You type. You hit send. An answer pops up like magic. Feels like nothing and costs you nothing, or so you’d think. But here’s the uncomfortable truth: the hidden cost of every AI prompt you send is very real, and it’s measured in electricity, water, and carbon.
That breezy little message you fired off? It just reached into a power grid and a water supply somewhere you’ll never see. Once you clock the environmental impact of AI, you can’t unsee it.
- One ChatGPT prompt swallows around 0.34 watt-hours of energy and 0.322 mL of water.
- Users fire off roughly 2.5 billion ChatGPT prompts a day.
- Data centre electricity use could top 945 TWh by 2030, more than all of Japan.
- The UN warns AI data centres could one day drink as much water as 1.3 billion people.
- Even saying “thank you” costs OpenAI tens of millions a year.
What a Single Prompt Actually Costs
Sounds harmless, doesn’t it? Every time you poke a question into ChatGPT, Gemini or Copilot, you’re spending real electricity and real water.
Forbes did the sums: one ChatGPT prompt burns about 0.34 watt-hours and 0.322 mL of water, going off figures shared by OpenAI boss Sam Altman. Gemini’s a touch leaner at 0.24 watt-hours and 0.26 mL a pop.
Well, multiply those droplets and sparks by billions of queries a day and suddenly you’re staring at terawatt-hours a year. That’s more energy than some entire nations get through. The little numbers are the trap.
Also read: Building In-Demand Tech Skills After Graduation
How Much Electricity Does AI Use?
The International Energy Agency’s April 2025 report predicts that global data centre electricity use will surpass 945 TWh by 2030. It is the current usage of Japan. And AI is the main cause, with demand from AI-tuned data centres set to more than quadruple.
The pace is borderline absurd. The Intelligent Computing Journal flags that the computing power needed to keep AI growing is doubling every 100 days.
OpenAI hit 700 million weekly users in August 2025, and per a July 2025 TechCrunch article, those users send 2.5 billion AI prompts a day. Now add in Microsoft cramming AI into Office and Google wedging it into Gmail and Search, too. The scale is genuinely hard to picture.
Also read: Apple’s £599 MacBook Neo Sounds Great—Until You Read This
The Water Problem Nobody Mentions
AI chips get blisteringly hot, and to cool them, water is used. NDTV World points to a chilling United Nations warning: by the end of this decade, AI data centres could use as much water as 1.3 billion people. No wonder communities near these sites are dry and bills are increasing.
NVIDIA thinks it has solved part of the problem. Its new servers pump a liquid coolant through a closed loop. Picture your car’s cooling system that is soaking up heat and whisking it away without forever slurping fresh water. It works at up to 45°C, so you can do without the energy-guzzling chillers.
NVIDIA says big sites could save millions a year. Lovely. Except the kit costs a fortune, ripping out the world’s old infrastructure will take ages, and the chips still need mountains of electricity, which itself can burn through water to be made. So, no silver bullet.
Also read: The AI Skills That Will Be Most Valuable in 2026 for Real Jobs
Even Your “Thank You” Adds Up
The Hindu reveals that even a quick ‘thank you’ wakes up the whole neural network, with a simple interaction sipping 0.1 to 0.3 watt-hours, roughly the same as a 10-watt LED bulb glowing for about 30 seconds.
And get this: Altman has admitted that polite little “please” and “thank you” messages cost the company tens of millions of dollars a year in electricity. Being nice has never been so pricey.
A 2024 Lawrence Berkeley National Laboratory report found that US data centres already munch through about 4.4% of the country’s electricity, possibly rocketing to nearly 12% by 2028.
McKinsey reckons global data centre spending could hit $6.7 trillion by 2030. And despite splashing out on clean energy, Google’s own data centre electricity use has more than doubled in just four years.
Also read: How Do High Heat Retention Storage Heaters Work
The Stat Everyone Gets Wrong
You’ve seen the comforting comparison. Training GPT-3 used about 5.4 million litres of water, while California’s almond farms slurp 1 to 2 trillion litres a year. AI’s basically a rounding error.
Not so fast. Training happens once. Inference, the model actually answering you, happens 900 million times a week. Weekly Opportunities does the maths: with ChatGPT’s roughly 900 million weekly users and OpenAI’s published 0.3 mL per query, researchers reckon the true figure is five to fifty times higher once off-site cooling’s counted.
That’s 4.4 to 44 billion litres of water a year from ChatGPT alone. Firms bang on about training because it’s a tidy, bounded story. Inference? That keeps climbing.
Why “More Efficient” Won’t Save Us
You’d assume slicker chips and smarter cooling would sort it. They help, but they can’t outpace the stampede. It’s the old Jevons paradox: make something cheaper and easier to use, and people just use loads more of it.
One prompt might sip a fraction of a watt-hour, but billions a day pile up into a thumping great power and water bill for the planet.
Also read: How to Use a Spark Plug Tester Like a Pro
What Can You Actually Do?
Don’t panic-quit AI. Just be a bit smarter. Reach for smaller models like Claude Haiku over Opus and GPT-5-mini before GPT-5 and save the big guns for jobs that really need them. Pause before you generate: if no one’s going to read, use or remember the output, skip it.
Lump related questions into one prompt instead of rattling off ten. And if your boss is buying AI in bulk, ask what the environmental audit actually says. Every prompt is a resource decision, and it belongs to all of us.
Sources & References
- Forbes – One ChatGPT prompt burns about 0.34 watt-hours and 0.322 mL of water.
- NDTV World – The United Nations warns that by the end of this decade, AI data centres could consume as much water as 1.3 billion people.
- The Hindu – A simple word like ‘thank you’ can wake up the entire neural network of AI.
- Weekly Opportunities – ChatGPT has roughly 900 million weekly users.
Disclaimer: The figures and estimates discussed in this article are based on publicly available reports, industry research and company statements available at the time of writing. AI energy and water consumption can vary significantly depending on the model, infrastructure and usage patterns. Readers should consult original sources for the latest data and developments regarding the environmental impact of AI.
