Congress moves to require big tech to pay for AI data-center power infrastructure upgrades.

A House subcommittee just took the first formal step toward turning Big Tech's voluntary "pay your own way" pledge into binding federal law. On June 24, 2026, the Energy and Commerce Subcommittee on Energy marked up the bipartisan Ratepayer Protection Act (H.R. 9340), a bill designed to stop AI data centers from quietly shifting their electricity costs onto household power bills.

The pledge that Microsoft and Anthropic pioneered earlier this year was always a promise. This bill is an attempt to make it a rule.

H.R. 9340 is a bipartisan federal bill, sponsored by Rep. Gabe Evans (R-CO) and Rep. Kathy Castor (D-FL), that would codify the principles behind the White House "Ratepayer Protection Pledge." It directs state utility regulators to consider forcing the largest power users—any non-residential site drawing 100 megawatts or more—to pay the full incremental cost of the grid upgrades built to serve them, rather than spreading those costs across every customer's bill.

The Subcommittee on Energy took up eight electricity and pipeline bills in a single markup on June 24, with the Ratepayer Protection Act as the headline measure. A markup is the first formal step a bill takes toward becoming law, where a committee debates and votes on the text before sending it up the chain. The significance is less about the procedure and more about the politics. This is the first time Republican leaders have rallied behind concrete legislation to address data-center-driven rate hikes, a notable shift for a party usually wary of anything that looks like cost-shifting onto private companies. Energy and Commerce Chair Brett Guthrie framed it around competition with China and protecting families in the same breath, arguing that families and small businesses should not foot the bill for infrastructure that benefits everyone.

Castor, the Democratic co-sponsor, was blunter about what is driving the bipartisan moment. She attributed Republican engagement to "populist anger" filtering up from constituents. "The public is up in arms," she said, adding that voters are wary of paying more for electricity regardless of party.

The mechanism is technical but blunt. H.R. 9340 amends Section 111(d) of the Public Utility Regulatory Policies Act of 1978, the federal law that sets the framework for how states regulate utilities. The amendment directs state regulatory authorities to consider establishing a "large-load standard" for customers drawing 100 megawatts or more.

That 100-megawatt threshold is the key design choice. It is set high enough to capture hyperscale AI data centers while leaving residential, commercial, and traditional industrial customers untouched. Once a customer crosses that line, the standard would obligate them to absorb the cost of the new generation, transmission, and local infrastructure their facility requires, rather than socializing those costs across the broader rate base.

This targets the more solvable of the two ways data centers push up bills. The first is interconnection: connecting a giant new load often requires expensive substations and transmission lines, and utilities have historically spread those costs across all ratepayers. The bill goes hard at that problem. The second—a tighter overall market lifting prices for everyone—is harder for any single rule to neutralize.

The bill did not appear from nowhere. It codifies a voluntary commitment that two companies pioneered months earlier. In January 2026, Microsoft launched its "Community-First AI Infrastructure" plan, pledging to pay electricity rates high enough to fully cover the costs its data centers create, replenish more water than it uses, create local jobs, and reject local property tax breaks. That made Microsoft the first major hyperscaler to put a comprehensive cost-recovery framework on paper.

A month later, Anthropic followed. In a February 11 blog post, the company behind Claude said AI firms should not leave American ratepayers to pick up the tab, and committed to cover 100% of the grid upgrades needed to interconnect its data centers, paid through higher monthly charges on its own bill.

Then came the formal pledge. On March 4, 2026, seven companies signed the White House Ratepayer Protection Pledge: Amazon, Google, Meta, Microsoft, OpenAI, Oracle, and xAI. Worth noting for accuracy: Anthropic was an early mover on the commitment but was not among the seven White House signatories. The pledge committed the group to a "build, bring, or buy" framework for their own power.

The throughline is a ratchet, from a single company's voluntary promise to a federal bill in roughly five months.

The political momentum is built on hard figures that voters feel directly. Power bills near data center hubs have jumped as much as 267% in five years, and roughly seven in ten Americans oppose local data centers. The story those numbers tell is simple: a cost that used to be invisible became visible, and it landed on monthly bills during an affordability squeeze. That is the political fuel behind both the federal bill and a wave of state action.

A sharp read demands taking the criticism seriously, because the pledge-to-law pipeline has real gaps. Critics of the original pledge argued it changed little in practice. Harvard's Ari Peskoe noted that neither the president nor the companies actually control who pays for grid expansions, since utilities and their regulators hold the pen on cost allocation. Consumer advocates made a related point: a voluntary pledge with no enforcement mechanism is a press release, not a protection.

The bill answers part of that critique by moving from voluntary to statutory. But it has its own soft spots. It directs state regulators to "consider" a large-load standard rather than mandating one, leaving discretion at the state level. And the corporate commitments underneath it have scope limits worth remembering: Anthropic's pledge, for example, applied to facilities it owns in a handful of states, not to the leased capacity that makes up most of its compute. A promise that covers owned infrastructure but not the bulk of leased capacity is a narrower commitment than the headlines suggest.

There is also the legislative reality. A subcommittee markup is the start of a long road, not the finish. The bill must clear the full committee, the House, and the Senate, where grid policy is tangled up in broader permitting fights that have stalled for years.

This is where the story matters for anyone tracking AI unit economics rather than just the politics. The pledge, and now the bill, formalizes a structural shift in the data center business model. Power is moving from a back-office line item to a core strategic variable. Hyperscalers and AI labs are being pushed from passive energy consumers into active grid stakeholders that co-invest in generation, deploy on-site power, and absorb delivery risk. Procurement of electrons is becoming a competitive moat alongside chips and models.

The financial impact is real but bounded. Forcing data centers to fund their own grid upgrades raises the per-megawatt cost of capacity in exactly the markets these companies most want to build in. For capital-heavy cloud models like Microsoft's and Amazon's, already pouring tens of billions into infrastructure annually, that is a marginal headwind layered on top of states pulling back the tax incentives that used to sweeten the same projects. The combination does not stop the buildout. It reprices it, shifting advantage toward states with reliable power and toward companies willing to fund their own footprint.