America’s biggest power grid operator is being remade by artificial intelligence, not through software, but through steel, concrete, and megawatts. A wave of new data centers built to feed AI models is straining the system that keeps the lights on for tens of millions of people, forcing hard choices about prices, reliability, and where new power plants get built. The crunch is arriving faster than regulators and utilities expected, and the fixes are far messier than the sleek marketing around AI would suggest.
The heart of the problem is simple: AI data centers consume staggering amounts of electricity, and they are clustering in a handful of regions that already sit at the edge of grid capacity. Nowhere is that more visible than in the territory of PJM Interconnection, which coordinates power for more than 65 million customers across a swath of the eastern United States. What began as a tech boom is now a full‑blown planning crisis for PJM Interconnection and the communities that depend on it.
The AI land rush meets an aging grid
The current surge in AI computing is colliding with a grid that was never designed for dense clusters of hyperscale facilities. Data centers that train and run large AI models draw so much power that, as CNBC noted, their energy needs are so great they are reshaping regional planning assumptions. For PJM, which already manages a complex mix of aging coal plants, nuclear units, and growing renewables, the sudden spike in demand from AI facilities has pushed reserve margins to levels that internal planners say have never been this short. Instead of gradually rising consumption spread across households and factories, PJM is confronting blocky, multi‑gigawatt loads that appear almost overnight when a tech company decides to build a new campus.
Those decisions are not random. Companies are clustering AI infrastructure near existing fiber routes and tax‑favored industrial parks, especially in parts of Virginia, Ohio, and Pennsylvania. In northern Virginia, the concentration of server farms has earned the nickname “Data Center Alley,” and PJM’s own capacity auctions now treat that area as a special case because of its voracious appetite for electricity. The result is a grid that must be reinforced at enormous cost to serve a relatively small number of corporate customers, while the risks of outages and price spikes are socialized across the 65 million people who rely on PJM’s network.
Prices spike as capacity gets scarce
One of the clearest signals that AI demand is biting into the grid is the behavior of PJM’s capacity market, the mechanism that pays generators to be available in future years. In the latest auction, prices to supply electricity in PJM’s footprint jumped to record levels, with the steepest increases in zones that include Virginia’s “Data Center Alley.” Analysts tie that surge directly to the projected growth in AI‑driven data center load, which forces PJM to procure more capacity and to value reliability more highly in constrained pockets of the grid. For generators, this is a windfall. For consumers, it is a looming bill.
Independent research has warned that the proliferation of data centers is driving up electric rates for households and small businesses, especially in the PJM region. One analysis of PJM’s capacity auctions found that projected data center growth helped push capacity prices up by a factor of 10 in some areas, with rate increases that customers will ultimately see on their utility bills. The same study highlighted how transmission upgrades to bring power to northern Virginia are being justified largely by data center demand, yet the costs are spread across a much broader customer base. In effect, the AI boom is being underwritten by people who may never set foot inside a server hall.
PJM’s scramble to rewrite the rulebook
Faced with this surge, PJM is trying to update its planning models and market rules on the fly. Earlier this year, the grid operator signaled that it would “ratchet down” some of its most aggressive AI demand projections after an internal analysis suggested that earlier forecasts may have overshot near‑term growth. Across the U.S., energy policymakers are wrestling with similar questions, trying to distinguish between marketing hype and realistic build‑out schedules, while also accounting for bottlenecks in transformers, high‑voltage electronics, and specialized construction teams. For PJM, getting those numbers right is not an academic exercise; it determines how many power plants clear its auctions and how much transmission it orders utilities to build.
At the same time, PJM is “Pushing Forward on Efforts to Meet Rising Data Center Load,” as internal planning documents describe it, by exploring new interconnection rules and incentives. In early January, regional stakeholders were briefed on Efforts to streamline how large loads connect to the grid while still meeting reliability requirements. Those efforts include revisiting how PJM studies co‑located resources, such as data centers paired with on‑site generation or battery storage, and how it treats flexible demand that can ramp down during peak hours. The goal is to avoid overbuilding while still ensuring that a sudden spike in AI traffic does not trigger rolling blackouts.
