Sam Altman thinks you’re being dramatic.
The OpenAI CEO took the stage this week to perform his favorite trick: telling the public that their eyes are lying to them. This time, the target wasn’t the existential dread of a rogue AGI or the slow death of the creative class. It was water. Specifically, the millions of gallons of it required to keep the world’s most famous chatbots from overheating.
Altman dismissed the growing concern over AI’s massive water footprint as "totally fake." It’s a bold stance for a man whose company relies on Microsoft’s sprawling data centers—facilities that, according to Microsoft’s own sustainability reports, saw a 34% spike in water consumption in a single year. That’s roughly 1.7 billion gallons of water. Or, if you’re looking for a mental image, about 2,500 Olympic-sized swimming pools evaporated into the ether to help people write slightly better emails and generate pictures of cats in space suits.
But sure. "Totally fake."
The friction here isn't just about the numbers; it’s about the audacity of the pivot. For years, the Silicon Valley elite have positioned AI as the ultimate clean-tech savior. We’re told these models will optimize our power grids and solve the fusion puzzle. But in the present, the reality is much thirstier. Every time you ask ChatGPT to summarize a meeting, a data center in a place like West Des Moines, Iowa, pulls a literal cup of water out of the local supply to cool the hardware.
Researchers at the University of California, Riverside, estimated that a single conversation of about 20 to 50 questions with GPT-4 costs the planet a 500ml bottle of water. That doesn't sound like much until you multiply it by hundreds of millions of users. Then, it starts to look less like a statistical error and more like a crisis.
Altman’s dismissal follows a familiar script. Tech giants love to talk about "efficiency" as a moral get-out-of-jail-free card. The logic goes like this: yes, we’re using more resources now, but the software is getting so smart it will eventually figure out how to use fewer resources. It’s a promissory note written in disappearing ink. We’re burning the house down to build a better thermostat.
The CEO’s "fake" comment likely stems from the industry’s obsession with "net-zero" accounting tricks. Microsoft and Google love to point toward their investments in water replenishment and carbon credits. They buy a forest here, they fund a pipe repair there, and suddenly, the billions of gallons pumped out of local aquifers don't exist on the balance sheet. It’s a shell game played with the fundamental building blocks of life.
There’s a specific kind of arrogance required to call an environmental cost "fake" when the local residents of the towns hosting these data centers are seeing their utility bills climb. In Iowa, the city of West Des Moines used about 6% of its total water supply for Microsoft’s clusters. When the heatwaves hit, the locals are told to conserve while the servers get a cold bath. That isn't a fake conflict. It’s a trade-off. We are trading local ecological stability for the privilege of automating the middle class out of existence.
Altman wants us to focus on the "energy breakthrough" he claims is coming. He’s heavily invested in Helion Energy, a fusion startup. He talks about a future where energy is so cheap and abundant that these resource concerns will seem quaint. It’s the classic Silicon Valley escape hatch: ignore the wreckage of the present because the future will be shiny enough to buff out the scratches.
But fusion is always a decade away. The water is disappearing now.
The industry’s refusal to be honest about its physical footprint is becoming its most defining characteristic. They want the tech to feel ethereal. They want it to be "the cloud." But the cloud is just a bunch of loud, hot boxes in a warehouse that need constant cooling to avoid melting into a puddle of expensive silicon.
If the water usage is truly "fake," one wonders why OpenAI and its partners aren't opening their telemetry data to independent auditors. Why not show us the real-time consumption per query? If the efficiency gains are as miraculous as Altman suggests, the data should be his best friend. Instead, we get hand-waving and dismissive quotes at high-society conferences.
It’s a neat trick. If you tell people the problem isn’t real, you never have to pay the bill for fixing it.
We’re being asked to bet the planet’s cooling systems on the hope that a chatbot will eventually become smart enough to apologize for drinking the well dry. It’s a hell of a gamble.
Is it still "fake" if the taps run dry while the server fans are still spinning?
