The rapid expansion of artificial-intelligence data centers helped drive a ninefold increase in the price paid to keep power plants available across the largest electricity grid in the United States.
PJM Interconnection’s capacity-market clearing price rose from $28.92 per megawatt-day for the 2024–2025 delivery year to $269.92 per megawatt-day for 2025–2026.
That represents an increase of approximately 833 percent in a single auction cycle.
The increase affects a region covering all or parts of 13 states and Washington, D.C., including Virginia, Maryland, Pennsylvania, Ohio, Illinois and New Jersey. PJM manages electricity supply and transmission for approximately 67 million people.
Artificial-intelligence data centers were a major cause of the demand growth behind the price shock. However, saying that AI alone caused the entire 833 percent increase would be too simplistic.
The grid was also dealing with retiring power plants, slower-than-needed construction of replacement generation, transmission limitations and changes to the rules governing how much reliable capacity PJM purchases.
The result was a market in which electricity demand was rising quickly while dependable supply remained increasingly constrained.
The 833% Figure Refers to Capacity, Not Ordinary Electricity
Electricity markets contain several different prices.
The energy market pays power plants for the electricity they produce and deliver at a particular time. Those wholesale energy prices can change hourly based on weather, fuel costs, local congestion and demand.
The capacity market is different.
PJM’s capacity auction pays generators and other eligible resources to guarantee that they will be available during a future delivery year, particularly when demand reaches its highest levels.
A power plant can receive capacity payments even when it is not producing electricity every hour. The payment compensates it for remaining ready to operate when the grid needs it.
The $28.92 and $269.92 figures are therefore measured in dollars per megawatt-day of committed capacity. They are not direct comparisons of the average wholesale price of each megawatt-hour actually consumed.
The Federal Energy Regulatory Commission’s PJM overview explains that PJM operates both competitive wholesale electricity markets and the regional transmission system.
The distinction matters because a headline saying wholesale electricity prices rose 833 percent may lead readers to believe every unit of electricity suddenly became nine times more expensive.
What rose ninefold was the cost of reserving enough dependable generating capacity to meet projected demand.
That cost still enters consumer bills, but it represents only one component of the total retail electricity price.
How the Price Rose From $28.92 to $269.92
PJM conducts what it calls a Base Residual Auction to secure electricity capacity several years ahead of the period in which it will be needed.
For the 2024–2025 delivery year, the main PJM region cleared at $28.92 per megawatt-day.
In the auction for the 2025–2026 delivery year, the price jumped to $269.92 per megawatt-day across most of the region.
The percentage increase is calculated by comparing the $241 difference with the original $28.92 price.
That produces an increase of approximately 833 percent.
Certain areas faced even higher capacity prices because of local transmission constraints and tighter supply.
The Baltimore Gas and Electric zone cleared at $466.35 per megawatt-day, while the Dominion zone covering areas of Virginia and North Carolina cleared at $444.26.
Northern Virginia is particularly important because it contains the world’s largest concentration of data centers.
Facilities serving cloud computing, online services and increasingly AI workloads consume enormous quantities of electricity around the clock.
Data Centers Were a Primary Driver
PJM’s independent market monitor concluded that existing and forecast data-center demand was the primary reason capacity-market costs rose so sharply.
Monitoring Analytics estimated that data-center load increased revenue in one recent capacity auction by approximately $7.3 billion, accounting for about 82 percent of the increase in auction revenue.
Across two consecutive auctions, existing and projected data-center demand was associated with approximately $16.6 billion in capacity-market revenue, roughly half of the auctions’ combined total.
Those figures do not mean every dollar went directly to an AI company or that data centers physically purchased all the capacity themselves.
PJM forecasts how much electricity the entire region will require, adds a reserve margin and then purchases enough dependable capacity to meet that forecast.
When data centers raise projected demand, PJM must acquire more generation capacity.
If available supply does not grow at the same pace, the clearing price rises for all capacity purchased through the auction.
