PJM's 6GW Capacity Shortfall: The Grid Crisis That Could Reshape AI Infrastructure

TL;DR

PJM Interconnection's December 2025 capacity auction failed to procure enough power for the first time in history—falling 6,625 MW short of reliability targets for the 2027/28 delivery year. Data centers account for 94% of projected load growth, driving capacity prices to a record $333.44/MW-day (up from $28.92/MW-day in 2024). This shortfall—roughly equivalent to Philadelphia's electricity demand—has triggered political backlash from Senator Bernie Sanders and Governor Ron DeSantis, while forcing Virginia to create a new rate class protecting residential customers from data center cost shifts. The crisis accelerates behind-the-meter generation, demand response programs, and questions fundamental assumptions about AI infrastructure growth.

6,625 megawatts.

That's how far short PJM Interconnection fell in its December 2025 capacity auction—the first time in the grid operator's history that it failed to procure enough supply to meet expected electricity demand. The shortfall represents nearly the entire electricity demand of Philadelphia, and signals what FERC Commissioner Judy Chang called "the grid reliability crisis that has been brewing for years."

PJM serves over 65 million people across 13 states and the District of Columbia. When the nation's largest grid operator can't secure enough power to keep the lights on, the implications extend far beyond regional electricity bills.

For data center operators and AI infrastructure planners, this shortfall represents more than a policy problem. It's a fundamental constraint that will reshape where facilities get built, how they get powered, and whether the current AI buildout trajectory remains viable.

The December 2025 Auction Results

PJM's 2027/28 Base Residual Auction procured 145,777 MW of capacity—about 6,625 MW below the grid operator's 20% installed reserve margin target. That target estimates how much capacity prevents more than one unexpected power outage every 10 years.

The auction delivered only a 14.8% reserve margin, well below the 20% reliability requirement. It marked the first time the entire PJM regional transmission organization, including Fixed Resource Requirement areas, failed to meet reliability standards.

Capacity prices hit the $333.44/MW-day price cap across PJM's entire region—a record-high price for the third consecutive auction. The total capacity bill reached $16.4 billion, up from $2.2 billion in December 2022.

Joe Bowring, president of independent market monitor Monitoring Analytics, told CNBC he's never seen the grid under such projected strain.

Data Centers: 94% of Load Growth

The forecast peak load for 2027/28 is approximately 5,250 MW higher than the previous auction's forecast. Nearly 5,100 MW of that increase—94%—comes from data center demand.

"Data center load growth is the primary reason for recent and expected capacity market conditions, including total forecast load growth, the tight supply and demand balance, the significant shortfall in cleared capacity, and high prices," stated Monitoring Analytics.

The numbers reflect an unprecedented shift in electricity demand patterns. PJM's 2025 Long-Term Load Forecast projects peak load growth of 32 GW between 2024 and 2030, with data centers responsible for 94% of this increase.

The trajectory in Virginia's "Data Center Alley" illustrates the acceleration. The 2022 forecast projected 5,700 MW of load growth by 2037 in PJM's Dominion Zone. The 2025 forecast: over 20,000 MW from data centers alone in the same zone by 2037.

$21.3 Billion in Data Center Costs

Data centers that haven't been built—but could come online by 2027/28—account for approximately $6.2 billion of the $16.4 billion auction costs.

Over the last three base capacity auctions, costs related to data center forecasts above existing loads totaled $21.3 billion—45% of the $47.2 billion in total cleared capacity costs.

The market monitor estimated data centers were responsible for 63% of the price increase in the 2025/26 auction alone, translating to $9.3 billion in costs recovered from all ratepayers across PJM.

Consumer Bill Impacts

The capacity market costs flow directly to residential electricity bills. Starting June 2025, Washington D.C. Pepco customers saw bills increase by an average of $21/month—approximately half attributable to the spike in capacity market prices from data center demand.

The average monthly residential bill increased from $121 in 2021 to $156 in 2025—a nearly 30% rise. Government analysts expect prices to continue outpacing inflation in 2026.

