Late last week, the Senate moved to compel data centres to disclose their electricity consumption, a step that has been debated for months but now has legislative momentum. The bill is narrow — it asks for transparency, not caps — but it marks a shift in how Washington is thinking about the energy footprint of the AI buildout. The question of who pays for it has been building quietly underneath the capital expenditure announcements for more than a year.
The scale of the problem is not in dispute. Data centre electricity demand is projected to nearly triple by 2035 according to BloombergNEF, and Goldman Sachs estimates it could surge 175 percent by 2030 from 2023 levels. In the short term, natural gas plants are carrying most of the new load. After a December auction by PJM Interconnection — the grid covering much of the eastern United States — costs attributable to data centres totalled $23.1 billion for the three-year period from June 2025 through May 2028. That is 49 percent of the total auction cost. The grid is not failing under this pressure, but it is visibly strained.
The cost question is more contentious than the consumption question. Utility commissions in multiple states are now facing a version of the same argument: tech companies want access to large quantities of power at competitive rates, while residential and small commercial customers worry they will absorb the infrastructure costs needed to deliver it. A CNBC investigation earlier this year found that the debate over ratepayer protection has become one of the most active areas of utility regulation in the country. A coalition that includes Bernie Sanders and Ron DeSantis — about as ideologically mismatched as a coalition can get — has spoken against the data centre boom's effect on electricity prices, a signal that the backlash is not confined to any particular political lane.
The Senate disclosure bill is a response to a genuine information gap. Utilities and grid operators have struggled to forecast actual power demand from data centres because the companies involved are not required to report what they expect to consume, and the gap between projected and actual demand has made capacity planning difficult. The argument for transparency is pragmatic: you cannot plan a grid for demand you cannot measure.
The tech industry's counter-argument is that data centres represent economic development, that many operators are pursuing renewable power purchase agreements aggressively, and that restricting the buildout would cede ground in the AI race to China. All three points have some merit. They also do not answer the question of who pays for the transmission infrastructure, the peaker plants, and the grid upgrades needed to serve a load that is growing faster than any single region's power sector was designed to accommodate.
Washington Post reporting from February described the buildout of private power arrangements — essentially shadow grids that large data centre operators are constructing to bypass the public utility system entirely. The practice is legal and, from the operator's perspective, rational. It also suggests that the companies with the resources to self-supply will do so, leaving the public grid to everyone else. The Senate bill, if it passes, would at minimum make the scale of that divergence visible. Whether visibility leads to policy is a different matter. But the period in which the energy cost of AI was treated as a background assumption rather than a political question appears to be ending.