The Water Was Never the Problem

The simple believeth every word: but the prudent man looketh well to his going. — Proverbs 14:15

A True Number and a False Story

Begin with the viral post, because it is an honest specimen of a dishonest genre. A man stands in a dry stretch of the Rio Grande in New Mexico, films the cracked bed, and tells his audience that the river is gone because the data centers drank it. He names a facility — Meta's campus at Los Lunas — and a figure: roughly seventy-five million gallons of water a year. Then he draws the line the footage was filmed to draw, from the cooling towers to the dry sand at his feet, and asks his viewers to pressure their representatives before it is too late.

Here is what makes the post worth examining rather than merely dismissing: the number is essentially correct. The Los Lunas data center did consume about that much water in the reporting year, by the company's own published account. The post did not fabricate its figure. What it fabricated was the relationship between the figure and the riverbed — and that is the more instructive kind of error, because it is the kind a sincere person makes and an audience rewards. A true number welded to a false story travels farther than an honest uncertainty, and it does more damage, because the people who later discover the weld stop trusting the number too.

The discipline this essay tries to model is the opposite move. Take the number seriously. Take it all the way down to where it actually lands. And let the conclusion be whatever the arithmetic says it is, even when that conclusion disappoints the people who were hoping for a villain.

Where the Water Actually Goes

First the mechanism, because the word everyone uses — consume — hides the whole story, and the hiding is what lets both sides talk past each other. There is a large difference between water that a facility withdraws and water that it consumes. Withdrawal is water you take and largely give back: it passes through, warms up, and returns to the river a little hotter than it left. Consumption is water that leaves the watershed entirely, and it leaves almost always by one route — it becomes vapor and rises into the sky.

The dominant way to cool a large data center is evaporative, and it consumes by design rather than by accident. The reason is a quiet piece of physics that turns out to be the same one the prophets reached for when they wanted an image of cleansing. To turn liquid water into vapor takes an enormous amount of energy — far more than merely heating it — and the departing vapor carries that energy away with it. So if you let a fraction of your cooling water evaporate, the rising vapor hauls the heat off the water that remains, and the water that remains comes back cold enough to use again. The evaporated fraction does not pass through. It is gone, lifted into the air, leaving its dissolved salts behind in the water that stays — which then has to be partly dumped, as a concentrated brine, before it scales the pipes. Water up the tower as vapor; water out the drain as brine. Roughly four-fifths of what an evaporative system withdraws is consumed this way, gone from the local basin until it falls again as rain somewhere, on someone, downwind.

There is a strange and exact inversion here, worth naming because it reframes the whole debate. Evaporation is nature's own purification — the one motion in the water cycle that goes up, the step that leaves every contaminant behind and returns the water clean as snow. In the cooling tower, that sacred-looking process is run as an industrial heat sink and then severed from its return. The vapor that rises from a healthy spring comes back down to recharge the ground it left. The vapor that rises from a cooling tower in a dry basin is debited from that basin and credited, as rain, to another. It is the redemptive lift of the water cycle with the homecoming cut off — a spring run backwards, lifting living water out of a place that needed it and posting it to an account somewhere else. That is the real cost of evaporative cooling, stated honestly. It is not that the water is destroyed. It is that it is relocated, and the basin that gave it does not get it back.

The Calibration That Ends the Panic

Now the scale, because scale is what the riverbed video is built to make you forget. The question to ask is not how big seventy-five million gallons sounds standing in a dry wash. It is how big it is against the resource it is drawn from. And the cleanest way to feel that is to measure the whole of the thing against a river.

Take every data center in the United States — every cooling tower in the country — and add up all the water they directly consume in an entire year. The figure, by the most cited national accounting, is on the order of seventeen billion gallons annually. Now set it beside the Mississippi River, which discharges in the neighborhood of four hundred billion gallons on an ordinary day and considerably more on a high day in spring, swollen with snowmelt. The arithmetic is almost rude in how decisively it settles the matter. A full year of every data center's direct water consumption in America equals roughly three percent of a single spring day's flow of the Mississippi. Turn it around and it is starker still: the Mississippi pours, in about forty-five to fifty minutes of one spring day, as much water as the entire national fleet of data centers consumes directly in twelve months.

