King Charles Summons AI CEOs After Trump's Pushback
On September 16, the House passed the Ratepayer Protection Act by a vote of 417 to 3. Four hundred seventeen to three. In a Congress that canât agree on what day it is, thatâs not a bipartisan vote â thatâs a near-unanimous one, the kind you get maybe once a year on anything with âAIâ attached to it. The bill would push state regulators to make data centers pulling 100+ megawatts off the grid pay for the power plants and transmission lines their own demand requires, instead of spreading that cost across everyone elseâs electric bill.
Then, on September 17, the whole thing ran straight into the Senate and died. Not the bill itself â a Republican senatorâs attempt to fast-track it, followed immediately by a Democratic senatorâs attempt to fast-track something tougher. Both got blocked, on the floor, within minutes of each other. Congress managed to agree almost unanimously that AI data centers shouldnât get a free ride on your power bill, and then, the very next day, proved it still canât get out of its own way to actually do anything about it.
Quick Summary: What Happened
Detail Info House vote 417-3, September 16, 2026 What it does Pushes state regulators to make 100+ MW data centers cover their own grid-upgrade costs Who voted no Reps. Summer Lee, Delia Ramirez, and Rashida Tlaib â all Democrats Senate fast-track attempt Blocked September 17 by Sen. Martin Heinrich (D-NM) Heinrichâs alternative The GRID Savings Act, blocked in turn by Sen. Bernie Moreno (R-OH) Official source H.R. 9340, Ratepayer Protection Act Bottom line: Congress found near-total agreement that AI shouldnât be allowed to quietly raise your electric bill, and then, one day later, proved that agreement doesnât survive contact with the Senate.
Start with what the bill actually does, because âmake AI pay for its own powerâ is the pitch, not the mechanism. H.R. 9340, sponsored by Rep. Gabe Evans (R-CO) and Rep. Kathy Castor (D-FL), amends the 1978 Public Utility Regulatory Policies Act to push state utility commissions toward requiring âlarge loadâ customers â facilities drawing 100 megawatts or more, which is squarely where a modern AI data center campus lives â to cover the full incremental cost of the generation, transmission, and distribution upgrades built to serve them. States get a year to start the process and two years to finish it. Large loads have to post financial assurances before utilities break ground, and if a data center bails on its power contract early, the bill guarantees the utility still gets made whole. Rep. Evans put it plainly: âHardworking families should not have to subsidize the energy demands of data centers.â Castorâs version of the same point: âIf Big Tech builds a massive AI data center, Big Tech â not hardworking families, farmers or small businesses â should pay for the power and grid upgrades it requires.â
That message cleared the House Energy and Commerce Committee 52-0 back in July, then cleared the full floor 417-3 in September. The three no votes â Reps. Summer Lee, Delia Ramirez, and Rashida Tlaib â werenât objecting to the idea. Tlaibâs complaint was that the bill doesnât go far enough: it only requires state regulators to consider adopting cost-recovery rules, not actually adopt them, which she argued would âfail to meaningfully protect our communities.â Thatâs a real distinction, and itâs the exact distinction that ended up sinking the billâs fast-track shot in the Senate a day later.
The backdrop here isnât abstract. Nationally, electricity prices have climbed roughly 27% since 2019 â well ahead of general inflation â and in the states where data centers have clustered hardest, bills are rising faster still: Virginia up 13% in a single year, Illinois up 16%, Ohio up 12%. Data center construction rose 57% between July 2025 and July 2026. None of that is a coincidence a state regulator has to squint to see.
Line up the two bills and the disagreement isnât really about whether AI data centers should pay for their own grid upgrades â everyone involved, House and Senate, both parties, agrees on that part. Itâs about how hard the requirement bites. The House bill nudges states to act over a one-to-two-year runway and stops short of forcing their hand. Heinrichâs GRID Savings Act skips the state layer entirely and puts FERC in charge of one binding national rule. âIf a company is driving the need for expensive new facilities, that company should pay for it â not families and small businesses,â Heinrich said when he introduced it. Thatâs not a different goal than the House bill. Itâs a different amount of trust in whether state regulators will actually use the discretion Congress hands them, and Tlaibâs three-vote objection in the House was the same bet made from the other chamber.
