Sam Altman doesn't care if you think ChatGPT is thirsty. In fact, he’s calling your bluff. During a recent talk at the AI Impact Summit in India, the OpenAI CEO took a flamethrower to the growing pile of "eco-guilt" being dumped on artificial intelligence. His message was blunt: the viral claims that every AI query drinks a gallon of water are "fake," and if you're worried about the energy it takes to train a model, you should look in the mirror.
Altman’s logic is simple, if not a bit inflammatory. He argues that we’re holding AI to a biological standard we don't even meet ourselves. "It takes like 20 years of life and all the food you eat before you get smart," Altman told the Indian Express. He’s right. We don't tally up the megajoules of a toddler’s cereal to see if their future PhD is "carbon efficient." So why are we doing it to a GPU?
The Water Myth is Dry
The most persistent stick used to beat AI companies lately is the "water footprint." You’ve probably seen the headlines: Your ChatGPT query is draining local reservoirs. Altman isn't buying it. He calls these claims "totally insane" and "disconnected from reality."
It’s not just CEO posturing. The technology has shifted. The industry used to rely heavily on "evaporative cooling"—basically sweating for servers—which did consume massive amounts of water. But the newest data centers are pivoting. Many now use closed-loop liquid cooling systems. This isn't a "one-off" use; the water stays in the pipes. It’s like the radiator in your car. Once it’s filled, you aren't "using" new water every time you drive to the grocery store.
Of course, the scale still matters. Microsoft, OpenAI’s biggest partner, saw its water consumption jump 87% between 2020 and 2023. You can't ignore that kind of spike. But Altman’s point is that the "per-query" panic is a distraction. If the world is going to use AI for everything from curing cancer to fixing the tax code, we need to talk about total infrastructure, not the imaginary gallon of water in your "write me a poem" prompt.
The Human Energy Bill
Altman is leaning into a "fair comparison" argument that makes a lot of people uncomfortable. He’s essentially commoditizing human intelligence to show how efficient AI actually is.
If you measure the energy it takes for a trained model to answer a question versus a human to do the same task, AI is arguably already winning. A human brain runs on about 20 watts. That sounds efficient until you factor in the 20-year "training phase" involving thousands of pounds of food, climate-controlled housing, and a massive transport infrastructure.
Why the Comparison Grates
- Biological vs. Synthetic: Critics like researcher Matt Stoller argue that equating a "big spreadsheet" (AI) to a human being is dystopian.
- The Scale Problem: One human brain doesn't scale to a billion users. A single AI model does.
- The Source: Humans run on calories; AI runs on a grid that’s still largely powered by coal and gas.
Altman admits that total energy use is a "fair concern." He isn't saying AI is free; he’s saying it’s a catalyst for the energy breakthroughs we already needed. He’s betting big on nuclear fusion, having poured $375 million into Helion Energy. For Altman, the answer isn't "use less AI." It’s "make energy abundant."
The Real Bottleneck is the Grid
The real story isn't about ChatGPT’s "thirst." It’s about a global power grid that wasn't built for the 2020s. We’re trying to run the most advanced technology in history on an electrical skeleton from the 1970s.
By 2026, data centers could account for 35% of Ireland's total energy use. In the US, the demand from AI is expected to triple by 2028. This is causing a literal power struggle. Amazon’s data center builds in Europe are hitting seven-year waitlists for grid connections.
Altman is pushing for a move toward nuclear, wind, and solar at a pace the world hasn't seen yet. He’s basically saying that if we want the benefits of AI, we have to stop being afraid of the "N-word": Nuclear. Whether it's the Small Modular Reactors (SMRs) Microsoft is exploring or the fusion "holy grail," the tech industry is no longer waiting for utilities to catch up. They’re becoming the utilities.
Stop the Micro-shaming
If you’re feeling guilty about using AI to help you summarize a meeting or code a website, honestly, stop. The "environmental cost" of your individual query is roughly equivalent to leaving an LED lightbulb on for a few minutes.
The real pressure belongs on the providers and the policymakers. We need:
- Transparent Audits: Companies shouldn't just say their water use is "fake." They need to prove it with location-based reporting.
- Infrastructure Reform: Faster permitting for clean energy projects.
- Efficiency over Size: Shifting from "bigger is better" to "smaller and smarter" models (like quantization and distillation) that run on less juice.
The debate isn't actually about water. It’s about our willingness to trade physical resources for digital intelligence. Altman has made his choice. He’s going all-in on the idea that the "intelligence" AI provides will eventually help us solve the very energy problems the technology is currently creating. It's a high-stakes gamble, but in his mind, it's the only one worth playing.
Check your local utility’s green energy options if you want to offset your own digital footprint, but don't let the "gallon of water" myths keep you from using the tools. The industry is changing faster than the headlines can keep up.