The higher total is then allocated through utilities and electricity suppliers, which can pass costs to households and businesses.
Not Every Data Center Is Exclusively an AI Facility
The phrases “AI data center” and “data center” are sometimes used interchangeably, but they are not identical.
Data centers have supported websites, streaming, banking, cloud storage, government systems and corporate software for decades.
AI is now accelerating their growth because training and operating large models require substantial computing power.
A facility may serve a mixture of traditional cloud computing and AI workloads rather than having one exclusive purpose.
It may also change uses over time as companies install new processors or shift computing tasks between locations.
For that reason, grid forecasts generally track data-center electricity demand as a category. They do not always provide a precise division between AI and non-AI activity.
The current expansion is nevertheless closely connected to the AI investment boom, with major technology companies spending heavily on new server campuses, processors, networking equipment and supporting electricity infrastructure.
AI Servers Require Large Amounts of Continuous Power
Advanced AI systems run on dense clusters of graphics processors and other specialized chips.
The computing equipment consumes electricity directly, while cooling systems remove the heat produced by servers.
Additional power is used by networking equipment, pumps, backup systems, lighting and power-conversion equipment.
Unlike a factory that may close overnight, many data centers operate continuously.
Large campuses can demand hundreds of megawatts. Several proposed developments involve one gigawatt or more when built in stages.
A one-gigawatt load is comparable to the output of a large conventional or nuclear generating unit.
Connecting one such project is difficult. Connecting many of them within the same region can transform the grid’s demand outlook in only a few years.
This creates a fundamental timing problem.
A data center can sometimes be planned and constructed faster than a new power plant or major transmission line can be approved, financed, connected and completed.
Northern Virginia Is at the Center of the Pressure
Virginia’s Dominion service territory contains “Data Center Alley,” a cluster centered in and around Loudoun County.
The region became attractive because of its extensive fiber-optic connections, proximity to government and business customers, available land and established data-center ecosystem.
That concentration now creates unusual electricity demand.
The Dominion capacity zone cleared at $444.26 per megawatt-day in the 2025–2026 auction, significantly above the $269.92 price across most of PJM.
The difference reflected local reliability needs and limits on how much electricity could be imported through the transmission system.
During a July 2026 heat wave, PJM also reported serious transmission congestion in northern Virginia as high air-conditioning demand combined with the area’s large data-center load. Spot wholesale prices rose from about $30 to more than $300 per megawatt-hour in stressed periods.
That later event was separate from the 833 percent capacity-auction increase, but it demonstrated how concentrated data-center demand can also affect real-time grid operations.
Power-Plant Retirements Also Tightened Supply
Data centers were only one side of the equation.
PJM had also lost generating resources as older coal, gas and other plants retired or announced plans to close.
Some plants became uneconomic. Others faced environmental requirements, maintenance costs or policy pressure.
Retirement is not necessarily a problem when new supply replaces the old plant promptly.
The challenge in PJM is that new generation has not entered service fast enough to match both retiring capacity and rapidly growing demand.
Thousands of proposed solar, wind, battery and gas projects have sought grid connections, but many have faced lengthy engineering studies, transmission-upgrade requirements, permitting delays or financing uncertainty.
A capacity market can send a strong price signal indicating that new generation is valuable.
It cannot make permitting, transformers, turbines, transmission lines and construction crews appear immediately.
Auction Changes Also Affected the Result
PJM changed how it calculated the amount and reliability value of available capacity before the 2025–2026 auction.
Those reforms followed concerns that previous rules did not adequately account for the risk of power plants failing during extreme weather.
Winter Storm Elliott in December 2022 caused a large number of generators to become unavailable at the same time that demand surged.
PJM subsequently revised assumptions about generator performance and reliability.
These changes reduced the capacity value assigned to some resources and increased the amount PJM needed to procure.
The revised design helped produce a more conservative reliability assessment, but it also placed upward pressure on the auction price.