Hustedâs political situation adds a layer worth being honest about. Heâs in a competitive Ohio reelection fight where data center costs are already an attack line, which is exactly why he wanted a clean, fast, bipartisan-looking win before it became a floor fight. Heinrich answering with âonly if itâs my stronger bill,â followed by Moreno tanking that bill on the very next request, turned a policy disagreement into something closer to the Senateâs ordinary reflex against letting the other side claim credit.
This also lands inside a broader story this site has tracked about who gets to set AIâs pace. Anthropicâs Dario Amodei has pushed labs to voluntarily slow down; Sam Altman told OpenAI staff heâs open to the same idea, then had to ask Congress if coordinating on it with rivals is even legal; Trump has dismissed the entire pacing conversation as a âsick conspiracy.â Those stories are all about whether labs should slow model releases. This oneâs different â itâs about who eats the electric bill while AI keeps moving â and itâs the first version of that question to get an answer nearly the whole House agreed on, only for the Senate to prove even the least controversial AI-adjacent idea in Washington isnât immune to normal gridlock.
If youâre a residential ratepayer in a state with heavy data center buildout (Virginia, Illinois, Ohio, and similar high-concentration states are the ones to watch), nothing changes on your bill yet. Both the House bill and Heinrichâs alternative are stalled, and states arenât required to act on either. Your state utility commissionâs public comment process â not Congress â is currently where this actually gets decided, and most of those proceedings are open to public input.
If youâre evaluating where to site or build an AI data center, treat âlarge load cost recoveryâ as a when, not an if, regardless of what happens in Washington. Roughly 18 states already have purpose-built data center tariffs in place or approved, with a median threshold around 50 megawatts â well below either federal proposalâs floor. Federal inaction doesnât mean state-level cost exposure isnât already here.
If youâre tracking AI policy for enterprise compliance or government-affairs reasons, watch the state commissions and FERC dockets more closely than the Senate floor for now. With the Senate leaving Washington in roughly two weeks and midterms looming in November, neither bill is likely to move again this session â the Houseâs near-unanimous vote was the easy part.
The uncomfortable read here isnât that Washington disagrees about AI and electricity costs. Itâs that it doesnât, really â and gridlocked anyway. A 417-3 House vote is about as close to consensus as modern Congress produces on anything, let alone something touching both Big Tech and utility regulation. The Senate had two separate chances, roughly ten minutes apart, to advance some version of the same basic idea, and both attempts died to single-senator objections that were more about process, timing, and an Ohio Senate race than about the substance either side claims to care about.
Thatâs a pattern this site has now documented across a few different flavors of AI governance this year: strong agreement on the problem, followed by breakdown over whose version of the fix gets to pass. It happened with lab pacing. Itâs happening again here, just with electric bills instead of model releases. The mechanism for making AI data centers pay their own way clearly exists â multiple versions of it, in fact, sitting in the same building on the same day. Whatâs missing isnât consensus. Itâs a Senate willing to let either version through without extracting something else first.
We think the 417-3 vote is the real story and the Senate collapse is the predictable coda, not the other way around. Itâs genuinely rare for Congress to agree this cleanly that AI infrastructure shouldnât get subsidized by people whoâll never use it â thatâs worth taking seriously as a signal, even with the underlying billâs soft âconsiderâ language doing a lot of the compromise that made 417 votes possible in the first place.
Heinrichâs objection is the more interesting substantive argument of the two, separate from how the Senate floor fight actually played out. âStates should consider adopting cost-recovery rulesâ and âFERC sets one binding national ruleâ produce meaningfully different outcomes for the same stated goal, and reasonable people can land on either side of state discretion versus federal mandate here. Whatâs harder to defend is Moreno tanking the stronger version in the same breath Husted was blocked on the weaker one â thatâs not two competing visions of good policy losing to each other. Thatâs neither version getting a real vote.
