Don’t ask if AI data centers use a lot of water. Ask if they’re worth it.
Plus: authority and escalation, AI “journalism,” and more
Good morning! It’s Wednesday, and here’s this week’s post. As a reminder, I wrote a new book this past fall. Please read about it, and:
This is substantially a post about journalism, so let’s start with some bad reporting: a July 2025 story from the BBC headlined, “‘I can’t drink the water’—life next to a U.S. data center.”1
The titular quote comes from Beverly Morris, a Georgia woman featured in the opening anecdote. Morris is a retiree who a decade ago bought a house in a small town southeast of Atlanta. Since then, a data center has been built about a quarter mile from her place. It’s owned by Meta, the company formerly known as Facebook, and is used in part to power artificial intelligence tools.
It’s not clear whether Morris can actually see the data center from her home, but she’s unhappy about its location and thinks it’s causing her plumbing problems:
She believes the construction … disrupted her private well, causing an excessive build-up of sediment. Ms Morris now hauls water in buckets to flush her toilet. She says she had to fix the plumbing in her kitchen to restore water pressure. But the water that comes of the tap still has residue in it.
“I’m afraid to drink the water, but I still cook with it, and brush my teeth with it,” says Morris. “Am I worried about it? Yes.”
Meta, however, says the two aren’t connected. In a statement to the BBC, Meta said that “being a good neighbor is a priority.” The company commissioned an independent groundwater study to investigate Morris’s concerns. According to the report, its data centre operation did “not adversely affect groundwater conditions in the area.”
While Meta disputes that it has caused the problems with Ms Morris’ water, there’s no doubt, in her estimation, that the company has worn out its welcome as her neighbor. “This was my perfect spot,” she says. “But it isn’t anymore.”
Now, at a personal level, I’m sympathetic to Morris, just as the BBC reporters intend. If I’d bought a house under these circumstances only to have what I considered to be disruptive and unanticipated development nearby, I too would be upset.
But I’m not so sympathetic to the reporters, because we have two testable theories of the case here, and the BBC—a large, well-resourced news organization—doesn’t appear to have tried to test either of them.
Is it possible that the Meta building somehow affected this well? Sure, maybe. I don’t know! But if there’s sediment in the water, it shouldn’t be that complicated to find out what the sediment is and whether it’s dangerous. Even if she doesn’t drink it straight, Morris is consuming this water when she cooks. Wow, that’s scary, wonder what it could be! is not a useful or sufficient response here. There are labs that handle this kind of thing, and it should also be possible to ask some knowledgeable third parties whether it’s likely that the construction had something to do with the sediment.
It should also be possible to make at least an informed guess as to whether the kind of construction matters—i.e., if the data center is the problem, is it a problem because it’s a data center, or would any building of this size and shape in this location affect Morris’s groundwater the same way? Again, I don’t know, but it seems to me that a report like this could at the very least solicit expert comment in an attempt to adjudicate these competing claims.
Maybe Morris is simply understandably miffed about a failing well—or maybe the headline quote is true, and Meta is lying and should face legal consequences for damage to her property. Would be good to know!
The BBC report doesn’t bother. It mentions Morris just once more after the portion I’ve quoted above. That mention is not more fact or context, just a note that future shifts toward sustainability from the tech industry are “little consolation to homeowners like Beverly Morris—stuck between yesterday’s dream and tomorrow’s infrastructure.” This is an implicit endorsement of her theory of the situation, a guilty verdict rendered after no pretense of a trial.
From Morris, as if playing a word association game, the BBC report leaps from bad AI thing with water number 1 to bad AI thing with water number 2: Data centers use water for cooling, sometimes in very large quantities. That could cause water shortages or higher prices for household consumers.
Then there’s a hiccup in the narrative—the story’s setting is Georgia, which evidently has a lot of water, and thus it seems there are no actual shortages or price hikes to mention—so we skip on to bad AI thing with water number 3: A creek near a data center construction site has brown water, and a nonprofit focused on local river health suspects the construction may be at fault.
Again: testable! In fact, a volunteer from the nonprofit names a kind of chemical he thinks may be involved. It is knowable whether his idea is correct, whether the construction is at fault, and whether any culpability is specific to data centers. This nonprofit might even have the resources to do the testing in-house; at the very least I expect they know who else could do it.
Yet again: no testing.
Instead we skip back to bad AI thing with water number 2 and hear about big tech companies’ projects and promises to help with water access in Africa and cleanup of polluted waterways like the Chesapeake Bay. Then comes the part about these future advances coming too late for Morris and others affected now. And then the closing line: “As AI grows, the challenge is clear: how to power tomorrow’s digital world without draining the most basic resource of all—water.”2
Now, I shouldn’t have to establish my bonafides to regular readers, but for those who are new here: I’m very far from an AI booster.
I do see its value for some uses—I use it to transcribe interviews and also find that the AI search functions some sites have introduced far outperform the older, text-based search. This is especially true when what you’re searching is images, because AI can search the images themselves, while a text-based search is limited to the text that the humans who uploaded the images thought to enter. Getty Images is a major source of photojournalism that we use at Christianity Today, for instance, and it’s amazing how much better its AI search is compared to its traditional search function.
