Congress, five states, and the White House all moved on AI data-center power costs this summer. The numbers they're citing don't measure the same thing.

In the space of five months, the US Congress, the White House, and at least a dozen state governments have all taken some action on the same underlying problem: AI data centers are pulling enough electricity off local grids to visibly move residential bills, and residents are noticing before regulators finish figuring out what to do about it. What's harder to find, reading the coverage of all of it together, is a single number anyone agrees on for how big the problem actually is — because the numbers being cited aren't measuring the same thing, over the same time period, in the same place.

That's not a minor caveat. It's the actual story. A bill that cleared a House committee 52-0 in July, a governor's executive order that sidestepped her own legislature, and a US senator's viral statistic that turned out to measure the wrong part of a power bill are all, in their own way, downstream of the same reconciliation problem: wholesale prices, retail bills, five-year windows, twenty-year windows, one city, one state, and one national average are all getting cited interchangeably as "what data centers are doing to your electric bill."

The federal push: a bipartisan bill, unanimous in committee

On July 21, the House Energy and Commerce Committee voted 52-0 to advance the Ratepayer Protection Act, H.R. 9340 — a genuinely bipartisan bill, introduced June 18 by Rep. Gabe Evans (R-Colo.) and Rep. Kathy Castor (D-Fla.). The bill would direct state utility regulators to consider a "large-load" standard: covered facilities drawing 100 megawatts or more at a single site would have to cover the full, incremental cost of any generation, transmission, or distribution upgrade their connection requires — and maintain financial assurances to cover those costs even if the facility later scales back or shuts down. It amends Section 111(d) of the 1978 Public Utility Regulatory Policies Act, giving state regulators one year to consider the standard and two years to make a determination.

The core idea is simple and, going by the unanimous committee vote, not especially controversial in principle: if a data center's connection requires the grid to expand, the data center's owner pays for that expansion, rather than spreading the cost across every residential ratepayer on the same grid — including the ones who never asked for a data center down the road. What the 52-0 vote doesn't tell you is whether the bill goes anywhere from here. It has not been scheduled for a floor vote, and it has no Senate companion.

Who supported the bill, until committee leaders changed one word

The 52-0 vote hides a real fight that happened just before it. The version of H.R. 9340 introduced in June applied to any "large load" customer of 100 megawatts or more — a category that would sweep in steel mills, hydrogen plants, and other heavy industrial users alongside data centers. That broader version had endorsements from [Microsoft](#/company/microsoft) and [Google](#/company/google). Ahead of the committee markup, leaders narrowed the bill's scope to apply specifically to data centers rather than large loads generally. The Data Center Coalition — whose members include [Amazon](#/company/amazon), Google, and Microsoft — came out against that narrowed version, having supported the original.

That reversal is easy to misread as the industry simply opposing regulation. It's more specific than that: the coalition's objection, per reporting on the committee process, was to being singled out as a named category of electricity customer rather than being treated under a general large-load standard that would apply the same cost-recovery logic to any heavy industrial user. Whether that's a principled objection to disparate treatment or a preference for a rule broad enough to be politically harder to enforce against any one industry is exactly the kind of question a committee markup transcript settles and a press release doesn't — and neither this piece nor the coverage it draws on has that transcript.

That last gap matters for a reason specific to this story: the same policy idea, introduced separately in the Senate by Sen. Jon Husted (R-Ohio) in July, currently has zero Senate cosponsors. Husted framed his version as continuous with his prior record in Ohio politics — "It's not really a shift," he told a local outlet. "I was always trying to hold these companies accountable by making them pay local property taxes... making sure that they hired local labor." Whether that framing helps him find cosponsors is, as of this writing, unresolved.

The federal effort isn't starting from nothing. In March 2026, the White House secured a Ratepayer Protection Pledge — a voluntary commitment, not a law — from seven hyperscalers: [Amazon](#/company/amazon), [Google](#/company/google), [Meta](#/company/meta), [Microsoft](#/company/microsoft), [OpenAI](#/company/openai), Oracle, and [xAI](#/company/xai). In July, the administration expanded that pledge to cover 187 additional organizations — 55 utilities, 105 electric cooperatives, and 27 data-center developers — representing what the White House describes as roughly 80% of the power delivered to American homes and businesses. H.R. 9340 is, in effect, an attempt to turn that voluntary pledge into an enforceable legal standard rather than a promise seven companies can walk back.

What "moratorium" means depends on which government you're asking

New York is the clearest case of how much these headlines flatten. On June 4, both chambers of the New York Legislature passed the Responsible Data Center Development Act (S10642/A11560) — a genuine one-year moratorium on new permits for data centers drawing 20 megawatts or more, clearing the Senate 44-16 and the Assembly 102-39. That bill went to Governor Kathy Hochul's desk. She has not signed it.

Instead, on July 14, Hochul signed Executive Order No. 62 — a different instrument entirely. EO 62 pauses discretionary state environmental permitting for data centers of 50 megawatts or more (a higher threshold than the legislature's 20 MW) for up to one year, while state agencies draft a broader environmental impact framework. Multiple outlets, including CNBC, reported this as New York becoming "the first state to impose an AI data center ban" — which is true in the sense that it's the first statewide permitting pause of its kind, but it is not the moratorium the legislature actually passed. Legal analysts tracking the bill describe Hochul as likely to veto the legislature's version by year end now that her own order is in effect.

