On October 9, 2026 — the same day FERC rejected PJM’s Interim Resource Adequacy Service — three Democratic senators released a 27-page report that takes the cost-allocation fight out of the dockets and into Congress. Power and Profits: How the AI Data Center Boom Costs Households and Communities, from Senators Elizabeth Warren, Chris Van Hollen, and Richard Blumenthal, is the product of a nearly yearlong investigation that began with letters to seven data center companies on December 15, 2025. Its central finding is a gap between what data center operators say when they promise to pay their “fair share” and what they told investigators they are actually willing to pay for.
That finding deserves a careful reading, because the report is an advocacy document built from company responses and staff interviews, not an adjudication. The companies dispute parts of it, and the report itself contains no new law, no fine, and no enforceable order.
The but-for line
The report’s answer turns on a two-word test. “But-for” cost allocation asks a simple question about any grid upgrade: would this upgrade have been built but for the data center? If the answer is no, the data center pays the full cost. If the upgrade would have been built anyway — to serve general load growth, replace aging equipment, or meet reliability needs — its cost is shared among ratepayers, as grid costs traditionally have been.
The distinction matters because most of what a data center requires falls into a gray middle. The report draws a line between two kinds of cost. The first is dedicated, customer-specific infrastructure: a substation built exclusively to serve one facility, a spur line that goes nowhere else. Every company surveyed told investigators it pays for that. Google wrote that it pays “for 100 percent of the electricity we use to power our data centers.” Microsoft said it was committed to paying “for the electricity that we consume.” Meta and Amazon offered versions of the same pledge.
The second kind is shared network infrastructure: transmission lines, substations, and generation that a data center triggers but that other customers can also use. Here, the report says, the companies’ commitments stop. In responses and interviews, the companies uniformly acknowledged they would pay some, but not all, of the upgrade costs attributable to their facilities, and they broadly opposed but-for standards that would assign them the full cost. Their argument, as the report presents it, is that upgrades which provide a system benefit should be shared among all ratepayers — even if the upgrade would not have been built but for the data center. One hyperscaler told staff that such infrastructure “often provide[s] a system benefit which should be shared among all ratepayers.”
That is a real argument, not a dodge, and it is worth stating fairly. A transmission line built to serve a data center cluster will typically remain in service for decades, carry power for other customers, and improve reliability for the surrounding system. Assigning 100 percent of its cost to the customer that triggered it can overstate that customer’s responsibility. The counterargument, pressed by consumer advocates and adopted by the senators, is that without a but-for rule, the timing and sizing of upgrades are set by one customer’s demand while the bill is spread across everyone — including households whose usage did not drive the investment.
What the report documents
The report’s most concrete illustration comes from Louisiana. Entergy Louisiana announced that its agreement with Meta for a $50 billion data center was structured so that Meta “covers its own costs” and “pays its full cost of service.” But in a case now before the Louisiana public utility commission, the report says, an independent consultant determined that a new power plant purchase was “driven primarily” by the Meta facility, while Entergy argued that no single customer was driving the acquisition and Meta called the consultant’s report “inaccurate speculation.” The consultant estimated that adding the new power would cost an average Entergy Louisiana ratepayer as much as $150 per year. Meta’s position, as the report presents it, is that the costs should be shared across the rate base. The dispute is the but-for question in miniature: everyone agrees Meta pays its own bills; the fight is over which bills are Meta’s.
The report also takes up the companion tool to but-for allocation: large load tariffs. States have begun requiring outsized users to accept minimum contract lengths, exit fees, and minimum billing demands — charges based on a fixed percentage of contracted peak demand whether or not the customer uses it. Consumer advocates and independent analyses cited in the report support minimum billing demands of 80 to 90 percent, levels considered sufficient to keep a data center’s departure or underuse from stranding costs on other customers. The companies, the report says, pushed for lower thresholds. One Big Tech firm called a 90 percent monthly minimum demand charge “insanely high.” Another called 85 to 90 percent “unreasonably high,” arguing that facilities typically do not use all of their projected demand. A third said it viewed 50 percent as reasonable, but would accept 90 percent if the utility could justify it.
On transparency, the report finds that all four Big Tech companies routinely request nondisclosure agreements from utilities, landowners, commercial partners, and in some cases government officials, and that several acknowledged the goal is to limit public scrutiny. Meta told investigators that “maintaining confidentiality during project development increases efficiency and speed,” a line the report reads as excluding local residents from the relevant stakeholders. The report credits two partial concessions: Microsoft announced in January that it would no longer seek NDAs with local governments, though it would continue seeking them with state agencies and utilities; Amazon announced in early October that it would no longer seek NDAs with the “government agencies” it works with, though the report says the scope of that commitment is unclear. Google and Meta declined to make such a commitment, according to the report.
The report’s remaining findings concern tax incentives and jobs. The senators say none of the seven companies provided comprehensive quantitative evidence on full-time job creation, and that several estimated permanent staffing at roughly one job per megawatt of power demand — so a 100 MW facility using as much electricity as 100,000 households would support around 100 permanent jobs after construction. They cite Virginia’s data center tax exemption costing more than $1.9 billion in fiscal 2025, Texas’s exemption projected at $3.2 billion over the next two years, and Meta’s Louisiana project receiving tax breaks worth as much as $10 billion against an estimated 1,000 permanent jobs. Those figures are the report’s, and the companies dispute the framing: Amazon pointed to $542.9 million in Virginia property taxes its facilities contributed in 2024.
