14 September 2026
Heard In AI

Data-center water totals don't settle the argument next door

On Moonshots with Peter Diamandis, one speaker argued that hyperscalers could win communities over with cheaper power, school programs and better-looking buildings. A fellow panelist replied that the fight has already been politicized beyond what benefits can fix — and the viral water figures at the center of it describe the whole country, not the town being asked to host the building.

A briefing reports one development when it happens. We correct or clarify it later; a new development gets a new briefing. How our formats work

If he were building a data center, one speaker on Moonshots with Peter Diamandis said, he would do it very differently. He would promise education programs in local schools. He would promise to make electricity in the community cheaper than it is today. And he would stop building black boxes: "We're going to make them look like cathedrals." Spending an extra 10% on all of that, he said, seemed obvious enough that he could not understand why it was not already happening.

A panelist the others addressed as Alex answered immediately. "My worry is that that wouldn't help. My fear is that this isn't because people think data centers are unsightly or unaesthetic." His concern was that opposition to data centers has been "overly politicized," and in the worst case stoked from outside the country. A number of U.S. adversaries, he said, would like nothing better than to slow America's data-center buildout, and his worry was that in a year or two some quantum of the opposition would turn out to have been encouraged by them.

The numbers behind the fight

Data centers are the buildings full of computers that train and run AI systems. They need electricity, and the equipment needs cooling, which is where water enters the local argument.

The host framed the segment with polling he attributed to Heatmap News. A year ago, he said, Americans were split 43 to 42 on whether they opposed data centers being built near them; asked again this month, 75% were opposed and 61% strongly opposed. He described the trend as close to a straight line heading toward everyone. He gave no survey method or sample size on air. In the same week, he noted, Senator Bernie Sanders again called for a nationwide moratorium.

Against that he set a post on X the panel described as viral, about what it called the data-center water myth. The figures he read out: data centers at 627 million gallons a day, golf courses at about 2 billion gallons a day — roughly three times as much — power plants at 133 billion and raising cattle at 137 billion.

What 627 million gallons a day actually counts

That headline number lines up closely with modeling published by Lawrence Berkeley National Laboratory in its 2024 report on U.S. data-center energy use. For 2023 the report estimates about 66 billion liters of water consumed directly at data centers, and close to another 800 billion liters consumed indirectly, through the electricity those facilities buy. Added together and divided across a year, that is roughly 630 million gallons a day.

Two details change what the figure means. The first is the word consumed. Consumption counts water taken out of the immediate water cycle — evaporated in a cooling tower, for instance — rather than every gallon withdrawn from a river or aquifer and later returned.

The second is that most of the total is not in the building at all. The indirect share, calculated from regional electricity mixes and where data centers actually sit, is water used at power stations. Cooling technology and local generation vary enough by region that the report treats geography as a major source of variation.

And the model produces national annual totals. Berkeley Lab built it from the ground up, equipment by equipment, for 2014 through 2023, with scenarios running to 2028 whose range depends heavily on how many AI servers ship and how hard they are worked. It finds infrastructure efficiency improving even as total demand rises. It does not tell a specific town what a specific proposed building would draw from its own supply.

Local decisions, national arguments

That gap between national accounting and local burden is close to the panel's disagreement about politics. Alex's point was structural: in the United States, siting a data center ends up a local decision. China, in his description, can simply declare that the east handles data and the west handles compute and energy, then build a national grid to join them. American municipalities and states decide for themselves what their land is used for — a system he considered far easier for an adversary to polarize, if the goal were to shut computing out of terrestrial deployment altogether.

The host pushed back: if the opposition rests on false data and an outrage cycle, it can be countered. Why, he asked, is the White House not getting out in front of an issue plainly gaining steam? He pointed to Mark Zuckerberg's video from the previous week, promising better schools, better access to jobs and a positive contribution to the community, and argued that a community could end up so clearly better off — cheaper energy above all — that residents would reject the scare stories themselves.

Another speaker thought the problem sat deeper than any single campaign. Information systems, he said, reward compelling narratives over evidence, and data centers are where that shows. Fifty years ago the posture was show me the evidence and I'll form an opinion; now, in his account, it is have an opinion and show me the evidence that confirms it — amplified by social media, where nuance does not spread. His complaint was about legislating from a story rather than a measurement: "we're making national policy based on these stupid innuendos and memes rather than measurement."

The facility that claims to make water

One concrete counter-example came from the panel rather than from any report. A speaker described Rob Fisher, formerly president of Link Studio, who left to start a data-center company called Provocative AI. As he told it, the facility is both carbon negative and water negative: it runs mostly on nuclear power from Seabrook, New Hampshire, does its own carbon capture using waste heat, and captures more carbon than the loop produces. On water, he said the company found it could use the humidity accumulating from the temperature gradient to produce more water than it consumes, and run on its own dripping condensate. "Then nobody can complain. We're actually water negative and carbon negative." That was his account of the company's claim; he cited no audit or outside measurement.

The segment ended where the panel's argument about narrative predicted it would — in branding. Rename them, someone suggested: intelligence foundries. Compute citadels. AI churches. "It's a branding problem. Again, it's not a factual thing."

Which leaves the question the national figure was never built to answer. Berkeley Lab's totals describe the country's data centers in aggregate, including water consumed at the power plants that supply them. The vote that decides whether a building goes up happens in one county, about one site, one grid connection and one water table.

Share this article

Go to the original

Sources & further reading

  1. 01

Connected ideas and articles

From the conversation

Podcast episodes

Article history

Updates to this article

Tags

Keen bets the last AI company standing will be Chinese

On The Diary of a CEO, economist Steve Keen argued that the only thing likely to slow the race for frontier AI is its sheer cost and the physical resources it needs — and that the company left standing will be Chinese. Other speakers pushed back with military necessity and the long unprofitable years of earlier internet giants, and Keen pointed to the recent Kimi release as his example.

5 min read

NVIDIA's open-model push is about GPU demand, the Moonshots panel says

On Moonshots EP #283, Peter Diamandis introduced a reported $6 billion NVIDIA arrangement with the coding startup Poolside as America's answer to Chinese open models. Emad Mostaque argued the real driver is selling more GPUs, while Alex and Dave disagreed about whether licensing-and-hiring deals exist to dodge antitrust review or simply to hire fast — and what happens to the half of Poolside that stays behind.

7 min read

Before a grand AI treaty, Graylin wants hotlines and shared safety tests

Alvin Wang Graylin proposes a practical starting point for US–China AI cooperation: an emergency hotline, shared safety tests and an agreement to keep talking. Speaking personally on Moonshots, ahead of a September 24 dialogue described in the episode, he connects those steps to a larger bargain—financing AI deployment abroad while spreading agreed safety standards.

6 min read

AI money can help nuclear scale—but not guarantee faster power

Ramez Naam sees AI demand as a powerful source of nuclear financing, but doubts new small reactors can supply electricity within the five-year window he considers reasonably predictable for investment. His argument turns on what can be delivered sooner—and whether repeated construction and factory production can make later plants cheaper.

6 min read