The 833 percent jump therefore reflected an interaction among higher forecast demand, reduced accredited supply and updated market rules.
The Auction Cost Rose Into the Billions
PJM’s 2025–2026 auction procured approximately 135,684 megawatts of unforced capacity.
The total cost to customers rose dramatically compared with the previous auction.
Capacity payments are ultimately collected from electricity users across the region.
Retail bills combine several charges, including energy generation, capacity, transmission, distribution, environmental programs, taxes and utility operating expenses.
Because capacity is only one component, an 833 percent capacity-price increase does not create an 833 percent household-bill increase.
The effect depends on the utility, state regulations, contract timing, customer class and the share of the bill connected to PJM capacity costs.
Some customers may see the increase introduced gradually as supply contracts and regulated rates are updated.
Others may experience a more noticeable adjustment in one rate case.
Household Bills Did Not Rise Ninefold
A family whose monthly electricity bill was $150 should not expect it to become $1,400 solely because of the capacity auction.
The capacity component is much smaller than the complete bill.
However, the increase was large enough to add meaningful pressure to rates already affected by transmission upgrades, fuel prices, storm recovery and other utility investments.
The impact can differ substantially across states.
Customers in zones with higher local capacity prices, such as the Baltimore and Dominion regions, may face greater pressure than customers in areas clearing at the broader PJM rate.
Utility structures also matter. Some states use regulated utilities, while others allow customers to choose competitive electricity suppliers.
The higher capacity cost eventually affects both systems because suppliers must recover the expense of serving peak demand reliably.
The Following Auction Set Another Record
The 833 percent increase was not a one-time return to normal pricing.
PJM’s next capacity auction, covering the 2026–2027 delivery year, cleared at the federally approved cap of $329.17 per megawatt-day throughout the region.
That was another 22 percent increase over the previous $269.92 regional price.
PJM said the higher price could add roughly 1.5 percent to 5 percent to some customer bills, depending on the state and utility.
The auction also procured more new, upgraded, reactivated and imported generation than the preceding year, suggesting that high prices were beginning to attract some supply response.
However, prices remained exceptionally high because demand continued growing and the grid still needed more dependable resources.
Prices Have Since Climbed More Than 1,000%
By July 2026, PJM’s capacity-price problem had intensified further.
The latest annual auction reached approximately $325 per megawatt-day, near a temporary price cap established to protect customers.
PJM still fell roughly 6.8 gigawatts short of its reliability target.
Reuters reported that without the price limit, the regional clearing price could have reached $554.72 per megawatt-day, while the Illinois ComEd zone might have exceeded $775.
Compared with the $28.92 price before the initial surge, recent capped capacity prices represent an increase of more than 1,000 percent over two years.
The continuing shortage suggests that the original 833 percent increase was an early warning rather than the peak of the problem.
Capacity Prices Pay Existing Plants More
High capacity prices increase revenue for generators that successfully clear the auction.
That includes natural-gas, nuclear, coal, hydroelectric, wind, solar, battery and demand-response resources, although conventional gas, nuclear and coal plants provide most of PJM’s accredited capacity.
The increased revenue can help prevent an existing plant from retiring and can support upgrades or new construction.
Power-generation companies may therefore benefit financially from a tighter market.
After the 2024 auction results were announced, shares of several major independent power producers rose as investors anticipated higher capacity revenue.
This is how the market is intended to work: scarcity produces higher prices, which encourage investment.
The concern is that customers must pay those higher prices immediately, while new generation can take many years to arrive.
Ratepayers May Be Funding AI’s Grid Requirements
The controversy is not merely that data centers use large amounts of electricity.
It concerns who pays for the infrastructure and capacity needed to serve them.
When PJM adds projected data-center demand to its regional load forecast, it procures more capacity through the market.
The resulting cost is generally spread among electricity customers according to established allocation rules.
