The House Voted 417 to 3 to Make Data Centers Pay for the Grid. The Bill Only Asks States to Consider It.

On Wednesday, September 16, 2026, the House passed H.R. 9340, the Ratepayer Protection Act, by a vote of 417 to 3. It moved under suspension of the rules, which requires two thirds of members present, and it cleared that bar with roughly 180 votes to spare. In a Congress that has spent the year unable to agree on what an AI model even is, 417 members agreed that a hyperscaler should pay for the transmission line it causes.

I have spent two days reading the text against what the states have already done, and the honest summary is that the headline and the mechanism point in different directions. The headline says Congress made data centers pay. The mechanism says Congress asked fifty public utility commissions to think about it, on no schedule, with no penalty for declining. Both of those are true at once, and the interesting part is which one actually changes a siting decision in 2027.

What the Bill Does

H.R. 9340 was introduced in June by Rep. Gabe Evans, Republican of Colorado, and Rep. Kathy Castor, Democrat of Florida. It amends the Public Utility Regulatory Policies Act of 1978 by adding a new federal ratemaking standard: customers with peak demand of 100 megawatts or more at a single site should bear the full incremental cost of the transmission, distribution, and generation capacity required to serve them, rather than spreading that cost across the residential and commercial rate base.

That is a clean statement of cost causation, and as policy it is close to unobjectionable. The problem is the verb PURPA uses. A PURPA standard is a consideration requirement. State commissions must take the federal standard up, hold a proceeding, and decide. They are not required to adopt it. They are not required to adopt anything resembling it. The statute has worked this way since 1978, through every standard Congress has bolted onto it, and the historical record on adoption rates is uneven at best.

The three members who voted no said exactly this. Reps. Summer Lee of Pennsylvania, Delia Ramirez of Illinois, and Rashida Tlaib of Michigan opposed the bill on the grounds that state discretion would fail to protect the communities actually absorbing the load growth. Whatever you think of the politics, the objection is technically correct about the mechanism. The bill directs states to consider a cost-allocation standard. It does not impose one, and it attaches no deadline to the consideration.

The States Already Did This

Here is the fact that reframes the whole vote. The Edison Electric Institute's large-load tracker had 23 states with at least one approved large-load tariff as of May 2026, with another seven pending. The federal government is not opening this file. It is arriving after the market moved, which is the normal sequence for utility regulation and the abnormal sequence for anything with the word AI attached.

Look at what Ohio approved. The AEP Ohio data center tariff, cleared by the Public Utilities Commission of Ohio in 2025, triggers at a monthly peak of 25 megawatts. It carries an 85 percent minimum demand charge, a 12-year contract term, exit penalties, and a load study fee running up to $100,000 for a 100 megawatt facility. Analysts modeling a 100 megawatt build under that tariff put first-year incremental cost near $10 million before a single token is served. Virginia, the largest data center market in the country, has its own approved large-load tariff through the State Corporation Commission.

Now put the federal standard next to it.

The federal trigger sits four times higher than Ohio's. A 60 megawatt inference site is already captured by the Ohio tariff and falls entirely outside the standard the House just passed. If a state with no tariff today adopts the federal standard verbatim, it produces a regime looser than the one already operating in a major Midwest market. That is not a gotcha, because the bill sets a floor rather than a ceiling and nothing in it preempts a stricter state rule. But it does mean the bill cannot be read as the country tightening. In the states that matter most for compute siting, it is the country catching up.

The Senate Math

S. 5028, the Senate companion, was introduced in July by Sen. Jon Husted, Republican of Ohio, and has not moved out of committee. The consensus read from energy analysts is that it does not pass before the midterms. A 417 to 3 House vote does not create Senate floor time, and Senate floor time in an election year is the scarcest commodity in Washington.

So the realistic near-term outcome is that H.R. 9340 sits as a passed-one-chamber bill for months, gets cited in campaign literature in roughly forty districts, and changes nothing about how a kilowatt gets billed. Anyone modeling 2027 power costs off this vote is modeling the wrong instrument. Model the state tariff docket in your interconnection queue.

The Voluntary Layer Underneath

There is a third track running parallel, and it is worth pricing because it moved first. The White House Ratepayer Protection Pledge launched in March 2026 with Amazon, Google, Meta, Microsoft, OpenAI, Oracle, and xAI signing a commitment to build, bring, or buy new generation and to cover the delivery infrastructure their loads require. In July it expanded to more than 200 additional utilities, cooperatives, developers, and state governments, including NextEra and Duke, with four governors signing on stage. The administration puts the pledge's reach at roughly 80 percent of electricity delivered to US homes and businesses.

A pledge is a pledge. It has no adjudication, no penalty, and no audited number behind the 80 percent coverage claim. But it means the largest buyers of grid capacity in the country publicly conceded the cost-causation principle six months before the House voted on it. That is why the vote was 417 to 3 and not 250 to 170. Nobody was defending the other side. The industry had already stopped arguing that ratepayers should carry the interconnection bill, because arguing that was costing more in local permitting fights than the transmission upgrade costs in the first place.

The Three Tracks

Read down that column and the federal bill is the last entry, not the first. The tariffs came before the pledge, the pledge came before the bill, and the bill asks states to consider doing what twenty-three of them finished doing. Congress is ratifying a settlement, not negotiating one.

Our Take

The number everyone will quote is 417 to 3, and it is genuinely the most important number in the file, just not for the reason the coverage says. It is not important because it creates an obligation, because it does not. It is important because it is a price discovery event on the politics of a gigawatt.

Four hundred and seventeen members of Congress decided that the safest available position on data center power costs was to vote against the data centers. Not one Republican broke for the industry. The three defections came from the left and were complaints that the bill was too weak. There is no constituency in either party for the position that residential ratepayers should subsidize an inference cluster, and every member of the House now has that vote on the record to point at in a town hall.

That is the thing that actually repriced on Wednesday. Not the cost allocation, which the state commissions were already handling and which this bill does not compel. The political cost of an interconnection request. If you are siting compute in 2027, the binding constraint was never going to be a PURPA standard. It was going to be a county board, a rate case intervention, and a local news segment about a power bill, and the House just told every one of those actors that Washington has no objection to them winning.

The practical read for anyone building: stop tracking the federal bill and start tracking your utility's tariff docket. The threshold that governs your site is 25 megawatts in Ohio, whatever the SCC settled on in Virginia, and an open question in the seven states with pending filings. The 100 megawatt federal number is a floor nobody has to stand on. Price the 85 percent minimum demand charge and the twelve-year term, because those are real dollars on a real contract, and the federal standard is a press release with a consideration requirement stapled to it.

Three signposts for the next 60 days. First, whether S. 5028 gets a committee markup before the midterm recess, which is the only signal that the Senate treats this as legislation rather than messaging. Second, whether any state commission opens a PURPA consideration docket citing H.R. 9340 by name before it has passed the Senate, which would tell us the House vote carries persuasive weight without statutory force. Third, whether any of the seven pending state tariffs lands at 100 megawatts rather than at the 25 to 50 megawatt band the approved ones cluster in, because a federal number becoming a state anchor is the one way this bill makes the regime looser instead of tighter.

We track provider infrastructure and capacity signals on the live status board and model serving costs on the cost calculator. Power is the line item that does not show up in either one yet, and that is increasingly the line item that decides where the next cluster gets built.