The Invisible Energy Problem
There's a strange irony at the heart of America's AI boom: nobody actually knows how much electricity it's consuming. Not officially, anyway. No federal agency collects standardized energy data from data centers. The numbers that circulate in headlines — about doubling consumption, tripling demand by 2035 — come from estimates and voluntary disclosures, not verified reporting requirements.
Senators Elizabeth Warren and Josh Hawley want to change that, and the coalition they're assembling to do it should get the attention of anyone building or funding AI infrastructure.
What the Letter Says
On Thursday, Warren and Hawley sent a joint letter to the Energy Information Administration urging it to establish mandatory annual reporting requirements for data centers. The ask is specific: hourly, annual, and peak energy loads; the rates companies pay; details on any grid upgrades triggered by new facilities and who foots the bill; and whether data center operators participate in demand response programs that help utilities manage peak load.
Critically, the senators want the EIA to distinguish between energy consumed by AI computing tasks and energy consumed by general cloud services. That's not a minor request — it's a direct attempt to quantify how much of the electricity burden is specifically attributable to the AI buildout, rather than the data center industry broadly.
The letter gives the EIA a deadline: respond by April 9.
Why the Data Gap Matters
The absence of reliable federal data isn't just a policy inconvenience — it's a structural problem with real consequences. Utilities use data center growth projections to plan grid investment. But because data centers shop around between utilities during the planning process, multiple utilities often count the same prospective facility in their forecasts. The CEO of retail electricity company Vistra said last year that utilities may be inflating projected electricity demand by anywhere from three to five times actual need.
This "phantom growth" problem means grid planners are making expensive infrastructure decisions based on noise. Meanwhile, behind-the-meter power — where data centers install their own generation capacity entirely separate from the grid — makes the true total consumption figure even harder to calculate.
The senators are essentially arguing: before you can regulate the impact of data centers on consumer electricity bills, you need to know what they're actually using. That's a harder argument to dismiss than most energy policy debates.
The Bipartisan Pile-On
What makes this moment different from previous attempts at data center accountability is the breadth of the political coalition forming around it. Warren and Hawley sit at opposite ends of the Senate's ideological spectrum — the fact that they're signing the same letter signals this isn't a partisan issue.
The same week, Senator Bernie Sanders and Representative Alexandria Ocasio-Cortez introduced legislation proposing a full moratorium on new data center construction until Congress agrees on AI regulation. Senator Dick Durbin introduced a separate bill mandating disclosure of both energy and water use. Hawley had already co-sponsored a February bill with Senator Blumenthal requiring data centers to supply their own power rather than drawing from consumer grids.
At the state level, at least a dozen states are considering construction moratoriums, and New York has a three-year pause bill under consideration.
This is a legislative environment that is moving from general concern to specific mechanisms — and moving faster than the industry has acknowledged.
The EIA's Awkward Position
The EIA already mandates energy reporting from oil and gas companies and manufacturers. Warren and Hawley argue the same statutory authority covers data centers — a reading that Harvard Law's Environmental and Energy Law Program director Ari Peskoe says is plausible under the existing provision.
The agency took a modest step forward this week by launching a voluntary pilot program covering roughly 200 companies operating in Texas, Washington, and Virginia. But the senators' letter makes clear they view this as insufficient: voluntary and geographically limited doesn't produce the national baseline needed for real grid planning or consumer protection policy.
The EIA administrator acknowledged last December that the agency expects to play an "essential" role in data center energy analysis, while also noting that launching a new survey from scratch takes roughly two years through standard OMB channels — though faster options exist for smaller-scope surveys.
The gap between "we expect to be essential" and "we have mandatory reporting in place" is exactly the space this political pressure is designed to close.
This Puts Pressure on the Industry's Self-Regulation Argument
Earlier this month, the Trump administration convened tech executives at the White House to sign a nonbinding agreement pledging that consumers wouldn't bear the cost of data center electricity expansion. The Warren-Hawley letter explicitly calls this out: without verified consumption data, there's no way to hold companies to that pledge.
This puts pressure on hyperscalers — Microsoft, Google, Amazon, Meta — and the wave of AI infrastructure companies building out new capacity to get ahead of mandatory disclosure rather than wait for it. Google's data centers doubled their electricity consumption between 2020 and 2024. By 2035, planned new data centers are projected to nearly triple the sector's total energy demand. Those are striking numbers, but they're self-reported. Mandatory, standardized, federally collected data would make it much harder to manage the narrative.
What This Means
- For AI infrastructure companies and hyperscalers: Mandatory energy reporting is probably coming. The political coalition is bipartisan, the legal authority may already exist, and the EIA has signaled receptiveness. Companies that voluntarily exceed disclosure expectations now will be better positioned than those that wait to be compelled.
- For founders building on cloud infrastructure: Your energy costs are going to become more visible — and more politically salient. If AI compute costs are tied to electricity costs, and electricity costs become a campaign issue, the regulatory environment around where and how you build will tighten.
- For investors in AI and data center infrastructure: The moratorium bills are unlikely to pass in their current form, but they signal a floor of political will that isn't going away. Factor disclosure requirements and potential demand response obligations into infrastructure valuations.
- For grid planners and utilities: The phantom growth problem is real and the EIA pilot is a start, but the actual signal won't improve until reporting is mandatory and national in scope. The Warren-Hawley letter is, among other things, a demand for the data utilities need to plan honestly.
The data center industry has operated for years in an information environment that suited it: projections were voluntary, comparisons were difficult, and the link between AI compute and consumer electricity bills was abstract. That's changing. The question now isn't whether disclosure requirements are coming — it's how granular they'll be, and whether the industry shapes them or has them imposed.
