The largest power grid in the United States is running out of room, and the federal government just put a clock on fixing it.
In late 2025, the Federal Energy Regulatory Commission ordered PJM Interconnection to overhaul its rules for connecting massive new electricity loads to the grid. PJM operates the power network that serves roughly 65 million people across 13 states and the District of Columbia, stretching from New Jersey to Illinois. The directive required PJM to file a detailed informational report by January 19, 2026, covering reliability risks, backstop procurement plans, and updated demand forecasts. That filing landed earlier this year, and the picture it paints is sobering: reserve margins across PJM’s territory are tightening faster than at any point in the grid operator’s two-decade history, driven overwhelmingly by data centers built to run artificial intelligence workloads.
The demand spike is already showing up in prices
PJM’s own capacity auction results tell the story in dollars. The 2025/2026 Base Residual Auction, which cleared in mid-2024, produced prices that jumped more than 800% in some delivery zones compared to the prior year. That auction is the mechanism through which PJM secures commitments from power plants to be available when demand peaks. When clearing prices spike that sharply, it signals that the margin between available generation and expected load is shrinking, and those costs flow directly into the bills of residential and commercial customers.
The driver is not a mystery. Northern Virginia’s “Data Center Alley,” located squarely within PJM’s footprint, is already the largest concentration of data centers on the planet. Loudoun County alone hosts more data center capacity than most countries. And the buildout is accelerating: hyperscale operators including Amazon, Microsoft, and Google have announced billions of dollars in new campus expansions across PJM territory in Virginia, Ohio, and Indiana, with much of that capacity dedicated to AI training and inference.
PJM’s interconnection queue reflects the pressure. As of early 2026, the queue contained more than 260 gigawatts of proposed new generation and storage projects, but only a fraction of those will clear the study process and reach commercial operation within the next three to five years. Meanwhile, new large-load interconnection requests from data center operators have surged, with individual facilities sometimes requesting 300 megawatts or more, enough to power a small city.
What FERC ordered and why it matters now
FERC’s directive was not a suggestion. It is a binding federal order that requires PJM to develop new interconnection and capacity market rules specifically designed for large, fast-growing loads. The commission told PJM to address how it will protect existing ratepayers from bearing disproportionate costs when a single customer adds hundreds of megawatts of demand in a compressed timeline.
The January 2026 informational report that PJM filed in response covers four key areas: reliability concerns tied to accelerating load growth, the status of expedited generation and transmission additions, backstop procurement plans for acquiring capacity outside normal auctions when reliability is at risk, and revised load forecasts that account for the AI-driven demand surge.
Backstop procurement is one of the most contentious tools in that list. It allows PJM to contract for additional capacity when market auctions do not secure enough resources, but the costs are socialized across all ratepayers. FERC has not publicly defined a threshold for when backstop measures become necessary or how those costs would be allocated, leaving a significant policy gap that PJM and stakeholders are still working to resolve as of June 2026.
Federal data confirms the scale of the problem
PJM is not operating in an information vacuum. The U.S. Department of Energy published a national assessment of data center electricity consumption that projects demand could more than double by 2028 under aggressive growth scenarios. That report draws on underlying research from Lawrence Berkeley National Laboratory, which modeled multiple scenarios based on AI chip efficiency trends, hyperscaler construction timelines, and regional power availability.
Under LBNL’s high-growth scenario, U.S. data centers could consume more than 300 terawatt-hours annually by 2028, up from roughly 175 terawatt-hours in 2023. Even the moderate scenario projects consumption rising to approximately 230 terawatt-hours. For context, 300 terawatt-hours is roughly equivalent to the total electricity consumption of a country the size of the United Kingdom.
These are national figures, not PJM-specific projections. But PJM’s territory hosts the densest cluster of data center activity in the country, which means it will absorb a disproportionate share of that growth. PJM’s own internal forecasts, portions of which have been shared in stakeholder meetings, suggest that data center load within its footprint could grow by 10 to 15 gigawatts over the next five years, a figure that would require the equivalent of roughly 10 large natural gas power plants or a comparable mix of renewables and storage.
What consumers and businesses should watch
For the 65 million people who get their power through PJM, the most immediate concern is cost. Capacity auction prices have already risen sharply, and those increases will begin appearing in electricity bills over the next 12 to 18 months as the 2025/2026 delivery year takes effect. If subsequent auctions clear at similarly elevated levels, residential customers in affected zones could see meaningful increases in the capacity portion of their bills, which typically accounts for roughly 10% to 15% of total electricity costs.
Reliability is the deeper worry. PJM’s reserve margin, the buffer between peak demand and available generation, has been narrowing. The grid operator has publicly acknowledged that without new generation coming online faster than current timelines allow, parts of its territory could face elevated risk of rolling blackouts or emergency load-shedding events during extreme weather by the late 2020s. That risk is most acute during summer heat waves and winter cold snaps, when demand spikes and older generating units are most likely to experience forced outages.
State-level permitting decisions will play a major role in determining whether new supply arrives in time. Transmission line approvals, gas plant siting, and renewable energy permitting all move through state and local processes that can add years to project timelines. Virginia, which hosts the bulk of PJM’s data center load, has been working to streamline permitting for both generation and transmission, but neighboring states have been slower to act. If permitting bottlenecks persist, the gap between demand and supply could widen faster than PJM’s planning models currently assume.
Demand-side tools offer a partial counterweight. FERC’s order requires PJM to consider demand response, energy efficiency, and load flexibility programs as part of its restructuring. Some large data center operators have signaled willingness to curtail usage during grid emergencies or invest in on-site generation, such as natural gas turbines or fuel cells, that could reduce their draw from the shared network. But voluntary commitments from a handful of operators will not close a multi-gigawatt gap, and mandatory curtailment provisions for large loads remain politically and commercially fraught.
A grid built for a slower world
PJM’s planning framework was designed for an era when electricity demand grew at 1% to 2% per year, driven by population shifts and gradual economic expansion. The AI data center boom has shattered that assumption. In some parts of PJM’s territory, load growth projections now exceed 5% annually, a pace not seen since the post-World War II electrification era.
The grid operator’s leadership has used increasingly urgent language in public forums to describe the mismatch. PJM officials have told stakeholders that the window for restructuring the grid’s planning and market rules is measured in years, not decades, a framing that reflects both the speed of data center construction and the long lead times required to build new power plants and transmission lines. A gas plant permitted today might not deliver its first electrons for four to six years. A major transmission upgrade can take a decade.
That timeline mismatch is the core of the problem. Data centers can be designed, permitted, and energized in 18 to 24 months. The generation and transmission infrastructure needed to serve them reliably takes far longer. Unless PJM and its state regulators find ways to compress those timelines, or unless data center operators accept constraints on how quickly they can draw power from the grid, the gap will continue to grow.
FERC’s order gives PJM a framework and a set of deadlines. The January 2026 report was the first major milestone. Subsequent filings and rule changes are expected throughout 2026 and into 2027. For consumers, the most important thing to track is whether PJM’s revised load forecasts match reality, whether new generation projects clear permitting and construction hurdles on schedule, and whether capacity auction prices stabilize or continue to climb.
The answers to those questions will determine whether the nation’s largest power grid can absorb the AI era’s appetite for electricity without forcing the 65 million people it serves to pay sharply higher bills or face the kind of reliability failures that were once considered unthinkable in the wealthiest power market in the world.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