Wholesale electricity costs as much as 267% more in areas near data centers than it did five years ago, with costs passed directly to customers.

Political Backlash: Sanders and DeSantis Align

The grid crisis has created unexpected political alignment. Senator Bernie Sanders became the first federal legislator to call for a national moratorium on AI data center construction in December 2025.

"I will be pushing for a moratorium on the construction of data centers that are powering the unregulated sprint to develop & deploy AI," Sanders stated. "The moratorium will give democracy a chance to catch up, and ensure that the benefits of technology work for all of us, not just the 1%."

Florida Governor Ron DeSantis endorsed similar restrictions within days. "You do not have enough grid capacity in the United States to do what they're trying to do," DeSantis stated.

Their reasoning differs—Sanders cites worker displacement and corporate consolidation; DeSantis emphasizes grid stability and electricity prices—but the policy prescription aligns. More than 230 environmental advocacy groups demanded a moratorium in December 2025, citing unsustainable water and electricity consumption.

"Our capacity market is breaking under the weight of data center demand and a dysfunctional interconnection queue," said CUB Executive Director Sarah Moskowitz.

Virginia's Rate Class Solution

Virginia hosts over a third of the world's data centers, with most concentrated in Northern Virginia. The state's response to cost shifts provides a model for other jurisdictions.

The Virginia State Corporation Commission approved a new rate class (GS-5) for customers demanding 25 MW or more, effective January 1, 2027. These large customers must pay a minimum of 85% of contracted distribution and transmission demand, and 60% of generation demand—regardless of actual usage.

The Commission rejected Dominion's requested rate increases of $822 million in 2026 and $345 million for 2027, approving $565.7 million and $209.9 million instead.

A JLARC study found that a typical Virginia residential customer could see generation and transmission costs increase by $14 to $37 monthly in constant dollars by 2040 without rate class restructuring.

The Interconnection Queue Problem

New generation takes eight years to come online in PJM—but capacity auctions secure future supply on a two-to-three-year forward basis. The mismatch creates structural shortfalls.

Before 2023, PJM's serial interconnection study system created a backlog exceeding 140 GW, with some resources waiting over five years for approval. The backlog has been reduced to about 46 GW through reforms, with processing expected to complete by end of 2026.

Approximately 57 GW of projects have completed PJM's study process and either signed or been offered generation interconnection agreements. But even with reforms, the timeline from application to commercial operation has risen from less than two years in 2008 to over eight years currently.

"We need PJM to file, without further delay, meaningful changes to its rules to overcome the interconnection and financing barriers to building new generation to close the resource adequacy gap before the end of this decade," FERC Commissioner David Rosner stated.

Behind-the-Meter Generation Accelerates

The grid bottleneck has accelerated behind-the-meter (BTM) power generation. By 2030, 38% of data centers are expected to use on-site generation for primary power, with 27% fully powered by on-site systems, according to Bloom Energy.

Natural gas leads the BTM transition. In 2024, orders for large gas turbines hit a 20-year high of 14 GW, driven by data center load growth. The trend accelerated in 2025, with 18 GW of US-bound gas turbine orders in the first half alone.

Small modular reactors (SMRs) represent the long-term bet. In January 2026, Oklo and Meta signed an agreement to build a 1.2 GW nuclear campus in Ohio, primarily for Meta's data centers, with first power targeted for 2030.

ENTRA1 announced an agreement with TVA to deploy up to 6 GW of nuclear capacity using NuScale's SMR technology—about 72 reactor modules across multiple sites.

However, first-of-a-kind SMRs could cost nearly three times more than natural gas ($331/MWh versus $124/MWh). Lux Research projects SMRs won't be cost-competitive before 2035.

Demand Response: The Flexibility Question

A Duke University study found data centers could unlock up to 76 GW of new capacity by curtailing energy use during grid stress. If new data centers met an annual curtailment rate of 0.5%—just hours per year—it could make nearly 100 GW of new load available without expanding generation.

The Aurora AI Factory in Virginia, opening H1 2026, serves as the first facility built to power-flexible standards. NVIDIA, Emerald AI, EPRI, Digital Realty, and PJM announced the 96-MW project in October 2025.