New York and Maine aren't outliers — they're two data points in a much wider scramble

More than 300 state-level data-center bills have been filed across more than 30 states in the first six weeks they were tracked this year alone, per a legislative tracker maintained by MultiState. Most of those bills are narrower than New York's or Maine's — tax incentive changes, disclosure requirements, water-use studies — but a specific subset, roughly a dozen states by one count, are actual moratorium or building-pause bills modeled on the same idea: stop approving new capacity until the state can measure the impact. Sources differ on the exact count (11 by one tracker, 12 by another, compiled on different cutoff dates), which is itself a small illustration of the piece's larger point — even the count of how many states are doing this isn't fully settled, let alone the dollar impact.

Oklahoma is a useful example of how easy it is to overstate where these bills actually stand. SB 1488 would impose a moratorium on data centers with an electrical load over 100 megawatts until November 1, 2029 — a genuinely aggressive proposal, four years longer than Maine's vetoed 18-month version. As of the most recent legislative tracking available, the bill's status is introduced, not passed, not signed, and not law. It is easy, skimming a list of state actions, to read "Oklahoma: data center moratorium" as an equivalent data point to New York's executive order or Maine's veto. It isn't — one is a governor's signed order currently restricting real permits, one is a vetoed bill that is now dead, and one is a bill sitting in committee that may never get a floor vote at all. Treating all three as the same kind of "state action" is the same category error as treating a 267% wholesale-price move as a 267% bill increase: technically each word is doing work, but the words next to each other imply an equivalence the underlying facts don't support.

How big is the actual number — and whose number is it

This is where the loudest public argument has been the least precise. In June, Sen. Elizabeth Warren stated: "If you live near one of these large data centers, your electricity bills over the last five years have gone up by as much as 267%." The figure is real — it comes from a Bloomberg News analysis of wholesale electricity prices at grid nodes near data-center clusters, which found wholesale prices up as much as 267% between April 2020 and April 2025. It is not, however, a measurement of residential bills. Wholesale prices — what utilities pay power producers — make up roughly 30% to 50% of a typical residential bill; the rest is transmission, distribution, and taxes, none of which moved anywhere near 267% over the same window. PolitiFact rated the claim Mostly False on exactly this distinction.

The actual residential picture, drawn from EIA-sourced data over the same recent five-year stretch (March 2021 to March 2026), is still a real increase — just a much smaller one, and concentrated in the specific states where data centers have clustered hardest: Washington, D.C. up 94%, Maryland up 74%, Maine up 73%, New York up 58%, against a 42% national average over the same period.

Ohio and Texas complicate the picture further, because both are measuring over entirely different windows than the five-year figures above — and mixing them into the same comparison, without saying so, is exactly the kind of thing that produces a headline number nobody can actually defend under questioning.

Texas is the single biggest bet, and the most exposed

No state's numbers are larger than Texas's. Under Senate Bill 6, signed in June 2025, any electricity customer drawing 75 megawatts or more at a single site is now a defined "large load" subject to disclosure and cost-recovery rules; the Public Utility Commission of Texas voted in March 2026 to publish a draft interconnection rule implementing it. The scale of what that rule is trying to manage is the real story: ERCOT, the grid operator for most of Texas, is tracking approximately 410 gigawatts of large-load interconnection requests as of its most recent hearing materials — and roughly 87% of that queue is data centers. For comparison, ERCOT's entire current generating capacity is well under half that figure; the queue alone, if even a fraction of it is built, would reshape the Texas grid more than any single infrastructure project in the state's history.

That queue is why Texas's own rulemaking — not the federal bill, not New York's order — may end up being the most consequential single document in this entire fight. A rule that makes data centers pay their own interconnection costs, on a grid where 87% of pending large-load demand is data centers, sets a cost-recovery template the rest of the country is likely to watch closely, for better or worse.

Why the federal and state pushes both exist: the voluntary pledge isn't holding

The White House's March pledge was always voluntary, and the gap between a pledge and an enforceable rule is the whole reason Congress and state legislatures are now moving in parallel. Of the seven original signatories' commitments, only a handful of states have actually converted pledge-style language into binding law or regulation — Virginia and Ohio are the furthest along. Virginia's rules require data centers to sign contracts of 14 years or longer and recover 85% of transmission and distribution costs and 60% of generation costs directly from the data center rather than the general ratepayer base. Most states with active data-center legislation this year don't yet have anything that detailed; a bill that says a data center "should" cover its costs is not the same as one that specifies a contract term and a cost-recovery percentage.

The public pressure behind all of this is real and measurable in places that publish polling on it. In Wisconsin, the share of voters who say a proposed data center's costs outweigh its benefits rose from 55% to 70% over six months, according to polling cited in policy analysis of the pledge's limits. Separately, at least 48 proposed data-center projects worth a combined $156 billion were blocked or stalled by local opposition in 2025 — before most of this year's state legislative activity even began. That's the political backdrop every one of these bills, orders, and vetoes is responding to: elected officials in both parties are reading the same rising opposition, which is a large part of why a bill sponsored by a Colorado Republican and a Florida Democrat could clear committee 52-0 without becoming a partisan fight.

What's actually established, and what's still someone's word

What ties all of this together isn't a single fact but a pattern: every level of government reaching for the same lever — make the data-center owner pay the marginal cost of the grid capacity it actually uses — while none of them agree yet on how to define the threshold (20 MW? 50 MW? 75 MW? 100 MW?), who enforces it (a state utility commission? an executive order? a federal statute?), or how large the underlying problem even is once wholesale prices are separated from what actually lands on a residential bill.