Project forward to the alarmed end of the forecasts, where that consumption quadruples by the end of the decade, and you have moved the figure to about thirteen percent of one spring day — a little over three hours of one river standing in for a year of the whole country's cooling. This is not a defense of waste and it is not the whole picture; the indirect water embedded in the electricity these facilities draw is several times larger than their direct cooling use, and an honest account multiplies the number accordingly. But even multiplied severalfold, the conclusion holds and is not close: the United States does not face a water-quantity crisis on account of artificial intelligence. The resource exists in overwhelming abundance. Anyone who tells you the nation is running dry to cool its machines is selling you a dry riverbed and hoping you will not ask how big the river is.

So Why Is Anyone Actually Suffering?

If the resource is so abundant, the fair challenge is the obvious one: then why are there real people with real grievances? And there are — this is where the doomers have a true thing the boosters would rather not discuss. The honest answer is that water is abundant nationally and scarce locally, and a national surplus is no comfort to a particular well in a particular dry county. The harm is real, but it is not the harm the panic names. It is not the cooling math overwhelming the continent. It is a handful of specific, avoidable, local failures, and they share a signature.

Consider the cases that actually went wrong. In an affluent Georgia subdivision, residents noticed their water pressure falling, and an investigation found industrial-scale hookups feeding a data center campus — one of them installed without the utility's knowledge — that had drawn tens of millions of gallons without initially being billed for it. The injury that enraged the community was not the volume in the abstract; it was that the same households had been told to stop watering their lawns to conserve while the largest user in the county was metering nothing. In Oregon, a single city of sixteen thousand saw roughly a quarter of its annual water go to one company's data centers — and learned the figure only after a public-records fight that the company funded the opposing side of, arguing the number was a trade secret. In Pennsylvania, a man's well ran dry for the first time in thirty-nine years, and while drought was the likely driver, he could not get a straight account of what the new water-hungry neighbors were taking.

Look at the signature on all of it. None of these is a story of the evaporative arithmetic being too large for the country to bear. Every one of them is a story of a knowable quantity that was concealed, unmetered, unauthorized, or simply never demanded — discovered by citizens at a dry tap rather than disclosed by the authorities who granted the connection. The water was always there to be accounted for. The failure was that no one with the right duty did the accounting before the valve was opened.

The Counter-Example That Proves the Point

The dry-riverbed post chose the wrong target, and the way it is wrong is instructive, because Los Lunas is closer to how this is supposed to work than to how it fails. The seventy-five-million-gallon figure exists in public at all because the company published it. The facility operates under a metered allocation with a contractual ceiling and a reporting process to enforce it. It runs a water-reuse program that drove its average daily demand far below the rights it holds. By the town's own accounting it consumes a single-digit percentage of municipal water — meaningful, worth watching, but not a river-killer. And the basin's genuine distress, which is real, traces overwhelmingly to causes that have nothing to do with the data center: decades of agricultural draw, interstate compact obligations measured in the billions of gallons, and a river that has run dry in stretches every winter for as long as anyone has kept records.

In other words, the man filmed his alarm standing on the one local example that most resembles competent governance, and blamed it for a riverbed that was dry for reasons older than the internet. This is what happens when scale and sourcing get collapsed into a single cartoon: the genuinely well-run case and the genuinely negligent case look identical from inside the panic, because the panic is not actually about how the water was governed. It is about the machines, and the water is a convenient stick.

The lesson runs the other way from the one the video teaches. The difference between Los Lunas and the Georgia subdivision is not the technology and not the volume; both use water to cool, and the negligent case used less than the well-run one. The difference is entirely in the governance — metered versus unmetered, disclosed versus concealed, capped versus open-ended, watched by someone with a duty versus discovered by a homeowner with low pressure. The variable that determines whether a community is harmed is not how thirsty the machine is. It is whether someone competent and accountable did the sourcing math before the water started to flow.

Whose Duty It Actually Is

Which lands the whole matter on a single, unglamorous point of responsibility. The water consumed by evaporative cooling is mathematically knowable in advance — knowable from the nameplate capacity of the facility and the wet-bulb temperature of the local air, computable before a single server is installed. There is no excuse for surprise. A dry well downstream of a data center is never a failure of arithmetic; the arithmetic was always available. It is a failure of someone to demand the arithmetic and match it to a source that can actually bear it.

The trouble is that the duty has too often sat with the wrong office. A municipal council chasing a ribbon-cutting and a tax base is structurally the wrong body to weigh a watershed, because its incentives run toward the press release and the jobs number, and the recharge rate of an aquifer does not fit on a press release. The water question is not a development question wearing a hydrology costume. It is a hydrology question that a development office has every incentive to wave through. Placed there, the projectable number does not get projected, because no one whose job depends on the deal wants to be the one who computed the figure that kills it.

