I knocked over a full glass of water about four paragraphs into writing this piece. That moment of clumsiness used more water than every question I’ve asked Claude to research the whole post — and almost certainly more than my entire month of AI use. But here’s the thing: if we actually cared about water, we’d look at the dinner table, where a single meatloaf matters orders of magnitude more than a lifetime of chatting up Chat.

And then the mystery that really gets me: how did a stat this shaky seep so deep into the zeitgeist — and what do we do about it?

Animal Factories Use Way More Water Than AI. Like, Way More

AI doesn’t really use that much water. I won’t harp on this point too much (mostly because it’s already been covered in more depth by Andy Masley), but here’s a brief overview: a ChatGPT query’s full water footprint is roughly 0.9 mL — about two teardrops — and ~85% of that is offsite at the power plant, not the cooling tower everyone pictures.

But there is an industry that does gulp down water like there’s no tomorrow: animal factories. Globally, animal agriculture uses over a quarter of the freshwater we use, and the water use of animal products is always higher than that of the best nutritionally equivalent plant-based options. Now, when people hear this, they might picture cows and pigs and chickens in a field tossing back water like Robert Hays on an airplane, but most of this water, 98%, is actually from animal feed. This is a recurring theme when analyzing the environmental hoofprint of animal ag — we have to include all the inputs for the animal feed as well, which tends to be plants like corn, soy, and alfalfa, which all need water for themselves. If instead we used that land to grow food for humans, we’d be able to save a ton of water.

How big is this difference? Massive. I charted the water use of a few animal-based and plant-based foods, along with the usage of 10,0001 AI prompts. To be extra fair, I plotted both the highest and lowest estimates of AI water usage I could find2.

And remember, this is compared against 10,000 AI prompts, which would still take well over a month, even for the hardest-core Claude fan. Swapping out just one beef burger for an Heura burger saves about as much water as vibe-coding to your heart’s desire3.

This argument holds up even if we examine the water use by type. Quick refresher: green water is rainwater soaked into soil and transpired by plants, while blue water is the more important metric: the surface and groundwater we actively pull from rivers and reservoirs. Still, the underlying facts stand if we look at just blue water: 10,000 prompts, which is roughly enough for a year’s use, is still, at the very highest estimate, worth 2 kilos of tofu (or, as I call it, a serving size). No matter how you cut it, your food choices matter more than your AI.

But let’s flip the way we look at it. What about from a national perspective, in terms of a nation’s total water usage? This is where the problem really starts to look absurd — animal agriculture is the single largest use of water, responsible for over 40%! That’s more than every shower, golf course, jacuzzi, slip ‘n slide, factory, faucet, swimming pool, sprinkler, toilet, water park, and TikTok rug cleaning video combined.

This gap is probably far higher in other developed countries — consider how the US is a net importer of beef and has an above-average number of data centers. The USA likely has the smallest ratio of animal agriculture’s water use to data centers’... and it’s still a factor of 130.

Many people would also argue that even though AI’s water usage is small in aggregate, it’s concentrated where it hurts. This is somewhat defensible. Estimates vary from 32% to 66% of data centers that are built or planned in areas of high water stress.

But let’s apply the same logic to factory farming and animal feed. Ceres found that 35% of global-corn-based feed is grown in regions of high or extremely high water stress. Zoom out to rivers and it gets worse. Livestock farming alone pushes 7% of the world’s rivers past the freshwater planetary boundary, and in another 34% it would blow past that boundary even if every other industry vanished. This is a far worse problem because there are way more animal farms than data centers. Again, orders of magnitude worse. This is a real problem, and I hope local ordinances get better at protecting the specific communities located near data centers and factory farms.

“We find irrigation of cattle-feed crops to be the greatest consumer of river water in the western United States, implicating beef and dairy consumption as the leading driver of water shortages and fish imperilment in the region.” — Richter et al., Nature Sustainability (2020)

The Villain Edit, In Numbers

It’s a pretty open-and-shut case: when it comes to water usage, animal agriculture is a far bigger threat than data centers. But no one seems to believe it. It feels like every single conversation nowadays irrevocably circles back to the idea of wasting water – something that never seems to happen when we’re munching on hamburgers.