Regulators push back, data centers told to bring their own power
Federal regulators have started to intervene as the scale of the AI build‑out becomes clear. The Federal Energy Regulatory Commission, often referred to as FERC, recently ordered PJM Interconnection to develop new rules for customers that share grid connections with generators, a group that increasingly includes data centers. In a directive summarized as Next Steps, FERC instructed PJM that, By January, it must file compliance plans describing how co‑located customers can access provisional interconnection service and how disputes over priority will be resolved. The aim is to prevent situations where a data center and a solar farm, for example, are effectively competing for the same limited grid capacity without clear rules.
Regulators in Washington have also signaled that they see AI‑driven demand as part of a broader shift in how infrastructure is planned. In a recent Washington update, The Federal Energy Regulatory Commission described how FERC is pressing PJM Interconnection and other regional operators to align their planning with long‑term decarbonization goals while still accommodating new industrial loads. That guidance came after more than 160 public comments, many of them focused on the tension between rapid data center expansion and state climate policies. The message from federal regulators is that AI cannot be treated as a special case exempt from the usual scrutiny around who pays for new wires and plants.
On the ground, some utilities and grid planners are going further, telling data center developers that they may need to supply their own power if they want to keep building at current speed. Reporting on PJM’s territory describes how grid officials have floated the idea that large customers should bring on‑site generation, such as gas turbines or large batteries, to reduce strain on shared infrastructure. In some cases, the message has been even starker, with warnings that facilities could face curtailment or delays if they do not invest in their own capacity. A separate account framed the choice bluntly, describing how some projects have been told to Bring Own Power or Shut Down if they cannot secure firm grid service. That kind of ultimatum would have been unthinkable a decade ago, when utilities were desperate for any new load.
Lessons from past tech booms and the road ahead
There is a historical echo here that energy veterans are quick to point out. In the late 1990s, the first wave of Internet infrastructure led utilities and tech companies to wildly overestimate how much power online services would need, only for demand to fall short and some plants to become stranded assets. A recent review of power sector trends noted that, at that time, the tech and energy industries assumed electricity use would soar, only to see it shrink by 50% or more in some segments as efficiency improved. The lesson is not that AI demand will evaporate, but that straight‑line extrapolations are risky. If planners overbuild again, households could be stuck paying for underused gas plants and transmission lines long after the current AI wave has crested.
Yet the scale and concentration of today’s AI data centers make this boom different from the early Internet. Facilities run by companies like Amazon Web Services are effectively small power systems in their own right, with a bottomless appetite for electricity and a preference for specific locations. That gives them enormous leverage over regional grids like PJM, but it also means they cannot be treated as just another industrial customer. As AI continues to spread into everything from chatbots to self‑driving trucks, the real test for PJM Interconnection and its regulators will be whether they can channel that demand into investments that strengthen the grid for everyone, rather than locking the country into a more fragile and more expensive energy future.
Global energy markets and the AI feedback loop
The AI data center boom is not unfolding in isolation from the rest of the energy system. It is intersecting with global fuel markets, refinery operations, and cross‑border trade in ways that complicate planning for PJM and its peers. One recent account of North American energy flows, for example, noted how shifts in Mexico’s oil sector, with Mexico seeing Crude Exports Slide as Refining Finally Reawakens, can ripple into U.S. fuel prices and generator economics. When refineries adjust output or export patterns, the cost of running gas and oil‑fired plants that backstop data centers can change quickly, feeding back into capacity prices and long‑term contracts.
At the same time, the broader push for sustainable infrastructure is reshaping how capital flows into grid upgrades that AI facilities depend on. Policy briefings in Washington have emphasized that any large‑scale reinforcement of PJM’s network must align with climate and resilience goals, not just the needs of AI clusters. In that context, the AI build‑out becomes part of a larger debate about whether new transmission lines should prioritize connecting wind and solar resources in the Midwest, offshore projects in the Atlantic, or high‑density loads in suburban corridors. The choices PJM Interconnection makes over the next few years will determine whether AI accelerates the transition to cleaner, more flexible power, or locks the region into a cycle of emergency fixes and rising bills.
More from Morning Overview