The Data Center Flexible Load Initiative (DC Flex) aims to deploy five to ten large-scale flexibility hubs by 2027, demonstrating how data centers can provide demand response and grid services.

But skeptics question hyperscale flexibility assumptions. "Downtime carries extremely high financial and operational costs, which limit how much demand they can realistically curtail or shift," one analysis noted. "This fully islanded large load just doesn't exist today."

PJM's market monitor has urged FERC to rule that large data centers can only come online if the grid operator can still meet reliability metrics—a significant policy shift from treating large loads as market participants to treating them as reliability risks.

National Grid Strain

PJM's crisis reflects broader national trends. Goldman Sachs projects global data center power demand will increase 50% by 2027 and 165% by 2030 compared to 2023.

US data center demand projections:

• 2026: 75.8 GW for IT equipment, cooling, and support
• 2028: 108 GW
• 2030: 134.4 GW

Grid Strategies' tally of utility forecasts indicates peak demand will reach 166 GW by 2030—a sixfold increase from forecasts three years ago. Data centers comprise roughly 90 GW of that growth.

The Department of Energy projects data centers could consume 6.7% to 12% of all US energy by 2028. Meeting this demand requires $720 billion in new utility infrastructure worldwide by 2030, with US utilities needing $50 billion for data center generation capacity alone.

What PJM's Shortfall Means for AI Infrastructure

The 6 GW shortfall creates immediate planning constraints for data center operators:

Location becomes power-constrained. Markets with available generation—Texas ERCOT, parts of the Southeast, Pacific Northwest—gain competitive advantage. PJM territories face either project delays or mandatory behind-the-meter generation requirements.

Interconnection timelines extend projects by years. Eight-year queues mean facilities planned today may not receive grid power until 2033-2034. Behind-the-meter generation or purchased power agreements become mandatory, not optional.

Capacity costs become major operating expenses. The 10x increase in capacity prices ($28.92 to $333.44/MW-day) translates to roughly $120/kW/year in additional costs for grid-connected loads.

Political risk enters project planning. Moratorium proposals from both Sanders and DeSantis signal that data center permitting may face new scrutiny regardless of which party controls state or federal government.

Mitigating Factors

Despite the shortfall, several factors may reduce the gap before the June 2027 delivery year begins.

PJM's demand forecast used January 2025 estimates. A new forecast incorporating stricter vetting of potential large loads and reduced economic outlook could lower projected demand significantly.

Some data centers in the forecast may not materialize. Speculative projects, financing failures, and timeline delays historically reduce realized load growth below forecasts.

The market signal works both ways. Record-high capacity prices should attract new generation investment, though the interconnection queue limits how quickly supply can respond.

Emergency demand response programs provide last-resort options, though reliability implications remain uncertain at the projected shortfall scale.

Key Takeaways

For Data Center Developers:

• PJM territories require behind-the-meter generation or long-term purchased power agreements
• Project timelines should assume 5+ year grid interconnection waits
• Site selection must weight power availability as primary constraint

For Grid Operators and Utilities:

• Data center rate classes protect residential customers but may accelerate BTM generation
• Interconnection reform remains critical—eight-year timelines are structurally incompatible with reliability needs
• Demand response programs need validation at scale before assuming flexibility

For Policy Makers:

• The Sanders/DeSantis alignment signals bipartisan concern about unconstrained data center growth
• Virginia's rate class model provides a template for cost allocation reform
• FERC intervention on large load interconnection may become necessary

For AI Infrastructure Planners:

• The current AI buildout trajectory assumes power availability that may not exist
• Regional diversification reduces concentration risk in constrained markets
• Power flexibility design should be standard for new facilities, not optional

About Introl

Introl deploys GPU infrastructure for AI data centers across challenging power environments. With 550 HPC-specialized field engineers operating in 257 global locations, Introl has deployed over 100,000 GPUs—including projects requiring behind-the-meter generation, grid interconnection coordination, and power-flexible designs. When power becomes the constraint, Introl ensures AI infrastructure still gets built.