The duty belongs with the people whose actual charge is the basin: county watershed authorities, hydrology offices, conservancy districts — bodies accountable to the aquifer's recharge and the compact's obligations rather than to the quarter's announcement. And it does not even require banning the thirsty design. Evaporative cooling is cheaper in energy, and energy carries its own water and carbon shadow, so the right answer is genuinely local. Beside a great river in a wet basin, evaporative cooling drawing on abundant flow may be the honest and even the greener choice. Over a stressed desert aquifer, the energy penalty of a closed-loop, near-waterless system is simply the price of not mining groundwater that will not come back. That determination — which cooling for which place — is a question for someone who reads water tables, not someone who reads press clippings. Put the decision there, require the projection in public before the allocation is granted, and the entire category of harm this essay catalogued mostly disappears. Leave it where it has been, and we will keep discovering by dry tap what a competent office could have told us by spreadsheet.

The Decision Worth Guarding

So the conclusion satisfies no one who came for a slogan, which is usually the sign it is close to true. The booster who says there is nothing to see here is wrong: real wells have run dry, real communities have been billed for water a concealed neighbor was taking for free, real figures have been buried as trade secrets. The doomer who says the machines are draining the rivers is wrong by an even wider margin: a year of the entire nation's cooling fits inside an hour of one spring river, and the loudest example of the panic turned out to be its best-governed case. Both are reaching past the actual lever to grab a more satisfying story.

The actual lever is the one quiet decision that keeps getting made in the wrong room: who accounts for the water, and before or after the valve opens. We can have data centers in great abundance — the resource is there, lavishly, and the work these machines do is worth real water honestly sourced. What we cannot afford is to keep letting a knowable input be granted by an office with every incentive not to know it. The evaporated water rises clean and comes down somewhere; that is the mercy built into the cycle. But it does not come down where it left unless someone planned for the basin as well as the building, and that planning is a duty, not a press release.

Keep the decision, then, in the hands that are accountable to the water. Let the developers build and the machines compute and the towers steam; none of that is the sin. The sin is the one trade that keeps getting folded invisibly into the welcome — the trade of an honest accounting of the source for the speed of the announcement. The water was never the problem. The water is abundant and the physics is knowable and the math was always there to be done. The only thing ever in short supply was the will to do the math out loud, in the right office, before the river was asked to answer for it.

---

On sources and figures. The Los Lunas consumption figure of roughly seventy-five million gallons (about 283 megaliters) for 2023 is from Meta's own published sustainability reporting as relayed in contemporaneous New Mexico local reporting; the metered allocation, contractual ceiling, reuse program, and single-digit share of municipal use are from the Village of Los Lunas' own statements and local coverage of the expansion agreement. The Rio Grande basin's stress predating and exceeding any data-center draw — the curtailed irrigation season, the multi-billion-gallon interstate compact debt, the river's seasonal winter dryness — is from regional reporting and is well anchored. The community-harm cases (the Georgia subdivision's unmetered industrial hookups and conservation-order inequity, the Oregon city's quarter-share consumption and the trade-secret records fight, the Pennsylvania dry well) are from 2025–2026 reporting; in each, note that the causal link between facility and specific harm ranges from documented (billing and metering failures) to contested (well contamination, where correlation is not established causation), and the essay treats them at the confidence the record supports. The national direct-consumption figure (~17 billion gallons in 2023, with projections that it could double to quadruple by 2028) derives from a Lawrence Berkeley National Laboratory analysis and corroborating reporting; the indirect water embedded in electricity generation is several times the direct figure and is acknowledged rather than dismissed. The Mississippi calibration uses a long-term mean discharge near the river's mouth of roughly 593,000 cubic feet per second and a representative high-spring-day figure on the order of 800,000 cubic feet per second; the spring figure is an estimate for illustration, and the conclusion is robust to a wide margin of error in it. The point of the calibration is proportion, not precision: even generous adjustments leave a year of national consumption inside hours of one river. This essay is offered as an invitation to right-size a real concern rather than to dismiss it — the harm is genuine and local, the resource is genuine and abundant, and the remedy is to move the sourcing decision to the office that is accountable to the watershed. In all thy getting, get understanding.