In 2026 the cost of the AI buildout started arriving on household electricity bills, and the argument over who should pay it moved from utility hearing rooms into the US Senate. A 27-page Senate report released on October 9, 2026 found that seven of the largest data center operators, Amazon, Google, Meta, Microsoft, CoreWeave, Digital Realty and Equinix, would not commit to paying for the grid infrastructure their facilities make necessary. In the PJM Interconnection, the grid region serving 67 million people across 13 states and Washington, D.C., the independent market monitor attributes roughly $29 billion in costs over about two years to data center demand. This article covers what the Senate report found, how capacity market math produced a tenfold price increase, what actually lands on a residential bill, and why a single, defensible number for "the data center share" is genuinely hard to produce.
The Senate report that turned rhetoric into receipts
The investigation began on December 15, 2025, when Sens. Elizabeth Warren (D-Mass.), Chris Van Hollen (D-Md.) and Richard Blumenthal (D-Conn.) sent letters to the seven companies. Staff requested information from each and interviewed employees over the following year. The result, "Power and Profits: How the AI Data Center Boom Costs Households and Communities," was first reported by TIME on October 8 and released on October 9.
The central finding is procedural but specific: none of the seven companies would agree to a standard that would require them to pay for building new power infrastructure that would not have been needed if not for their data centers. The companies generally said they would cover the direct costs of serving their facilities, but argued that larger grid investments such as new power plants and transmission lines benefit other customers and should not automatically be assigned to the company that drove the need for them.
Warren framed the stakes directly. "Congress must hold Big Tech accountable so these companies pay their fair share," she said. She has separately called for a national moratorium on new AI data centers until developers agree to cover the full costs. Van Hollen put it this way: "This report lays bare what we have long known: working Americans and local communities are footing the bill for Big Tech's massive expansion of data centers, while these companies continue to operate without transparency."
The report's most useful numbers concern employment and incentives. Data center developers routinely emphasize construction jobs when seeking subsidies, but when the senators' staff asked for comprehensive permanent employment data, several companies refused. Some told investigators permanent staffing runs at roughly one worker per megawatt. A 100-megawatt data center, by that measure, consumes about as much electricity as 100,000 homes while employing about 100 permanent workers after construction finishes. On incentives, the report found the most lucrative breaks are not the attention-grabbing property tax abatements but sales-tax exemptions on computer equipment. GPUs account for an estimated 39% of the spending at an average 1-gigawatt AI data center.
Transparency came in for its own section. All four of the major technology firms in the investigation, Amazon, Google, Meta and Microsoft, routinely sought nondisclosure agreements during data center development, and some acknowledged requesting them from government officials. Microsoft told investigators it would stop seeking NDAs with local governments while continuing to use them with state agencies, utility commissions and utilities; Amazon announced a similar policy. Google and Meta declined to commit. A University of Mary Washington study found NDAs had been signed in 80% of Virginia localities with existing, approved or proposed data centers.
One case study in the report is Louisiana. Meta's planned $50 billion data center in Richland Parish is expected to draw 4,500 megawatts, about four times the peak electricity demand of the entire city of New Orleans. Analysts cited by the report estimate that the local utility Entergy's purchase of a power plant tied to that load could raise the average Entergy customer's bill by $8 to $13 per month. Meta has disputed that its project is responsible for those costs.
How capacity prices rose tenfold in two years
PJM coordinates the electric grid and wholesale power market across 13 states and the District of Columbia, and it supplies northern Virginia, the largest concentration of data centers in the world. Its capacity market is a forward auction that pays generators to guarantee power is available at peak demand. That cost ultimately lands on electricity consumers.
The clearing prices tell the story. The 2024/2025 delivery year cleared at $28.92 per megawatt-day in most zones, according to IEEFA. The 2025/2026 auction cleared at $269.92, the sharpest single-year jump in PJM's history. The 2026/2027 auction cleared at $329.17 per megawatt-day, hitting the FERC-approved price cap. The December 2025 auction for 2027/2028 cleared at the updated cap of $333.44 per megawatt-day and, for the first time ever, fell short of reliability requirements, by 6,625 megawatts.
Monitoring Analytics, the firm acting as PJM's independent market monitor, quantified the data center contribution. In the 2025/2026 auction it attributed 63% of the price increase to data center load, which translates to $9.3 billion in costs recovered from customers across the region in higher electric rates. Across the last two auctions, the monitor found existing and forecast data center load drove $16.6 billion in capacity auction revenue, about half of the $30.8 billion total.
The monitor was pointed about causation. "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, and high prices," it stated, adding that it is "misleading" to describe the results as ordinary tightening: "The current conditions are not the result of organic load growth."
The mechanism matters because it is regional, not local. "Everyone pays those costs," said Ari Peskoe, director of the Harvard Electricity Law Initiative. "They're spread across the region to every business and resident that has an electricity meter."
