The House votes 417-3 to make data centers pay for their own power
Washington's answer to the data-center backlash landed Wednesday night, and the margin is the story — the AI buildout now has its first piece of federal cost legislation.
The U.S. House passed the Ratepayer Protection Act 417 to 3, requiring state utility regulators to consider whether large electricity users, data centers among them, should bear the incremental cost of the power infrastructure built to serve them. The bill, introduced by Republican Rep. Gabe Evans of Colorado, is the first legislation the chamber has advanced that speaks directly to the economic fallout of the AI boom, and Republican leaders moved it before lawmakers left Washington ahead of the November 3 midterms. Speaker Mike Johnson framed it as a guarantee that "hardworking American families won't foot the bill for the buildout of AI infrastructure." It now goes to the Senate.
What the bill actually does is narrow. The text asks regulators to consider charging large-load customers for new generation and interconnection — no mandate, no rate formula, no enforcement mechanism. Public Citizen's energy program director Tyson Slocum said that gives it "very limited ability to actually protect consumers from higher utility rates," while the Energy and Commerce Committee's ranking Democrat, Frank Pallone, called the measure "imperfect" and "only one small part of the puzzle." The political context is less ambiguous than the text. President Trump said this week that data centers make people and states wealthy and called them "the oil of the next 20, 25 years," while a University of Massachusetts Amherst poll published days earlier found just 11% of Americans would support construction of one in their community.
Our read: a 417-3 margin on an AI-infrastructure bill tells you the pass-through model is politically finished — but "consider" is a message, not a brake, and the interesting number is the three. Voting against ratepayer protection is running on the cheapest attack ad in American politics this cycle, so the Senate now inherits the harder question: whether it strengthens the only data-center cost bill that can actually pass. We tracked how that arithmetic reaches the ballot box — Pennsylvania voters move AI data centers to the top of the ballot.
The Basel Action Network put a number on AI's hardware tail: 395 to 617 million metric tons of AI-related electronic equipment retired between 2025 and 2050, with global e-waste tripling to as much as 211 million metric tons a year by 2050 and 15 to 20 percent of it attributed to AI. The analysis, the first part of BAN's "The Coming AI Waste Wave," works out to roughly 70,000 metric tons of e-waste per gigawatt of data-center capacity, and it deliberately counts the categories earlier studies skipped — power supply and distribution, cooling systems, backup power and networking gear, plus a catch-all BAN calls "AI Waste Contagion" for telecom infrastructure and personal devices pushed into early retirement. BAN's claim is that estimates built on servers and accelerators alone miss about 87 percent of a data center's electro-mechanical infrastructure.
The counterweight is peer-reviewed and much narrower. A study published in April by Alex de Vries-Gao in Resources, Conservation and Recycling derives its range from actual chip-production capacity and server-lifespan assumptions rather than projected computing demand, and lands on 131,000 to 224,800 metric tons of AI-server e-waste a year by 2030. The two figures are not measuring the same thing, which is the point: one counts everything replaced because AI exists, the other counts only the servers themselves.
Our read: the multiplier argument cannot be settled with public data, because no operator publishes an inventory of what it retires or where it goes. What every estimate agrees on is the shape of the curve — AI hardware turns over on a two-to-four-year cycle, faster than any previous computing buildout, and the field's own supply-side study says the precision problem is missing public information about installed hardware and replacement cycles. Expect the industry's answer to be a procurement standard rather than a reuse commitment. The same buildout's fuel side is already quantifiable — AI data centers will out-burn Germany and Japan on gas by 2035.
What to watch: whether the Senate keeps the word "consider" or replaces it with a cost-allocation standard, and whether any hyperscaler publishes retired-hardware volumes for its own fleet.
If a bill that only asks regulators to "consider" the cost passes 417-3, what would an actual mandate get? Tell us in the comments.
Sources: Reuters — US House advances first bill addressing economic impact of data center boom · CBS News — House passes bill to shield Americans from data center energy costs · Politico — House passes GOP-backed bill targeting data center power costs · The Verge — The AI data center e-waste problem is huge — and getting bigger · Basel Action Network — AI E-Waste May be 40 to 60 Times What Has Been Estimated · Basel Action Network — The Coming AI Waste Wave, Part 1 (PDF) · Let's Data Science — Study recalibrates AI server e-waste estimates