How this connects to PJM
For readers of this desk, the report’s most useful passages are the ones that tie the but-for fight to PJM’s territory. The report cites an estimate that data center demand will add $6.3 billion in charges for ratepayers in the PJM states in 2028 and 2029. It cites a Union of Concerned Scientists study finding that utilities charged ratepayers approximately $5 billion between 2022 and 2024 to fund more than 150 transmission projects “needed only to connect data centers.” And it cites Dominion Energy’s Virginia experience: roughly three-quarters of the utility’s 2025 transmission spending was directly or indirectly due to data centers, a total of $6.2 billion socialized across all ratepayers, with state PUC staff testifying that nearly 45 percent of those costs were attributable to lines “directly related to specific data centers.”
Those transmission figures are the retail and transmission-side complement to the wholesale story this desk has covered. As reported here, PJM’s market monitor calculates that data center load added $29.4 billion in capacity costs across the last four capacity auctions, and that 4,017 MW of the forecast load behind those costs has already been canceled or delayed. The capacity market dispute is about pricing a forecast; the but-for dispute is about allocating the wires and generation the load requires. Both turn on the same underlying question — how much of the cost of serving one very large, very new customer should be borne by everyone else — and both are now running in parallel.
The report also shows that the but-for standard is already being litigated inside the PJM footprint. In May 2026, the Maryland Office of People’s Counsel argued at FERC that charging a data center 100 percent of upfront network upgrade costs was necessary to deliver on the White House’s Ratepayer Protection Pledge. Pennsylvania’s Office of Consumer Advocate has recommended that a large load customer that is the but-for cause of distribution or transmission upgrades should bear 100 percent of those costs. Consumer advocates and the governor’s office have made a similar argument in a Virginia PUC case. And a utility has asked FERC to require data centers to pay for their transmission upgrades rather than spreading the cost across the rate base.
That record arrives at a moment when the federal toolkit is otherwise thin. On October 9, the same day the report was released, FERC rejected PJM’s IRAS proposal on jurisdictional grounds, while accepting the Large Load Registry. The backstop procurement designed to fill the 6,831 MW shortfall from the July auction remains suspended pending a paper hearing on cost allocation. The December auction for the 2029/2030 delivery year will model all forecast demand, because FERC rejected the demand-curve exclusion.
What a minority report can and cannot do
It is worth being precise about the report’s legal weight, because the politics around it invite overstatement. The report is the work of three senators in the minority. It proposes no legislation, creates no obligation, and binds no commission. Warren has separately called in interviews for a national moratorium on new AI data centers until developers demonstrate that full costs are covered. The report itself contains no moratorium proposal — it notes only that local communities are “fighting for moratoriums” — and no senator can impose one unilaterally. A moratorium would require legislation passed by both chambers and signed by the president, or action by the states and commissions that actually control siting and rates.
What the report can do is three things. First, it feeds future legislation. The House has already passed the Ratepayer Protection Act (H.R. 9340) by a 417–3 vote on September 16, 2026 — a bill that would require state regulators to consider, though not necessarily adopt, standards making large-load customers cover the full incremental cost of grid upgrades and provide financial assurances before construction. The bill has stalled in the Senate, where Senator Martin Heinrich blocked a unanimous-consent attempt and has countered with the GRID Savings Act, which would give FERC direct authority over large-load interconnections.
Second, the report supplies material for state proceedings. Cost allocation for transmission and distribution is overwhelmingly a state matter, and the venues where but-for standards will be won or lost are PUC dockets like the Louisiana Entergy case and the Virginia proceeding the report cites. Intervenors in those cases can now quote company statements gathered by Senate staff — the “insanely high” dismissal of a 90 percent minimum bill, the refusal to endorse but-for allocation — alongside the monitor’s PJM arithmetic. That is how a report with no legal force still moves outcomes: by putting admissions and figures into a citable, public document.
Third, it sharpens the political pressure on voluntary pledges. The White House’s Ratepayer Protection Pledge, under which hyperscalers committed to build, bring, or buy the generation their data centers need, has been the industry’s answer to the cost-shift critique. The report’s argument is that the pledge’s language — “full energy costs,” “full cost of service” — covers the dedicated-infrastructure category and leaves the shared-infrastructure question open. Whether that critique sticks depends on what the companies do next in rate cases and tariff proceedings, not on what the report says.
What to watch
Four things will determine whether the report is remembered as a turning point or a press release. The first is the FERC paper hearing on backstop cost allocation, where the but-for principle will be tested against PJM’s proposal to procure the auction shortfall. The second is the December Base Residual Auction, which will show whether the capacity market keeps clearing at its cap with all forecast demand in the curve. The third is the state dockets — Louisiana, Virginia, Ohio, Pennsylvania — where large load tariffs and but-for rules are actually written, and where the report’s most detailed evidence will be tested against utility testimony and company rebuttal.
The fourth is the Senate itself. If the Ratepayer Protection Act moves, the question will be whether the Senate strengthens its “consider” standard toward something closer to but-for allocation, or whether the bill’s state-discretion framework survives intact.
The but-for fight has been running for years in utility dockets, in PJM filings, and in state legislatures. What changed on October 9 is that it now has a Congressional record, a title, and a set of company statements that cannot be unsaid. The report does not settle who pays for the grid the AI buildout requires. It makes it harder for anyone to pretend the question has already been answered.