Why does the public water figure for a single data center keep changing, and why does it only ever seem to go up?
It’s a question worth asking plainly, before getting into any of the technical explanations offered for it, because the pattern itself tells you something the explanations often don’t. That’s the question hanging over Project Jupiter, a roughly 1,400-acre data center development from Oracle and OpenAI going up in Doña Ana County, in the southern New Mexico desert near the Texas border.
The answer says less about any one company lying and more about how these projects get approved before anyone outside them really knows what they’ll cost a community.
The number started small. An early report in the Santa Fe New Mexican put the facility’s expected water use at a million gallons a day, a figure that got repeated widely as the project’s water footprint.
Then, after Oracle announced it was switching part of the facility’s power supply from water-intensive natural-gas turbines to fuel cells, the company put out a new figure: about 11 million gallons of non-potable, recycled water for the data center and fuel cell system combined, according to reporting by NPR. NPR’s own framing of the shifting figures was blunt: the numbers keep changing, and residents trying to plan around them are left chasing a moving target instead of a fixed answer.
Worth pausing on what “non-potable, recycled water” actually means here, since it sounds reassuring in a press release and does real work in that sentence. It’s water that isn’t drawn straight from a community’s drinking supply, and closed-loop recycling does mean less water gets used once than an older, open-loop cooling system would use.
It doesn’t mean the water involved comes from nowhere, or that a region in a documented drought can treat 11 million gallons committed to one industrial project as a rounding error. Non-potable and finite aren’t the same axis, and a project can be responsible on one and still meaningfully strain the other.
Doña Ana County is already in a documented, long-running drought. Eddie Estrada, who lives six miles from the site, put the fear in plain terms. “Being that we’re in a drought, and then to allow a project like this to use that much water, the fear is that we’re going to run out,” he said.
That’s not an abstract worry in a region where agriculture and household wells already compete hard for a limited, shrinking supply. A number that starts at one million and climbs toward eleven, whatever the technical reasons behind the jump, reads very differently to someone whose well could run dry than it does in a corporate press release.
The jobs argument is the other half of why a project like this gets approved in the first place, and it’s not a cynical one. County Commission Chairman Manny Sanchez was straightforward about what drew local officials to the deal: “We’ve never had that type of money here in Doña Ana County.” For a county that hasn’t had this scale of investment before, the appeal of new jobs and a larger tax base is real, not manufactured.
The tension has nothing to do with officials being foolish to want it. It comes from the water numbers attached to that promise not being stable enough, early enough, for anyone to weigh the tradeoff honestly before signing on.
Commissioner Susana Chaparro asked the question that actually matters more than any single figure: “What is it gonna look like in one year? What is it gonna look like in three, five, 20, 30 years?” A single day’s water use is a snapshot. A drought-stressed county’s water table is a decades-long story, and a data center, once built, doesn’t get less thirsty as the region around it gets drier.
The mismatch between how fast these projects move and how slowly water questions get real answers is the actual story here, more than any single disputed number.
Doña Ana County isn’t an isolated case, either. A separate Texas Tribune analysis of the Stargate data center project in Abilene found state planners lack solid data on future water demand because companies are required to report past water use but not projected future use, a gap that shows up again and again across the region’s data center boom.
The pattern repeats with enough consistency to stop looking like a coincidence: an early, modest-sounding water estimate gets used to win local approval, and a larger, more complicated figure arrives later, once the project is already too far along to meaningfully renegotiate. That pattern is the actual policy problem, more than any single company’s specific number in any single county.
None of this means AI data centers can’t be built responsibly near a stressed water supply, or that every job they bring is a bad trade. It means the trade only works if the numbers a community is asked to accept are the real ones, disclosed before ground breaks rather than revised upward afterward. Doña Ana County said yes to a number that turned out not to be the number.
That’s the part worth remembering the next time a similar project shows up promising jobs, an investment nobody’s ever seen locally before, and a water figure that sounds small enough not to worry about.
I don’t have any technical standing to say whether 11 million gallons of recycled, non-potable water is a responsible amount for a project this size.
What I do know, from watching how numbers get treated across very different kinds of announcements for a living, is that a figure repeated confidently in a press release isn’t the same thing as a figure that’s been independently verified, and a community deserves the second kind before it agrees to anything, not after.
