meeting all this new demand, pressure is mounting for data center operators to solve their own power problems. Despite taking a generally supportive stance toward the AI boom, President Trump called on tech companies to build their own power plants for data centers in his February State of the Union address.

And it’s already happening. Energy consultant Cleanview says 46 data centers with a combined capacity of 56 gigawatts plan to build dedicated power infrastructure. This trend is giving birth to a “shadow grid”—a parallel energy system that operates alongside public power infrastructure.

This could still have knock-on effects for the rest of us. For a start, due to the difficulty managing the variable output of renewables, most projects rely on natural gas generators, which could lead to a spike in carbon emissions.

And because the most efficient turbines are hard to source on short notice, facilities are using more polluting generators. What’s more, tech companies are now competing with utilities for equipment. This could lead to ballooning costs that are then handed on to consumers.

Altogether, it’s become increasingly clear that the AI boom will fundamentally reshape the US energy system. And the speed at which companies are seeking to deploy new facilities is leaving little room for the work to be done in a considered and sustainable way.