Everyone Agrees Data Centers Should Pay. Washington Won’t Make Them.

Last month some answers landed to the question I have been tracking all year: who pays for the power the AI buildout demands? The White House expanded its Ratepayer Protection Pledge to more than 200 additional utilities, developers, cooperatives, and states, until it covered roughly 80% of the power delivered to American homes and businesses.1 A House committee advanced the Ratepayer Protection Act by unanimous vote.2 There’s real bipartisan support behind both instruments, yet neither one has defined how ratepayers will be protected. What they do instead is defer to the states, to the utilities, to the rate case hearings most people never hear about.

We heard about the pledge the first time it was announced during the State of the Union. I noted then, like now, how data center deals are opaque and lack transparency. The mechanics have not changed much. The pledge is voluntary and carries no penalty or enforcement. Only a public utility commission or the Federal Energy Regulatory Commission can turn a rate into an obligation. The Ratepayer Protection Act goes further. It amends the Public Utility Regulatory Policies Act so that every state regulator must consider a standard recovering the full incremental cost of serving a large load. Consider. That is the operative word. Let’s examine the federal regulatory authority and what it does.

In June, FERC issued show-cause orders to all six of the nation’s regional grid operators — the markets that move roughly two-thirds of the country’s electricity — directing each to justify its rules for connecting large loads or reform them.3 Not one national standard has been set. With six separate proceedings, each market was told to make its own case, with responses due this month. Even the federal regulator with the clearest authority to decide on cost-allocation chose to defer to the states. Decision power sits with the states — away from Washington. Here is the part of the story that gets lost in the noise about federal gridlock: the states have not been waiting.

Share

Oregon showed the way first. Its POWER Act, signed in 2025, directed regulators to build a separate rate class for the largest energy users — the data centers drawing twenty megawatts or more — on a simple principle: the customer driving added cost should carry it.4 Assign the cost correctly, and everyone else’s bill can come down.

Oregon did not stay an outlier for long. By this summer, roughly two dozen states had at least one large-load tariff on the books, with more pending.5 The instruments differ, and revealingly so — FERC would define a large load at 50 MW, Oregon draws the line at 20 MW, Texas at 75 MW, the House bill at 100 MW. We have not yet agreed as a country on what even counts as a large load, let alone what it should pay. But the direction is unmistakable.

Ohio’s regulators approved a tariff requiring data centers above 25 MW to pay for at least 85% of the capacity they subscribe to, for as long as twelve years, whether they use it or not.6 New Jersey enacted a Data Center Fair Share Act creating a dedicated ratepayer class for the largest facilities.7 Virginia enacted the first per-kilowatt-hour tax on data-center electricity in the country, which notably does not offset ratepayer costs but does generate tax revenue for the state.8 Four states, four different tools, one shared conviction. The costs are real, and addressing cost allocation has meant pursuing state-by-state solutions rather than a unified national agenda.

The question of who pays to build the grid is not new. When the country set out to electrify rural America in the 1930s, we decided — deliberately, through public policy — that the cost of running a line to a distant farmhouse should not fall on that farmer alone. We built cooperatives and federal loan programs to spread it.9 Cost allocation has always been political. Pennsylvania nearly did this first. Governor Gifford Pinchot’s Giant Power plan proposed cooperative distribution in the mid-1920s; the legislature rejected it, and the man who designed it went to Washington and built the federal program instead.10 What is new is the scale, the speed, and who stands to benefit.

The cost is not theoretical, and it is not small. PJM’s independent market monitor attributes $29.4 billion in capacity charges across its last four base auctions to data-center demand — 46% of every dollar of capacity cost in that period, in a market serving thirteen states and the District of Columbia.11 That lands, eventually, on ordinary electricity bills. The average American household is on track to pay around $792 for electricity this summer alone, up 10.5% from a year ago and the steepest such increase in at least a decade.12 Numbers like that do not stay abstract. They arrive on a bill that a family cannot explain and did not know was coming.

The capital flowing into these projects, meanwhile, flows to private developers who keep the asset and the returns. The cost of the grid that investment requires gets spread across every household’s bill. That split is not a law of nature. It is a choice of instruments, and it holds only until a state or a utility writes something binding that reassigns it. This is the state of play on energy cost allocation for data centers.

This is also how trust erodes — not in a single dramatic act, but in the quiet arrival of a cost no one will claim. When a household’s bill climbs and every institution involved can point somewhere else — the utility to the regulator, the regulator to the market, the market to the legislature — people stop believing the system is being run on their behalf. Naming who pays is not merely an accounting exercise. It is how a government keeps faith with the people it serves.

Leave a comment

Which brings me home to Pennsylvania. No state has more riding on this. More than $90 billion in private investment in energy and AI-related projects has been announced for the Commonwealth, and Pennsylvania sits inside PJM — the regional grid operator now under FERC’s order to show its work.13 The one binding protection Pennsylvania households have today did not come from the legislature. It came from a rate case: the PPL settlement that created the state’s first large-load rate class and, in the same proceeding, raised residential rates more than 3% this summer. Beginning in 2027, those large-load customers, data centers among them, will fund low-income assistance for households.14

It would be wrong to say Harrisburg has done nothing. The Pennsylvania House has voted for data center accountability repeatedly, and often overwhelmingly. It passed Matzie’s House Bill 1834, which would have the Public Utility Commission keep data-center connection costs off ratepayers’ bills, back in March.15 It passed the governor’s GRID standards as House Bill 2650 in June, with roughly a third of the Republican caucus voting yes.16 That bill conditions Pennsylvania’s data center tax break on meeting the governor’s standards — a meaningful lever, given that the exemption is projected to cost the Commonwealth roughly $2 billion by 2031. It governs who gets public money. However, it does not decide who pays the power bill. The House also voted to let municipalities pause data center applications for 180 days with a single dissenting vote, and to repeal the sales-tax exemption.17 The disagreement in Pennsylvania is not really between the parties. It is between the chambers. None of those bills has reached the governor’s desk, and HB 1834 has been sitting in the Senate for months now.

So the voluntary pledge still points to Harrisburg, the federal bill points to Harrisburg, and even FERC's process runs through PJM and back to Harrisburg. And Harrisburg, so far, has taken one action. Folded into the 2026-27 state budget, a single measure became law: facilities using more than 10 MW must now file an annual report on how much energy and water they consumed. That is the entire body of Pennsylvania statute on data centers.18

On July 22, Governor Shapiro commented on the action he will pursue. “I am now exploring everything I can through executive authority,” he told reporters, “since the Senate Republicans refused to act.” I take him at his word, and I understand the impulse. I think one prediction feels safe: Shapiro is not going to issue a moratorium. He has been too committed to that investment — the $90 billion and the jobs behind it — to slam the door.

Pennsylvania already said yes to the buildout without deciding how to regulate it. New York met the same stalled-statute problem and its governor answered with an executive order rather than wait for a bill that wasn’t moving. An order can hold the line while the Senate decides, but it can’t create a rate class, and it doesn’t survive the next governor. Both Governors Shapiro and Hochul are up for reelection in November. A binding answer does not have to choose between the investment and the ratepayer. It can welcome the $90 billion and still decide, plainly, that the load driving the new costs will carry them — the way Oregon did, the way New Jersey did, the way a utility rate case in Pennsylvania has already begun to.

I launched this platform with a reflection on my disbelief as Hurricane Ida turned a Philadelphia expressway into a river — about what decades of deferred infrastructure decisions look like when the storm finally arrives. The stakes are the same here. And the Senate returns at the end of September — just in time for peak hurricane season.