Sam Altman would like you to stop worrying about how much water ChatGPT drinks. On a podcast on 1 September, the OpenAI boss offered a soothing comparison: 38,000 ChatGPT queries, he said, use about the same water as growing a single California almond, and a data centre uses no more water than an office building. Relax, in other words. The internet did not relax. It turned him into a meme called “Sam Almond,” complete with AI-generated pictures of his face on a nut.
The mockery is fun, but the interesting bit is what the fact-checkers found when they actually did the sums, because the truth is messier and more interesting than either Altman’s spin or the panic he was trying to calm.
The claim, checked
PolitiFact rated Altman’s statement “Mostly False,” and CalMatters walked through why. His comparison leans on a very low, unconfirmed estimate of ChatGPT’s water use and an inflated figure for almonds. One California almond takes roughly 6 litres of water to grow. Independent estimates suggest you would need somewhere between 1,000 and 10,000 conversational AI prompts to match that, not 38,000. So Altman’s number is off by a factor of several, in the flattering direction, and he later admitted he was recalling it from memory and might be wrong. He was.
It is worth being precise about how he cheated the comparison, because it is a classic move. Pick a genuinely thirsty product (almonds are a punchline in California water debates for good reason), use the rosiest possible figure for your own product, and present the ratio as if it settles the argument. It does not. It just launders a real question through a tidy soundbite.
Now the inconvenient part
Here is where an honest piece has to annoy both sides. Altman’s specific number is bogus. But the wider panic that AI is single-handedly draining the reservoirs is also overstated, and pretending otherwise is its own kind of dishonesty. The blunt truth is that nobody has reliable public figures for how much water a given AI query uses, because the companies do not disclose it in any consistent way. Altman exploited that fog; so do some of the scarier headlines pointing the other direction.
What we can say with more confidence is that the water story is real but local and lumpy. A single hyperscale data centre plonked in a drought-stressed county can put genuine strain on that community’s water, which is exactly why local opposition has erupted, while in aggregate AI’s water use is still small next to agriculture and power generation. Both things are true: the sector’s thirst is a legitimate local problem and a convenient national scare, and the almond meme is funny precisely because it flattens all of that into a nut.
What this means
Do not let a CEO’s cherry-picked ratio talk you out of a real concern, and do not let the backlash talk you into believing AI is why your tap runs dry. The useful posture is to demand the one thing that would end the argument: actual, audited, per-query water and energy disclosures from the labs. Until Altman and his rivals publish real numbers instead of almond metaphors, everyone is arguing about a figure nobody can see, and the person with the friendliest soundbite wins by default. That is not a water shortage; it is a transparency shortage. (Sources: PolitiFact and CalMatters fact-checks, September 2026.)
Related on Top Tool Stack: why AI’s real costs are catching up with it.