What actually shows up on a bill
Concrete household figures vary by utility and state. Maryland's Office of People's Counsel estimates that customers there pay somewhere around $168 to $216 more per year on electricity bills primarily due to the data center boom, according to NPR. IEEFA estimated capacity market prices add about $18 per month to the average residential bill in western Maryland and about $16 per month in Ohio. The D.C. Office of the Consumers' Counsel estimated that around $10 of a $21 monthly increase for Pepco residential customers was due to the capacity market spike.
Longer-run projections are larger. NRDC has projected that by 2028 an average family in the PJM region could be paying around $70 a month extra because of forecast data center growth, and that cumulative customer costs could reach up to $163 billion through 2033 if regulators do not intervene. Separately, the Union of Concerned Scientists found that ratepayers in PJM will pay about $4.4 billion for data center-related transmission projects approved in 2024.
For a consumer-facing check, Consumer Reports cited a Bloomberg analysis finding that areas with high concentrations of data centers saw electricity prices jump 267% over the past five years.
Getting to a single national number, though, is not possible. Peskoe said "a single, clean number" is unavailable because "a lot of the data about how much particular data centers are paying is hidden behind non-disclosure agreements." Connor Waldoch, co-founder of the electricity-data firm Grid Status, described the scale of the exposure: "We're talking tens to hundreds of millions of dollars [of infrastructure] that are being built by these local utilities. In many cases, it is just ultimately going to end up in customers' bills." Lucy Qiu, a professor at the University of Maryland, noted that how much of the increase reaches residential customers depends on utilities' power-purchasing contracts, retail rates and regulatory decisions about who pays.
The legislative fight stalled
Congress tried and failed to settle the question. The Ratepayer Protection Act, sponsored by Sen. Jon Husted (R-Ohio) and introduced July 16, 2026, would require state utility regulators to consider standards under which large-load customers such as data centers pay the incremental cost of the generation, transmission and distribution infrastructure needed to serve them. It passed the House on September 16, 2026 by a 417-3 margin, then failed in the Senate on September 30 in a 57-43 vote that fell short of the 60 needed to advance. Husted's framing was blunt: "In plain language, if you create the cost, you should pay the cost."
Democrats objected that the bill was toothless because it required states only to consider the standards rather than to impose them. Senate Minority Leader Chuck Schumer called it a "toothless messaging bill," and Sen. Martin Heinrich (D-N.M.) has pushed an alternative, the GRID Savings Act, that would require FERC to set rules on large-load interconnections, including cost allocation, financial security requirements and cost-estimate disclosure. Because most rate-making authority sits with state regulators, the practical battles will continue in statehouses and public utility commissions.
Why the cost split is genuinely hard to compute
It is worth being precise about the complications, because they cut in both directions. First, separating infrastructure that serves only data centers from infrastructure that is shared is an accounting judgment, not a fact. Qiu noted that the answer "really depends on the scale of the upgrade and also how utilities are doing the accounting of that different share." Second, disclosure is limited, with NDAs and confidential utility filings hiding the underlying figures. Third, not every recent increase is attributable to data centers: fuel costs, extreme weather and equipment that needed replacing anyway all push bills up. Qiu also noted that reliability can improve near data centers because those facilities require robust infrastructure, and that the AI buildout can accelerate grid upgrades the system needed anyway.
There is also a perception problem. Large customers like AI data centers pay less per kilowatt-hour than households, a normal volume discount. But in an environment where 70% of Americans worry data centers will raise their electricity bills, per a Third Way survey, and about half say data center construction is bad for the country, per an Economist/YouGov poll conducted August 28 to 31, 2026, the optics are difficult. President Trump has warned that opposition to data centers could push the AI buildout overseas.
The demand curve behind the bill
The bills are rising because the load is rising. Gartner projects worldwide data center electricity consumption will climb from 448 terawatt-hours in 2025 to 980 TWh by 2030, roughly doubling, with AI-optimized servers alone growing from 93 TWh to 432 TWh. Gartner also predicts 40% of existing AI data centers will be operationally constrained by power availability by 2027. Deloitte estimates US AI data center power demand could grow more than thirtyfold by 2035, reaching 123 gigawatts, up from 4 gigawatts in 2024. Husted's own floor remarks put the buildout in perspective: the United States is expected to build the equivalent of 1,000 major data centers over the next five years, with more than 2,000 projects currently proposed.
What to watch
The action is shifting to state regulators. Virginia regulators ordered Dominion Energy to directly assign some transmission costs to data centers in August 2026, a template other states may follow. Monitoring Analytics has recommended requiring new data centers to supply their own generation rather than drawing on existing PJM supply. FERC has already rejected most of PJM's large-load interconnection plan, and the Department of Energy has pressed PJM on ratepayer protections. The next flashpoint is scheduled: the negotiated price caps that held PJM's 2027/2028 costs down expire before the 2028/2029 base capacity auction, when the market monitor expects the impact of data center load to increase sharply.
The through-line for 2026 is that AI's cheapest-looking costs were never only about chips and electricity. The grid, the rate case and the NDA are all part of the bill, and the fight over who signs it is now a mainstream political issue, not a footnote in an infrastructure filing.