FERC Punted, Trump Pressed, Utilities Pushed: The AI Power Bill Now Has a Payer

The AI power story used to be a permitting story: how fast could a substation be built, how many transformers could be found, which state would say no. As of Oct. 20, 2026 it is a cost-allocation story. The three moves that day all resolved the same way. The data center ends up funding the infrastructure that connects it, and the money follows the build rather than the tariff.

Washington's retreat from writing one national rule is the most consequential part of the three-day story. FERC's pending rulemaking in Docket RM26-4-000 still asks, in its own words, whether large loads should pay the full cost of the grid upgrades needed to interconnect them and how any overpayment should be credited back. It defines a large load as demand above 20 MW. That docket has not produced a rule, and FERC has missed the Energy Department's April 30 deadline for national standards.

While FERC stalled, states and utilities wrote the operative rulebook. The clearest example is Duke Energy, which on Oct. 7 reached an agreement with the North Carolina Public Staff, Amazon, Alphabet, Meta, Microsoft, the Carolina Industrial Group for Fair Utility Rates and the U.S. Department of Defense. Data centers must pay upfront for the grid facilities that serve them alone, and post deposits and security guarantees for upgrades that benefit everyone.

Virginia ran the same experiment a step further. A State Corporation Commission order filed July 31, 2026 requires Dominion Energy to assign the cost of transmission infrastructure directly to new large-load facilities, with mandatory contributions in aid of construction covering the substations and lines that connect a data center to the grid. The case involved a proposed $1.5 billion increase to Dominion's Rider T-1 line item, and an amended allocation factor cut the typical residential customer's monthly increase from $2.90 to $0.94, a 67.5% reduction. Dominion must file a line-extension policy amendment in a new docket within 90 days, and the commission reserved the question of extending direct assignment upstream.

In 2024 alone, regulators approved about $4.4 billion of local transmission projects needed to bring data centers online across Illinois, Maryland, New Jersey, Ohio, Pennsylvania, Virginia and West Virginia, where utilities had initiated more than 150 such projects since 2022 and nearly half the money landed in Virginia. Those costs reached residential bills because PJM's transmission planning treats data-center projects as informational, and state reviews of retail rates never saw them. PJM's independent market monitor has since estimated data centers drove 63% of the price increase in the 2025/2026 capacity auction, worth about $9.3 billion to ratepayers, on an auction that cleared at a record $329.17 per MW-day.

This is the layer the AI-power trade has missed. Cost-allocation reform moves grid construction off utility rate base and onto private balance sheets: when a data center prepays for its substation, the utility recovers its cost without funding construction itself and then billing other customers for it. The flow of money tilts toward whoever gets paid for building and equipping the grid, and away from utilities whose earnings growth depended on data-center load flowing into rate base. Quanta Services carries a $53.4 billion backlog after $9.56 billion of second-quarter revenue, up 41% year over year; Vertiv grew revenue 24% to $3.27 billion at a 37.7% gross margin; GE Vernova posted $11.10 billion of second-quarter revenue across power, wind and electrification.

The strain lands where the build is biggest. Dominion Energy spends 2.27 times its operating cash flow on capital, Entergy 1.52 times and Duke Energy 1.37 times. Amazon is the only hyperscaler here running negative trailing free cash flow, at negative $11.6 billion, against Alphabet at positive $53.3 billion. Utilities in the middle of the largest data-center build-outs cannot fund it from internal cash, which is why they now insist the customer pays upfront and why the tariff template is a financing tool rather than a fairness gesture.

There are two clocks running, and they are not the same clock. The docket calendar decides who pays in a specific state, in a specific month. The load forecast decides how much power is being repriced over a decade. Investors should track both, because the first is what changes utility rate cases and project economics, while the second is what changes the multiple on the equipment suppliers.

Three things could break the cost-allocation conclusion, and each is checkable. The evidence base is not neutral: the study utilities lean on was commissioned by their trade group, and PJM's independent market monitor reaches the opposite conclusion. The dollar amounts behind the biggest template are not public. And most of the pipeline is announced rather than contracted.