That means households, small businesses, schools and factories can help pay for capacity needed partly because a large technology company is building a new data center.
Critics argue that unusually large customers should bear more of the incremental cost they create.
Technology companies respond that they already pay substantial utility bills, fund connection infrastructure and create tax revenue, construction work and economic investment.
The dispute centers on whether their payments fully cover new generation, transmission and reliability risks or whether a portion is shifted to existing customers.
Forecasting Creates Another Risk
PJM must purchase capacity before it knows exactly how much electricity every proposed data center will use.
Technology companies may submit plans to more than one utility or explore multiple locations before selecting a final site.
Projects can be delayed, reduced or canceled.
If PJM assumes all proposed facilities will arrive, it may procure too much capacity and charge customers for demand that never materializes.
If it discounts too many projects, the grid may have insufficient supply when new facilities begin operating.
This problem becomes more difficult when data centers request extremely large connections and expect service within a few years.
The Institute for Energy Economics and Financial Analysis warned that capacity markets may be preparing for worst-case data-center forecasts even when some projected demand does not ultimately appear.
Accurate, binding development commitments are therefore becoming increasingly important.
AI Companies Are Exploring Their Own Power Supplies
Major data-center operators are seeking ways to secure electricity without relying entirely on existing grid supply.
Some companies have signed long-term agreements with nuclear plants.
Others are supporting new natural-gas generation, renewable-energy projects, batteries, geothermal systems or proposed small modular reactors.
On-site generation can reduce certain grid constraints, but it introduces other concerns.
Gas turbines create emissions and may operate for decades. Nuclear projects can take years to license and construct. Renewable projects still require transmission and dependable backup when wind or solar output is low.
A data center may also remain connected to the grid for reliability even when it claims to match its annual consumption with renewable contracts.
The source used on paper does not always reflect what physically supplies the facility during every hour.
Flexible AI Workloads Could Reduce Pressure
Not every computing task must run at an exact second or in one particular location.
Some AI training, data processing and other non-urgent workloads could be shifted to periods when electricity is cheaper and the grid has spare capacity.
Companies operating several data centers might also move computing between regions.
Flexible operation could reduce peak demand and allow more efficient use of renewable generation.
Critical services requiring immediate responses cannot be delayed in the same way.
Data centers would also need financial incentives and technical systems that reward flexibility without compromising customer privacy, security or performance.
PJM and other grid operators are considering rules that would allow very large customers to accept interruptions or bring their own generation instead of being treated as entirely inflexible demand.
More Generation Alone Will Not Solve Every Problem
Building new power plants is essential, but location matters.
A generator hundreds of miles away cannot always solve a local shortage when transmission lines are congested.
Northern Virginia’s data-center concentration has created a need for both additional generation and stronger transmission connections.
New power lines can face community opposition, land-acquisition disputes and lengthy state approvals.
Transformers and other high-voltage equipment also have extended manufacturing lead times.
In some locations, the fastest solution may involve locating data centers where spare generation and transmission already exist rather than continuing to cluster them in one saturated area.
Electricity pricing and connection rules can influence those decisions.
Reliability Is Becoming a Bigger Concern
Higher prices signal that electricity capacity is scarce.
The more serious issue is whether the grid can maintain service during extreme heat, severe cold or unexpected power-plant failures.
PJM’s recent auction failed to acquire its complete target reserve margin, leaving a multigigawatt shortfall.
That does not mean blackouts are certain.
Grid operators can use emergency imports, voluntary demand reductions and other measures when conditions become tight.
A persistent shortfall reduces the safety cushion available when several problems happen at once.
Data centers can intensify the risk because their load is large and often continuous, while residential cooling demand can rise dramatically during heat waves.
AI Is Not the Only Source of Growing Demand
Electricity demand is also increasing because of new factories, electric vehicles, building electrification, cryptocurrency mining and expanding industrial production.
Federal incentives and corporate investment have encouraged semiconductor plants, battery factories and other large manufacturing projects.