For ratepayers, the practical signal is this: donât wait on Congress. State utility commission dockets are where the actual cost allocation gets decided either way, federal bill or not, and that process is already underway in the states carrying the heaviest data center load.
H.R. 9340 pushes state utility regulators to require data centers and other âlarge loadâ customers drawing 100+ megawatts to cover the full incremental cost of generation, transmission, and distribution upgrades built to serve them, rather than spreading those costs to residential ratepayers. States have one year to begin the process and two years to finalize rules, and large loads must post financial assurances before utilities start construction.
The bill drew broad bipartisan support because it addresses a concrete, visible problem â rising electric bills in states with heavy data center buildout â without mandating a specific outcome; it only directs state regulators to consider adopting cost-recovery standards, which kept objections to a minimum. The three no votes, from Reps. Summer Lee, Delia Ramirez, and Rashida Tlaib, came from the opposite direction: they argued the billâs âconsiderâ language is too weak to guarantee results.
Sen. Jon Husted (R-OH) tried to pass the House bill through unanimous consent on September 17, but Sen. Martin Heinrich (D-NM) objected, arguing it lacked enforcement teeth. Heinrich then sought unanimous consent for his own, stronger bill, the GRID Savings Act, which Sen. Bernie Moreno (R-OH) blocked in turn. Both attempts died within the same floor session.
Sen. Martin Heinrichâs GRID Savings Act, introduced in August 2026, would give the Federal Energy Regulatory Commission direct rulemaking authority over facilities drawing 150 megawatts or more, requiring them to absorb grid-upgrade costs under one binding national standard rather than leaving it to individual states.
Not directly, not yet. Both federal bills are stalled, and neither has changed a single utilityâs rate structure. But roughly 18 states already have their own data center-specific electricity tariffs in place or pending, with thresholds well below either federal proposal, so cost allocation may already be moving in your state regardless of what Congress does.
Unlikely. With the Senate scheduled to leave Washington within about two weeks of the September 17 floor fight and midterm elections in November, neither the House bill nor Heinrichâs alternative has an obvious path to a floor vote before the current sessionâs window closes.
The House billâs threshold is 100 megawatts of demand; Heinrichâs competing bill sets it at 150 megawatts. Both thresholds are designed to capture large hyperscale AI training and inference campuses â the kind Amazon, Microsoft, Google, Meta, and OpenAI-aligned partners have been building at record pace â while leaving smaller commercial and industrial facilities out of scope.
Data center construction is accelerating fast enough to strain grid capacity in concentrated regions, which is the direct cause of the rate pressure this bill responds to. Itâs the electricity-cost side of the same infrastructure boom covered in Amazonâs $200 billion AI infrastructure commitment and the chip supply deals, like Qualcommâs $60 billion agreement with AWS, feeding it.
Last updated: September 19, 2026. Sources: H.R. 9340 â Ratepayer Protection Act ¡ Rep. Gabe Evans â press release ¡ Utility Dive â House passes ratepayer protection bill ¡ Yahoo News â Senate Democrat blocks Ratepayer Protection Act ¡ Sen. Heinrich â GRID Savings Act introduction ¡ Latitude Media â One Senate Democratâs answer to Trumpâs ratepayer protection pledge ¡ CNBC â Whatâs happening to electricity prices in data center states.
Related reading: Dario Amodei Told AI Labs to Slow Down. OpenAI Blinked. ¡ Altman Wants to Slow AI Down. Antitrust Says Maybe Not ¡ Trump Calls AI Slowdown Warnings a âSick Conspiracyâ ¡ Amazon Bets $200B on AI: What Changes for Users ¡ Qualcommâs $60B AWS Deal Cracks Nvidiaâs AI Grip