But such very select uses aside, I’m an AI skeptic. I’m especially wary of any AI use that would:
do my thinking for me
simulate human relationality
have any entanglement with matters of faith
I see “Christian AI” products as particularly pernicious and get annoyed at the souped-up AI versions of autocomplete that every text editor now seems to include.
All that is to say that I’m not coming to this out of some default affection for AI. If it all went away tomorrow, I would shed no tears. My life would not be worse. Cheap transcription aside, it would be better. Even so, I want us to think about AI and its uses factually and coherently, and that’s not what’s happening here—not in the BBC story and not the matter of AI and water more broadly.
It’s true that data centers, whether they’re purpose-built for AI or not, often use water for cooling. It’s also true that what these usages vary widely by build type and physical environment. Undark has what strikes me as a nuanced article explaining some key differences here, and this (very long!) Substack post on AI and water from Andy Masley seems well-researched and sensible, deserving of the attention it has received, even if I don’t share the wider conclusions of that attention.
From reading these and other sources, my sense is it’d be prudent to use closed-loop cooling designs (instead of designs where a lot of water evaporates) and to concentrate centers in colder places with lots of water whenever those choices are available.
What seems rather less prudent is a lot of reporting on this topic. It’s sensationalist and ill-informed, tossing out lots of big numbers (millions of gallons a day!) without explaining exactly what those numbers mean for a specific center design or how they fit into extant water usage and needs in a specific geographic location.
Worst of all, much of the reporting on this water use fails to even consider the most important question: Is it worth it?
Water use is not inherently good or bad, not even in places where water is scarce. How we’re using it matters immensely. This is obvious with a moment’s thought. When there’s a drought, cities will temporarily ban lawn sprinkler use to conserve local water supplies because green grass doesn’t matter that much. They’d never tell people to stop drinking or cooking with water, though, because those uses are literally vital. Likewise, data centers are far from being the only businesses to use a lot of water. As the Masley post notes, in many locales they aren’t even the largest industrial water consumer, and golf courses also very thirsty.
So the plain use numbers may tell us whether a specific location’s existing water facilities and supply would be overwhelmed by a new center construction in the near term. But they can’t answer that bigger question of value.
As Masley is right to observe, “if you believe AI is entirely valueless, then any water used on it is wasted.” But if it has some value, that suggests that using some water is defensible. If it has a lot of value, then using a lot of water may be not merely defensible but compelling or even imperative. Rice cultivation typically uses a lot of water, but you don’t see debates about that because rice is a staple food for billions worldwide. Or consider the mills with which I’ve illustrated this post. Or toilets! The point is: Water use is good if it does good things.
So is AI good? Does it have value? How much? And how much water use does that value warrant in this location or that one? These are questions to be answered together. The merits of AI data center water use are downstream of the merits of AI itself. I tend to be pretty down on those merits, but even I can see that starting with the water is wrongheaded and confused.
Odds & ends
Rereading The Private Patient, by P.D. James
“As a former cop, I have to ask: What the hell is ICE doing?” by Michael Bollentin for Reason
A federal badge does not confer universal authority or “absolute immunity,” as some wrongly claim. Authority is contextual; it must be tied to a lawful mission, jurisdiction, and conduct. When those elements are absent or unclear, it is not sufficient to dismiss the resulting conduct as an unfortunate mistake. Many of these encounters appear to involve agents operating outside their authority, in ways that should demand discipline and accountability.
In many of these videos, encounters are initiated with U.S. citizens engaged in constitutionally protected activity: filming, speaking, standing nearby, or questioning authority. Those actions do not create detention power. Filming officers is not interference; verbal criticism is not obstruction; refusing an unlawful command is not resisting.
“‘You’re going to see more defections’: Thomas Massie’s ominous predictions for the GOP,” an interview with Rep. Massie at Politico
When was the last time you heard directly from the president or his team about anything?
Does the prayer breakfast count? I mean, he called me a moron at the prayer breakfast.
Just on the stage, thousands of pastors, including some from my own district, who apologized to me. They were literally here in my office that day and praying with me. And then they go to the prayer breakfast and hear the president say that. They are not impressed and I don’t think anybody was impressed by his performance at the prayer breakfast. It was completely political. But to answer your question, last time I heard from him was at the prayer breakfast and people said well what’s your response to that? And I just said I’m glad to know I’m in his prayers.
Actually, the headline uses a hyphen where I have more correctly inserted an m-dash, though it really ought to be a colon, but I thought doing [sic] would be too mean.
This was also a hyphen. What is even going on over there?
Hi Bonnie. I am so with you regarding AI, especially when it comes to ministry. I am aware of some believers using it in different ways that are not interrupting true fellowship. That has mitigated my resistance beyond myself. For me, I want to avoid it because we know that it is only the Spirit that brings life. Nothing compares with sitting at Jesus feet and hearing His words.
That said, I have followed local issues on the incursion of "hyperscale" data centers. Amazon has rented an entire building. My 40 year background in energy, and the environment and 7 years service on the Fredrick County (MD) Sustainability Commission have yielded an acute sensitivity to the environmental impact of all of our industry. AI's impact is abysmal compared to its benefit. We can go faster while consuming much water and extreme amounts of electric power. The latter has resulted in backtracking on efforts to protect the environment. I will restrain myself from detailed comments on crypto mining which uses power in patently wicked amounts for pure greed. Another side of the current administration that escapes too many among us.