The Ratepayer Pays First

A household in Maryland paid 13.2% more for each kilowatt-hour of electricity in the twelve months to July 2026 than in the twelve months before, by DeepStack's reading of the Energy Information Administration's monthly sales and revenue data, first published for July on September 24. The country paid 6.6% more. Rank all 50 states and Washington, D.C., by that change and seven of the nine jurisdictions PJM serves wholly or largely land in the top nine: the District at 21.9%, then Maryland, Pennsylvania at 12.8%, New Jersey at 12.2%, Ohio at 11.7%, Virginia at 11.5% and Illinois at 11.4%. Delaware rose 5.7% and West Virginia 0.6%. The median of the nine was 11.7%; the median of the other 42 was 5.6%. PJM runs the grid and the wholesale power market for all or part of 13 states and the District, and carries the country's densest cluster of data centers.

The bill starts in an auction. PJM buys capacity, a plant's promise to be ready at the hour of peak demand, in an auction held before each delivery year, and the price reaches every customer's bill. The auction of July 2024 set it at $269.92 per megawatt-day for the year from June 2025, nine times the $28.92 of the auction before, and every auction since has cleared at the price ceiling, as tscsw recounted on August 11. At $269.92, a 1,000-megawatt plant collects about $98.5 million a year for standing ready. The auction for the year from June 2027, tscsw wrote on February 26, was the first since the market began in 2007 to miss PJM's reliability requirement: a reserve margin, the cushion of plants above expected peak demand, of 14.8% against a 20.0% target, 6,623 megawatts short. Part six, on September 19, followed the price to the ceiling and relayed an estimate that data centers caused 63% of the first jump. Part thirteen asks who carries that bill under the rules.

DeepStack read 92 filings, posts and episodes by 60 voices across two clusters of its vault, among them the regulatory record: briefs at the Federal Energy Regulatory Commission and the Texas commission, and the grid planners' reliability assessments. It verified 106 agreements and 70 disagreements, 10 of them direct contradictions.

The load pays its way

The first reading holds that the data center pays for what it causes. Amin Vahdat, who runs Google's AI infrastructure, told Sonya Huang on Sequoia's Training Data on October 6 that a utility building lines and plants for Google could in principle lift other customers' rates because of how billing works, so Google pays for the transmission lines, upgrades and added utility stations it requires. Google and Lancium made the same case to the Texas commission on December 18, 2025: assign costs to whoever causes them, with minimum transmission charges large enough that households and small businesses are shielded from shifts. Crusoe Energy told FERC on March 16 that co-located generation, a power plant built beside the data center it feeds, lowers ratepayer costs when it is set up correctly. EDF Power Solutions said the same day that such plants can ease congestion and cut the network upgrades the grid would otherwise need.

Utilities and generators describe the same contract from the other side. NRG Energy told the Texas commission on April 17 that a large load, the regulators' term for the biggest new customers, spends far more bringing its own generation than the $50,000 per megawatt of security the proposed rule demands, and asked that such projects be encouraged as an answer to affordability. Kingswell, on May 11, credited Berkshire Hathaway Energy with joining the buildout only where the largest cloud companies carry the full cost of building and running it.

DeepStack's reading is that this side pays for what it causes at its own site, from the substation to the deposit that holds its place in line. A data center also pays the capacity price on its own load, like any customer. What none of these filings offers to pay is the rise its demand causes in the price every other customer pays on the same kilowatt-hours as last year. The closest any comes is bringing its own new generation, which NRG and Crusoe propose.

The bill no state can shield

The second reading says the bill arrives through the market before any invoice is drawn, and years before the rules that would assign it to large loads take effect. Monitoring Analytics, PJM's independent market monitor, told FERC on December 5, 2025, that the central problem is a shortage of generation to serve the new loads reliably, and that the failure to impose a reliability requirement had driven large increases in capacity prices, and in energy prices and transmission costs as well. It added that whatever tariffs the states write, state regulators cannot protect retail customers from what those loads do to PJM's wholesale prices for capacity, energy and reserves. Its remedy was a queue for large data center loads, with a faster path for those that bring their own new generation.

The Virginia State Corporation Commission made the distributional case on December 15, 2025. If existing plants may leave the market and wire their output straight to data centers, behind the meter in the industry's term, the data centers get partly depreciated, cheaper generation, and the customers who remain pay for the costlier plants built to replace it. Unless such arrangements protect existing customers, the commission warned, they work as a wealth transfer from retail customers to generators such as Constellation, which it said was already profiting from a market short of resources and expensive. The capacity price is revenue to the generators that clear the auction. The Maryland Office of People's Counsel put a number on the transmission side on June 11: $1.6 billion of added cost on Maryland bills over 10 years, from $2 billion of investment in PJM's three most recent cycles of regional projects that it says was misallocated to the state, projects it says data center growth outside Maryland drove. The Indicated PJM Transmission Owners, in a rehearing request Exelon filed with FERC on January 16, said a co-located load that funds its own upgrades but pays nothing toward the existing grid leaves other customers holding the bag.

Read side by side, two filings that describe the harm alike disagree about the remedy. Virginia's commission saw no defect in PJM's own tariff that called for a PJM-only fix, and wanted the question settled in FERC's general rulemaking on large loads, with state authority over retail service intact. The market monitor had argued ten days earlier, in a separate proceeding, that state tariffs can reach the cost of connecting a load but not what the load does to wholesale prices. Robin Millican, on Columbia Energy Exchange, which Jason Bordoff hosts with Bill Loveless, on June 30, named tariffs that make large loads pay their incremental grid costs as one of the structural fixes. Maryland's counsel adds a limit: without regional reform, a state tariff can magnify the incentive to overbuild and move the cost onto customers elsewhere in PJM.

