Arizona Is About to Lose Nearly 30 Percent of Its Colorado River Water — as New Data Centers Are Planned
Arizona faces the loss of nearly 30 percent of its Colorado River water at the same time that new hyperscale data center proposals are meeting resistance across the state.
Arizona is becoming a test case for the question of how much water artificial intelligence actually costs. The state hosts 172 data centers — and at the same time, a cut of nearly 30 percent to its Colorado River allocation lies just ahead, according to The Verge (eded97c7). In the final part of October 2026 alone, two major reports on the same subject appeared, with partially conflicting conclusions.
The Verge, on October 29, describes large direct water withdrawals. KJZZ, the Phoenix area's public radio station, wrote on October 28 that data centers use less water inside their four walls than most people think — and that the real footprint lies somewhere else (38338b42). The tension between them is not necessarily an error. It may say something important about what kind of technology is actually behind the walls — and about how little we actually know.
The direct picture: millions of gallons per day
The Verge's report from TSMC's Arizona facility gives a daily water withdrawal of nearly 5 million gallons, drawn from Phoenix's municipal water supply (eded97c7). For comparison, Google's data center in The Dalles, Oregon is put at around 1.5 million gallons daily — roughly one-third of TSMC's Arizona consumption.
Worth noting: the TSMC facility is a chip fab, not a data center, and the figures rest on The Verge's reporting — not on public water permits or the company's own numbers. They should therefore be cited with that caveat. The fab appears in this story because, according to The Verge, it is part of the same infrastructure pressure driving data center construction in the state.
The counter-image: the hidden footprint through the power supply
KJZZ's report, based on interviews with water researchers, flips the perspective. Direct water use inside data centers is, according to the researchers, lower than most people think. The big consumption is indirect: data centers require enormous amounts of electricity, and power plants can use a great deal of water to deliver it (38338b42).
And here the knowledge stops. According to KJZZ, there is relatively little data on how much water Arizona's electricity generation uses, and analysts point out that it is difficult to get a complete picture of data centers' water needs without more transparency from the utilities. That is an important conclusion: the debate is not only about what companies won't say, but about what no one actually has numbers for.
Whether the two accounts contradict each other is unclear — they may well reflect different facility types and cooling technologies. A facility with a closed cooling system and a facility with other cooling and production processes can have very different direct water consumption, without either account therefore being wrong.
Viral numbers and actual limitations
KJZZ also uses its report to clean up one of the most widespread narratives about AI and water: claims that every AI query "drinks a bottle of water." Sarah Porter, director of Arizona State University's Kyl Center for Water Policy, tells KJZZ that such estimates from social media can be misleading. According to Porter, they are back-of-the-envelope calculations that can get people thinking — which she considers a good thing — but, as she puts it: "They are not accurate. They are not meant to be precise."
That is a useful distinction in a debate where both sides have incentives to simplify. Activists need forceful numbers to mobilize, and companies need to appear water-friendly. What both have in common for now is that they do not rest on a thorough public data foundation.
The companies' assurances — and what they are worth
Aligned Data Centers sent, according to KJZZ, a written statement in which the company claims its facilities are "engineered to require zero water for operational cooling" and use a closed loop that cools equipment without continuous withdrawals from the city's water pipes (38338b42).
It is worth taking seriously as a technological principle — closed cooling systems exist and are used — but it is, again, a company statement, not independently verified. KJZZ reports it as it is. For Arizona's residents, this is a concrete democratic problem: they are being asked to evaluate planned facilities on the basis of brief statements, not documented water data.
The politics shifting
Behind the technical disagreement lies a clear political shift. The Verge describes how the established facilities — both the chip fab and the 172 data centers — met little resistance, but that the newer and larger developments from Phoenix to Tucson require more of everything: more water, more electricity, more land than residents are willing to give (eded97c7).
The resistance is described as bipartisan, and it is easy to see why. In 2020 it was Phoenix's economic development chief, Christine Mackay, who according to The Verge led the effort to bring TSMC to the desert — at that time, such projects met little opposition. Now the same residents are to take a position on new facilities at the same time as nearly 30 percent of the Colorado River water is about to disappear. The Verge's reporter describes city council meetings where residents show up to protest data center projects, and conversations with local activists, elected officials, and their angry voters.
What a data-driven debate would require
If there is one actionable takeaway from the two reports, it is not "AI is dry" or "AI drains the desert." It is that Arizona lacks the data foundation needed to determine anything at all: utilities that publish water consumption tied to electricity generation, companies that document their cooling technologies and water sources, and municipal decisions that require such figures.
Until then, the debate will continue to be more about what people fear and what companies claim than about what is actually flowing through the pipes.