What does your prompt actually cost?
Every AI request draws real electricity and real water. It just never shows up on a bill you can read. Pick what you generate, and see what it equals in the physical world.
Generated images, upscales, and reworked variations.
One full battery, 0 to 100%.
Build your habit
Enter what you use in a typical day, and see the total add up over a week or a year.
The Chatterbox
Chat replies, quick drafts, everyday questions.
The Coder
Code generation, tool use, long outputs, retries.
The Artist
Logos, avatars, concept art.
The Director
Short clips and animations, per clip.
Loading…
0phone charges
0bulb-minutes
0miles driven
0litres of water
0plastic straws made
0tree absorption daysWhat it looks like in real life
Vivid ways to picture the invisible.
One ghost phone charge
Generate around 44 images and you've spent as much electricity as charging a phone from empty. A single busy afternoon of logo concepts can rack that up without you noticing.
An evening's worth of light
Ten rounds of back-and-forth with a Reasoning Agent, the kind of editing session that goes in circles, burns enough power to keep a bulb lit for the rest of the evening.
A trip to the shops
A one-minute AI film in Reasoning mode can carry roughly the same tailpipe emissions as driving to the local shop and back. All for footage nobody drove anywhere to get.
A shot glass, evaporated
About thirty reasoning-mode prompts, roughly a long brainstorming session, evaporate a shot glass of fresh water to keep the servers from overheating.
How to shrink your AI footprint
Six small habits that actually add up.
Be specific the first time
A vague prompt gets a vague answer, which usually means a second attempt to fix it. The more detail you give upfront, tone, format, the exact ask, the more likely the first result is the one you actually keep.
Match the tool to the job
Reasoning Agent mode exists for genuinely hard problems, not for fixing a typo or rephrasing a sentence. Save the heavier model for the moments that actually need the extra thinking.
Edit, don't restart
If a result is close but not quite right, ask for the specific change rather than regenerating the whole thing. A full re-run spends the entire cost again; a targeted edit spends a fraction of it.
Treat images and video as material
Generate on purpose, the way you'd use any other resource you can see running out. A dozen quick throwaway variations can cost roughly as much as one carefully planned image.
Know when to stop
The tenth variation rarely beats the third. Give yourself a rough cap before you start generating, the same way you'd budget paint, film, or any other physical material.
Reuse before you regenerate
A good output doesn't expire. Save it, reuse it across drafts and projects, and only generate a fresh one when the old one genuinely won't do.
Understanding AI's environmental impact
Why a calculator like this exists in the first place.
Search interest in green AI has climbed fast, and for good reason. Every prompt, image, and video you generate draws power from somewhere, and that AI energy consumption adds up across billions of requests worldwide every day. The Invisible Cloud translates that abstract AI power consumption into things you can actually picture: a phone charge, a lit bulb, a mile of driving.
The environmental impact of AI isn't limited to electricity either. Running large models depends on data center energy usage that also demands real water for cooling, and that electricity is tied to the same grid behind a country's wider carbon footprint of AI. As generative tools scale, questions about AI sustainability and AI's climate change impact are no longer a niche concern. They're becoming part of how people think about the tools they use every day.
This site doesn't ask anyone to stop using AI. It asks a smaller question: what if the cost were visible? Enter your own habits above, choose Standard or Reasoning Agent mode, and see what your own AI energy consumption actually adds up to, sourced from published research rather than guesswork.
How fast AI is growing
The scale here is worth sitting with. The global AI market is projected to grow from roughly $189 billion in 2023 to nearly $5 trillion by 2033, a 25-fold increase in a decade, according to a 2026 report on AI's environmental cost from the United Nations University's Institute for Water, Environment and Health (UNU-INWEH), led by Professor Kaveh Madani. Generative AI alone made up over a fifth of the global AI market in 2026 and is projected to reach 40% by 2030. Global AI expenditure is projected to exceed $2.5 trillion in 2026, with corporate AI investment topping $580 billion in 2025.
What it costs to train a frontier model
Training happens once, but at enormous scale. GPT-4's training run is estimated to have used 50 to 70 GWh of electricity and produced around 25,000 tonnes of CO2e, work that would take roughly 420,000 tree seedlings a decade to offset, about the number of trees across 105 Hyde Parks. Its water footprint came to about 600 million litres, enough to fill 237 Olympic-sized pools. Projections for GPT-5-class models put training electricity closer to 100 GWh, with a water footprint near 1 billion litres. None of that touches actually using the model afterwards. It's the bill before the first prompt is ever typed.
Inference: the part nobody sees
Training gets the headlines, but per the same UNU-INWEH report, the ongoing inference phase, generating responses for billions of everyday interactions, accounts for an estimated 80 to 90% of AI's total energy use. ChatGPT alone is estimated to process around 2.5 billion prompts a day. A typical AI text query is roughly 200 times more energy-intensive than a simple spam filter, and image and video generation sit well above that again. That's the exact gap The Invisible Cloud is built to close: not the training run nobody sees, but the everyday habit that quietly adds up.
Data centers, at a planetary scale
Zoom out further and data centers become infrastructure in their own right. Global data centers consumed an estimated 448 TWh of electricity in 2025, enough to rank 11th in the world if data centers were their own country, with a carbon footprint of around 189 million tonnes of CO2e. By 2030, that could exceed 945 TWh, close to 3% of projected global electricity use, as AI's own share of data center power climbs from roughly 20% to 40%. The report also flags a less-discussed cost: the hardware itself, projecting up to 2.5 million metric tons of e-waste a year by 2030, equivalent to discarding around 250 Eiffel Towers annually.