'Data Centers Don't Use Water': Inside Lutnick's False Claim and the House Bill That Says Otherwise

On Wednesday, September 2, 2026, US Commerce Secretary Howard Lutnick sat down for a CNBC interview about the Trump administration’s push to accelerate AI infrastructure buildout. When the conversation turned to community backlash over data-center water consumption, his answer was categorical — and it ignited a week of fact-checks, mockery, and an awkward collision with his own department’s data.

“I mean, data centers don’t use water,” Lutnick said, laughing. “This is propaganda by our adversaries to try to slow us down.”

The claim is false, and demonstrably so. It also landed at the worst possible moment for the administration: the same week the House Energy and Commerce Committee began drafting legislation whose entire premise is that data centers use enormous quantities of water — and that local ratepayers should stop paying for it.

What the record actually says

The numbers come from the government’s own researchers. A 2024 Lawrence Berkeley National Laboratory report, cited this month by the House Energy and Commerce Committee, found that US data centers directly consumed 21 billion liters of water in 2014. By 2023, that figure had more than tripled to 66 billion liters — roughly 17.4 billion gallons — and the trajectory points steeply upward as AI training and inference clusters come online. Hyperscale facilities are projected to dominate that growth.

Per-facility numbers are harder to dismiss as rounding errors. New AI data centers consume between 1 million and 5 million gallons of water per day under peak conditions — the daily usage of a town of tens of thousands of people. A widely cited Instagram fact-check in the days after Lutnick’s appearance put the US average at around 2 million liters per facility per day. In Northern Virginia, the world’s largest data-center hub, facilities consumed nearly 2 billion gallons in 2023 alone, with Loudoun County bearing the brunt.

The mechanics are not mysterious. Most large data centers use evaporative cooling: hot water absorbs heat from servers, then passes through cooling towers where some of it evaporates away. Closed-loop and air-cooled systems exist and use far less water, but they trade water savings for higher electricity consumption — and the industry’s aggressive build schedules mean cheaper, thirstier designs still win many bids. The water is not exotic; it is the same treated municipal supply that households drink from, which is precisely why residents in data-center counties from Virginia to Arizona have organized against new projects.

The contradiction that made it worse

What turned a false claim into a genuine political story was the discovery that Lutnick himself had previously touted water — lots of it — when selling these projects to states.

As Mediaite and others documented with side-by-side clips, Lutnick in 2025 described exactly how data centers consume reservoirs’ worth of water while pitching governors on hosting facilities. In one widely circulated contrast, he is seen in 2025 explaining how a data center “sucks” water from local supplies, and in 2026 dismissing the identical statement as enemy propaganda. The Morning Herald-style montage circulated across social platforms under headlines like “Lutnick caught in 4K.”

The fact-checks piled on. MeidasTouch’s research team concluded that while water use varies by facility, “the evidence clearly contradicts Lutnick’s categorical claim.” The Hill, Yahoo News, and BBC reality-check segments all ran the Berkeley Lab figures against his statement. Even outlets sympathetic to the administration’s AI agenda treated the quote as an unforced error.

There is a defensible narrow version of Lutnick’s point — that the water footprint of AI is frequently exaggerated relative to agriculture or industry, and that adversaries do amplify American infrastructural NIMBYism. But the Secretary did not make the narrow version. He made the categorical one, on camera, while laughing.

The House moves the other direction

While Lutnick was doing damage control, Congress was moving in exactly the opposite direction — and, unusually, with bipartisan urgency.

On September 2, the House Energy and Commerce Committee opened work on legislation to ensure that data centers do not drive up water costs for local communities. The legislative principle, as Committee Republicans framed it at the hearing: developers should cover the cost of their water. Draft provisions would require facilities using more than 200,000 gallons of water per day (averaged over 30 days) to report consumption and, critically, would bar utilities from passing data-center water costs through to residential ratepayers.

The committee’s framing memo was blunt: “U.S. data centers in 2014 directly consumed 21 billion liters of water. By 2023, total water use had grown to 66 billion liters.”

This is not a fringe concern. The Congressional Research Service’s July 2026 report “Data Centers and Water: Frequently Asked Questions” (R49057) notes that the House-introduced Water Resources Development Act of 2026 (H.R. 9497) would amend federal water-supply purposes to account for data-center demand. A separate bill, H.R. 9629, would require assessments of the environmental and public-health effects of data centers. State-level transparency bills — like the one Senator introduced in March requiring operators to report energy and water consumption to host states — have been accumulating all year, tracked by initiatives like datacentertracker.org.

And the industry itself is split. In the most surprising twist of the week, Microsoft publicly backed the water-cost bill. The company, which has faced sustained criticism over its own data-center water usage in Arizona and Wisconsin, endorsed the principle that data centers should not pass water costs to consumers. For a hyperscaler to side with ratepayer-protection legislation against its own industry’s lobbying posture signals that at least parts of Big Tech have concluded the political risk of water conflicts now exceeds the cost of paying their own bills.

Why this matters beyond one gaffe

The temptation is to file this as another entry in the administration’s fact-check ledger. But the substance underneath is one of the central unresolved questions of the AI buildout: who pays for the physical externalities of intelligence?

Electricity got the attention first — gigawatt campus announcements, nuclear restart deals, the DOE’s loan programs. Water is the quieter crisis because it is local, visible, and personally felt in a way electrons are not. A resident whose well runs low or whose utility bill ticks up does not experience “aggregate data-center water consumption.” They experience a specific facility, built by a specific company, permitted by a specific county board. That is why water, more than carbon or grid load, is the vector through which AI infrastructure meets genuine grassroots opposition — including from communities otherwise supportive of the industry.

The administration’s political problem is that “propaganda by our adversaries” does not work as a message against a Berkeley Lab report, a House committee of its own party, and Microsoft. When the Energy and Commerce Committee’s own ICYMI post contradicts the Commerce Secretary’s on-air claim within the same news cycle, the coordinating problem becomes the story.

The likely endgame is the one the House bill preview: consumption transparency mandates, cost-allocation rules that keep data-center water bills off household statements, and eventually water-efficiency requirements baked into state permitting. The UK’s data-center water debates and the Greens’ proposed moratorium in Ireland show the same pattern internationally — the industry’s social license is being renegotiated utility bill by utility bill.

Lutnick’s laugh will be a footnote. The 66 billion liters — and the question of who pays for the next 66 — is the actual news.

Lessons for the AI industry

Three takeaways are worth separating from the political noise:

1. Get the deniable claims right. There are honest arguments for why AI water use is manageable: cooling technology is improving, hyperscalers have committed to water-positive targets, and per-query water footprints are small. A Commerce Secretary saying “data centers don’t use water” forfeits all of them in seven words.
2. Cost allocation is the new frontier of AI policy. The electricity fights of 2024–2025 established that ratepayers should not fund stranded grid capacity. The water fights of 2026 are establishing the same principle for municipal supplies. Companies that get ahead of this — as Microsoft just did — buy political capital cheaply.
3. Local opposition is now the binding constraint. Between the $265M midterm ad campaigns, the Pentagon’s reprogramming fights, and county-level water revolts, the AI buildout’s chokepoint is no longer chips or capital. It is community consent, and consent is bought with accurate numbers.

The secretary’s office did not respond to requests for clarification by press time. The Berkeley Lab report, the CRS FAQ, and the committee’s hearing record — the same documents that contradicted him — remain publicly available.