How AI uses our drinking water - BBC World Service
Five Drops of Water per Prompt: The Dirty Secret of “Green” Artificial Intelligence
Google wants you to think artificial intelligence is the savior of humanity. The company finally admitted that its Gemini app—used by more than 450 million people per month—slurps down energy and water with every prompt.
Until recently, the actual numbers were treated like state secrets, locked away tighter than the nuclear launch codes.
Now, under mounting pressure from scientists and activists, Google has cracked the door open. But only a sliver.
And what do we see through this crack? PR-friendly numbers polished to a shine.
Google claims that a “typical short text query” in Gemini consumes:
-
0.24 watt-hours of electricity
-
0.26 milliliters of water
-
0.03 grams of CO₂ equivalent
That, they say, is like watching nine seconds of TV or wasting five drops of water (NZZ, Aug. 2024). Five drops. Cute, right? The PR machine couldn’t resist this poetic comparison, dripping with corporate innocence. Five drops. Who wouldn’t sacrifice that for AI-powered poetry, shopping lists, and dubious dating advice?
But here’s the part they don’t tell you:
Those “five drops” don’t include image generation, video generation, or long reports.
And those are exactly the kinds of tasks that turn servers into electricity-guzzling, water-chugging beasts.
Jeff Dean, one of Google’s AI gurus, admitted as much to MIT Technology Review (source): upload a stack of books and ask for a summary, and the consumption skyrockets.
So why the Disney version of the numbers? Because Google, like every other tech giant, is desperate to maintain the myth of AI as a “green” miracle machine.
Meanwhile, the servers sweat in their warehouses, guzzling water for cooling systems and pulling electricity from grids still fueled by coal and gas.
The Water Problem No One Wants to Talk About
Google’s own Sustainability Report admits that its data centers increased their total water use by nearly 30% between 2023 and 2024 (Google 2024 Report). Thirty percent.
But don’t worry, says Google: efficiency is up! Which is a bit like saying, “Sure, I drink more whiskey every night, but I’m using a thinner straw.”
Researchers like Shaolei Ren (University of California, Riverside) point out that Google only counts the water used inside its data centers.
It ignores the water used in the power plants that actually generate the electricity (Ren’s analysis on LinkedIn). If those plants run on hydropower, the real water footprint balloons even more. That’s not a detail—it’s the elephant in the server room.
Without standardized reporting, comparing Google’s sanitized numbers to those of other companies is like comparing organic apples to radioactive oranges.
The Carbon Sleight of Hand
Then there’s carbon. Google pats itself on the back, insisting emissions are “low.”
But buried in the fine print is a trick: they offset emissions by purchasing renewable energy credits, essentially creative accounting for pollution (NZZ). It’s like eating three Big Macs, jogging for five minutes, and calling yourself “net healthy.”
AI is hungry, and the bigger the model, the bigger the appetite. Training a large model is like feeding a starving mammoth Red Bull 24/7.
Once trained, these beasts never stop eating—they must be powered constantly to spit out our texts, memes, and “Which Hogwarts House Are You?” quizzes.
And yet, Silicon Valley CEOs keep selling AI as the messiah of sustainability. “Green intelligence,” they call it.
Except the only thing green here is the PR slime coating the press releases.
The French Did It Better
In July 2024, the French AI company Mistral actually released numbers that made sense: the environmental cost of generating a 300-word page of text (Mistral blog).
Not vague “five drops” metaphors. Hard, tangible data. That’s transparency. Google’s global empire? Still hiding behind smoke, mirrors, and water droplets.
Where Is the Green Intelligence?
Let’s cut the nonsense. If we’re serious about sustainable AI, we need:
-
Full standardized reporting – No more cherry-picked “five drop” fairytales.
-
Accounting for ALL resources – Energy, water, training costs, cooling systems, AND the upstream impact from power generation.
-
AI that powers itself – Yes, I said it. If AI is so “intelligent,” why can’t it run on its own recycled energy loop? Solar panels on every data center, heat recovery systems powering nearby homes, water cooling that reuses wastewater instead of draining rivers.
If AI is supposed to be the future, then let it invent its own damn power supply. Until then, it’s just another greedy, fossil-fueled junkie in a shiny hoodie.
The Punchline
So next time you type a prompt into Gemini, imagine those five drops of water falling.
Not onto your thirsty houseplant, not into the parched soil of a drought-hit farm, but into the belly of a data center in Iowa or Frankfurt, where servers sweat so you can ask, “Write me a poem about my cat in the style of Shakespeare.”
We keep pretending this is progress. But maybe the question isn’t what AI can do for us—it’s what we’re willing to sacrifice for AI.
And right now, the sacrifice looks like water, energy, and honesty.
Where is the “green intelligence”? Because all I see is an industry spraying perfume on a sewage leak.
Maybe it’s time to demand not just artificial intelligence—
but artificial responsibility.
“Efficiency is meaningless if the overall footprint keeps growing.” — Shaolei Ren, UC Riverside
💀 Five drops per prompt. Multiply that by billions. Then tell me again this is sustainable.
yours truly,
Adaptation-Guide