Maryland's counsel cites the industry's answer, the Ratepayer Protection Pledge, under which the largest data center developers commit to pay for new power delivery upgrades, network upgrades included, so the expense does not reach households. Market Sentiment, on July 10, read two moves of January 2026 as the end of the best case it had modeled for the power companies it follows: the governors of all 13 PJM states declared with the administration that data centers pay their grid costs, and the auction's price ceiling was extended to 2030. DeepStack reads the pledge as a promise about the wire. A household in Baltimore pays the capacity price whether or not a data center in northern Virginia funded its substation. In a separate federal proceeding, Earthjustice argued on April 24 that no commenter could name the penalty that would make a customer with virtually unlimited capital and strict uptime contracts cut its draw when the grid runs short.

The plant is late

The third reading says neither the load nor the plants arrive when the bills assume. The North American Electric Reliability Corporation, NERC, noted in an assessment of the Midwest grid that for two years industry projections of new resources had outpaced what was actually completed, held back by supply chains, permits and developers' own decisions. Its 2025 long-term assessment warned that large loads make forecasts volatile; revised ERCOT and PJM forecasts showed some projects slowing or failing even as requests for later years kept rising. The Electric Reliability Council of Texas, ERCOT, reported about 438 gigawatts of large-load requests on May 24, against about 452 gigawatts of generation requests, most of the load requests with no study submitted. Reyk Knuhtsen of SemiAnalysis, on June 18, counted 311 gigawatts of the more than 410 gigawatts ERCOT was assessing in April as phantom data center demand, against an all-time Texas peak of about 85 gigawatts.

What is real is slow. Jeremie Eliahou Ontiveros, also of SemiAnalysis, wrote on June 25 that utilities are pushing promised load ramps out by years for lack of long-lead equipment, while buyers carry the risk through letters of credit, deposits and take-or-pay contracts and the utilities, in many cases, face no penalty for delivering late. Mia Silverio of Prof G Markets relayed on September 28 an estimate that 30% to 50% of the data center megawatts planned to open this year will be delayed, with opposition groups in the United States above 840, from fewer than 50 at the start of 2024.

The strongest form of this side cuts against the house. tscsw's February 26 post held that if 15 to 20 gigawatts of data center load moves behind the meter by 2030, the grid's deficit and the scarcity premium paid to generators in PJM and ERCOT would shrink. Millican relayed industry estimates on June 30 that roughly 260 gigawatts of new load could be met by using the existing grid better, through dynamic line ratings, reconductoring and demand flexibility, while granting that new building is still needed. Eliahou Ontiveros puts the limit elsewhere: a major bottleneck is generation and the equipment to connect it. The same February 26 post argued that queues, turbine shortages and permits keep projects as large as a $33 billion gas plant from becoming megawatts, and NERC's forecasts show some of the demand fading on the buyers' side.

DeepStack's reading is that lateness on both ends lengthens the household's bill. A deposit buys a data center its place in line; until the plant behind it runs, the scarcity sits in everyone's capacity price.

The grid without an auction

Texas runs an energy-only market with no capacity auction. Its households paid 4.8% more per kilowatt-hour over the year to July, below the national 6.6%. Under the proposed rule NRG commented on, a large load posts security of $50,000 per megawatt, and ERCOT's September 17 update showed requests to connect gas generation up 23% since June, to 80 gigawatts. DeepStack reads that as supply answering the price. Doug Arent offered a frame for the difference in the first part of the June 30 conversation. Where spare generation and wires already existed, as in Nebraska, North Dakota and New Mexico, new load spread fixed costs over more sales and prices fell. In Virginia and the rest of PJM, the load still to come needs new generation that costs more than today's average.

The comparison falls short of proof. Retail prices also move with fuel and with rate cases for local wires, and the high risers outside PJM say so: New Hampshire rose 13.2%, Washington 12.7%, New York 11.1% and Maine 10.0%. Texas carries its own transmission bill as well; Google and Lancium counted about $24 billion of approved 765-kilovolt projects in December 2025 and asked that large loads connect quickly enough to pay their share of it. What the table shows is a median gap of six points between the jurisdictions where the market monitor says the transfer happens and the rest of the country.

The tests ahead

The Energy Information Administration publishes August's figures in late October. The line to track is whether the median twelve-month rise in household prices across PJM's nine jurisdictions stays at least four points above the median of the other 42, from 11.7% against 5.6% in July. Four points is the line because a gap that closes by a third or more would put the work back on the fuel and wires every state shares. PJM's capacity auction for the year from June 2029, scheduled for December 9, is the second check: a clear below the price ceiling would say the scarcity is easing. The data for all of 2026 arrive in late February 2027 and carry the same four-point test.

What would prove DeepStack wrong is the gap falling below four points while PJM's data center load keeps growing, or a rule that makes large loads bring their own capacity followed by an auction that clears below the ceiling. A third exit sits in Texas: if household prices there rise as fast as PJM's, the cause lies in what both grids share. The Data Center Coalition noted on March 16 that PJM had proposed June 1, 2029, as the date its new transmission services for co-located loads take effect; until then, the market price is the bill.

The record answers Part thirteen's question in the order the bills arrive. The data center pays for its wire. The ratepayer pays for the scarcity.