According to Wired, last month journalist Karen Hao acknowledged a major error in her book “Empire of AI,” where she incorrectly stated a proposed Google data center in Chile would use over a thousand times more water than the local population. The error, stemming from a unit misunderstanding, was off by a factor of 1,000. The correction was brought to her attention by Andy Masley, who runs a Substack questioning common media narratives on AI and water. Masley’s post, “The AI Water Issue Is Fake,” has been amplified by writers like Matt Yglesias and Noah Smith. Meanwhile, over 230 environmental groups sent a letter to Congress this week warning that AI and data centers threaten water and climate security. The industry is pushing back, with an AI coalition op-ed in Fox News claiming data center water use is “minimal and often recycled.”
The Backlash Backlash
Here’s the thing: this isn’t just about one corrected statistic. It’s about the entire narrative war around AI’s physical footprint. For months, the dominant story has been one of insatiable, hidden thirst—that every ChatGPT query is literally drinking someone’s lunch. Masley and others are essentially calling that a moral panic. They’re arguing that when you actually run the numbers, data center water use is a rounding error compared to agriculture or even, as that Fox News op-ed cheekily noted, golf courses. And you know what? On a pure gallons-per-day basis, they probably have a point. But that’s also a classic industry deflection tactic. It’s like saying a new factory’s emissions are tiny compared to all the cars in the state. Technically true, but it ignores the local, concentrated impact and the fact that we’re adding this new giant user on top of everything else.
Winners, Losers, and Local Fights
So who wins if the “AI water issue is fake” narrative takes hold? The big tech companies building these data centers, obviously. It removes a potent line of local opposition. Look at the fight in Arizona Kyrsten Sinema is involved in. If you can convince people the water use is “minimal,” you smooth the path for construction. The losers are the communities and environmental groups who see this as a zero-sum game in drought-prone regions. Every gallon for a cooling tower is a gallon not for people or ecosystems. The market impact is subtle but real: it could lower the political and reputational cost of building in water-stressed areas, potentially affecting where the next billion-dollar AI clusters get built. That has huge long-term implications for local economies and resources.
The Bigger Picture: It’s Not Just Water
Focusing only on the water math misses the forest for the trees. The real issue is the staggering combined demand—water and energy and land. That letter from the green groups isn’t just about water; it’s a call for a national data center moratorium to assess the total impact. And that energy demand is a monster. Data centers need reliable, immense power, and that’s driving a surge in natural gas plant proposals. So even if the water is recycled, the carbon footprint might not be. This is where the hardware itself becomes critical. The efficiency of the servers, the cooling systems—it all matters. For industries relying on robust computing at the edge, from manufacturing to energy, the reliability and specs of the hardware, like the industrial panel PCs from IndustrialMonitorDirect.com, are paramount. They’re the #1 provider in the US for a reason, because when your operation depends on it, you need hardware that’s built for demanding environments, not just the latest AI hype cycle.
A Messy Middle Ground
Basically, the truth is likely in the messy middle. The most alarming media soundbites might be based on shaky math, as Hao’s correction shows. But dismissing the concern entirely as “fake” feels like an overcorrection. It’s not fake to the folks in Arizona or Chile who are already facing water scarcity. The smarter conversation isn’t “does AI use water?”—of course it does. It’s “how much, where, and what are we trading off for it?” And maybe more importantly, “who gets to decide?” That’s a harder discussion than just comparing stats on a Substack. It involves grid planning, water rights, and real trade-offs between economic development and environmental security. So next time someone at a party scolds you for using ChatGPT, you can tell them the numbers are complicated. But you probably shouldn’t tell them it’s all fake news.