Some aging urban areas also require additional electricity as populations and economic activity grow.
AI-related data centers stand out because of the speed, scale and geographic concentration of their demand.
In parts of PJM, data centers account for most forecast load growth.
Nationally, the effect varies considerably. Regions with more available generation and transmission may absorb new facilities without experiencing the same price shock.
The 833 percent figure should therefore not be applied to every US electricity market.
Other Regions Have Produced Different Results
Research on data centers and retail electricity prices has produced mixed findings.
A 2026 working paper estimated that data centers may have modestly lowered average US retail electricity rates between 2015 and 2024 by spreading existing fixed grid costs across more electricity sales. The authors cautioned that future supply constraints could reverse that effect.
This does not contradict PJM’s recent experience.
A new large customer can lower average costs when the grid has spare capacity and infrastructure is already available.
The same customer can raise prices sharply when demand arrives faster than new generation and transmission.
The effect depends on timing, location, regulation, grid conditions and whether the data center pays for the incremental infrastructure it requires.
What Regulators Are Considering
Policymakers are examining several approaches to protect customers while allowing data-center development.
Utilities may create special rate classes for extremely large loads.
Data centers could be required to make long-term financial commitments before their projected consumption enters grid forecasts.
They could pay for dedicated generation or accept curtailment during emergencies.
Connection rules might require companies to cover transmission upgrades and provide deposits that are lost if a project is canceled.
Regulators could also accelerate approval of new generation and transmission while imposing stronger environmental and consumer protections.
No single policy will solve every problem.
Rules that are too weak may shift costs and reliability risks onto households. Rules that are too restrictive could push investment to other regions without improving the grid.
Consumers Should Expect Continued Rate Pressure
Capacity costs for several delivery years have already been determined through PJM auctions.
That means some increases will continue flowing into retail bills even if data-center construction slows tomorrow.
Utilities may also seek approval for transmission projects, power-plant contracts and distribution upgrades connected with load growth.
The exact customer impact will depend on state regulators and local utilities.
Households can review utility notices and public rate cases to determine how much of a proposed increase is attributed to capacity, transmission or large-load growth.
The broader political question is whether existing customers should pay higher bills so the grid can serve a rapidly expanding technology industry.
That debate is likely to intensify as more projects seek electricity connections.
The Headline Needs Precise Wording
It is fair to say that AI data centers helped push one major power market’s price up 833 percent in a year.
It is less accurate to say they caused wholesale electricity prices generally to rise 833 percent.
The increase occurred in PJM’s capacity auction, which purchases future grid readiness.
The price rose from $28.92 to $269.92 per megawatt-day between consecutive delivery-year auctions.
Data-center demand was identified by PJM’s independent market monitor as the primary driver of the increase in auction costs, but reduced supply, power-plant retirements and market-rule changes also contributed.
A more precise headline is therefore: AI data centers helped push PJM’s power-capacity price up 833 percent in one year.
The Main Takeaway
PJM’s regional capacity price increased from $28.92 per megawatt-day for 2024–2025 to $269.92 for 2025–2026.
That was an 833 percent increase.
The price measures what consumers pay to keep enough power-generation capacity available for periods of peak demand. It is not the same as the hourly wholesale price of electricity and does not mean household bills increased by 833 percent.
Rapidly growing data-center demand was a major driver. PJM’s independent market monitor estimated that data centers accounted for approximately 82 percent of the increase in revenue from one recent auction.
Power-plant retirements, slow construction of replacement resources, transmission constraints and new reliability rules also tightened the market.
The problem has not disappeared. Subsequent auctions set additional records, and PJM’s latest results remained near a regulatory price cap while still falling short of the grid’s reliability target.
The AI boom is no longer only a technology story.
It is changing how quickly the United States must build power plants, transmission lines and other grid infrastructure and raising an increasingly urgent question about who should pay for them.