But you know me, feeling vibes isn’t actually enough to be sure. So I sat down and counted. I pulled about 400 water-and-industry stories from nine big US and UK papers — the Post, the Journal, the Guardian, and similar outlets — from July 2024 through June 2026, and read each one to see when they mentioned AI or Animal agriculture as a threat4. Here’s what I found:

AI gets framed as a threat to water roughly a third of the time. Animal agriculture? Only about one story in twenty. Toggle between the two tabs to see just how disproportionate this coverage actually is — you are nearly six times more likely to see an article about AI’s water use in a mainstream paper even though animal farms drink up 130 times more water. Oof.

While this data doesn’t come from a preregistered study, it maps nearly perfectly onto existing data. Several outlets and researchers have estimated that only a very small percentage — 1.5%, 3.6%, 3.8%, 4%, and 7% — of climate stories mention animal agriculture. My 6.7% figure sits nicely within that range.

It gets worse when I examined the type of coverage. Animal farms are usually portrayed as victims: stories where drought shrinks the herd, ranchers cull5 their animals, or your breakfast McMuffin gets more expensive. AI, meanwhile, gets the villain edit — “your cloud is drying my river,” as one Spanish activist put it.

If you don’t believe me, run the experiment yourself. Go to any social media site, search “AI water use,” then search “animal agriculture water use,” and start scrolling. It won’t take long before you see what I mean.

Anatomy of a Faulty Stat

This is…strange, no? I’d argue that a switch away from data centers, or an extreme reduction in their use, would change the fabric of industrialized society far more than swapping from animal-based to plant-based foods. Would you rather eat soy-based chicken nuggets and oat milk lattes or have to send carrier pigeons instead of DMs for the rest of your life? It’s worth investigating where exactly this all came from.

The viral “one email, one bottle of water” claim started with a genuine 2023 preprint: half a liter for a whole conversation of 10–50 prompts on GPT-3. Then, in September 2024, the Washington Post wrote up that data into a fancy infographic, where the idea started to make a splash. Funnily enough, that original post argued that training GPT3 used the same amount of water as the amount of beef two US Americans eat in a year. But this caveat only appeared at the very end of the piece, and readers didn’t seem to put the water usage of AI into that perspective — only the “one bottle of water” claim stuck.

From there, it spread like wildfire (you know wildfire, the stuff that animal agriculture is fueling). Tom’s Hardware put the worst case in the headline (three bottles!), a news TikTok with over a million likes carried it further, and Snopes, an actual fact-checking site, repeated it without much scrutiny. By 2026, Claude’s water use had clawed its way into becoming folk belief, beyond any attempt at numeracy. Some say that writing "please" and "thank you" to a chatbot is an act of environmental vandalism. TikTok is full of vague water-use claims. Brands are slapping it on billboards. Politicians are capitalizing on it. Millions of people now feel a small pang of ecological guilt over an amount of liquid that couldn’t even water a houseplant.

The punchline is that the researcher who produced the original figure has since walked it back to roughly 15 mL for GPT-4. This correction, needless to say, has not gone viral.

So What Do We Do With This?

I have two conflicting feelings seeing a fact like this spread so rapidly. First, and most obvious, the data scientist part of my brain is recoiling in horror at seeing some “alternative facts” slowly become part of the zeitgeist. It is unbelievably frustrating to see genuine political power6 rally around a low-key environmental villain when that energy could instead go towards fighting the number one cause of deforestation across the world.

But there’s another part of me that’s sitting up and paying attention. If a narrative like this can come out of seemingly nowhere — based on facts that crumble apart as easily as Nature Valley bars — what’s stopping us from doing the same thing7?

Part 2 of this story — where I will analyze what types of facts tend to go viral, and what we need to learn from them — comes out in a few days.