Your washing machine has an energy label. Your AI chat doesn’t.
We are asking the European Union to require AI providers to show the energy use of every chat. In the app. In the moment.
this evening, so far
Roughly equal to
Here’s what that could look like:
Example from inside the chat
Done — here’s a 5-second clip, 1080×1920.
Conversation usage
Example:
Say someone is running a half marathon and looking for sponsors.
- Fifteen questions to the chatbot: how to approach sponsors, what a realistic target amount is, how to write a message people won’t swipe away.
- Then the flyer. Three images before one is good enough.
- And finally, one five-second clip for Instagram.
Source: MIT technologyreview.com
That’s what that evening cost.
That number was nowhere to be seen. Not while working, not afterward. It wasn’t in the app, not in a setting, not in a monthly summary. Nowhere.
No one knew the clip was the most expensive part. No one knew fifteen questions cost almost nothing. There was no reason to think anything of it, because there was nothing to see.
How many loads of laundry did you run tonight?
You don’t know. And that’s no accident.
The numbers above come from measurements on open models — software researchers can download and measure themselves.
The chatbots hundreds of millions of people use every day are not open. ChatGPT, Claude, Gemini and Grok publish no numbers. Not to users, not to regulators, not to researchers.
MIT Technology Review asked OpenAI, Google and Microsoft directly. All three declined to share numbers.
"The closed AI model providers are serving up a total black box."
— Boris Gamazaychikov, Salesforce
"We should stop trying to reverse-engineer numbers based on hearsay, and put more pressure on these companies to actually share the real ones."
— Sasha Luccioni, Hugging Face
The Lawrence Berkeley National Laboratory, which forecasts power demand on behalf of the US Department of Energy, reached the same conclusion: so little is disclosed that a decent forecast can’t be made. Data centers are being built without anyone able to calculate what they’ll cost the grid.
This isn’t a technical problem. Companies measure this internally down to the kilowatt-hour — they have to pay the bill themselves. They just don’t publish it.
This has been done before
Europe has had an energy label on your fridge for more than thirty years. Later it appeared on washing machines, light bulbs, TVs, cars and homes. Flight tickets show CO₂ emissions. Your box of cookies shows what’s in it.
The argument was always the same: if you use something, you get to know what it costs. And every time, it turned out to be possible.
No one today finds it unreasonable that a fridge has a label on it.
Only for the fastest-growing power draw right now does no one know a thing.
We are asking the EU to require AI providers to show the energy use of every chat.
Sign the call4 people signed before you
What we’re asking for
One thing, and nothing more: make the energy use visible the moment it happens.
- The energy use of this answer, and of this conversation, in kilowatt-hours
- A comparison that means something — laundry, driving, a fridge. A bare number helps no one, as shown above
- Where the power came from: what share was renewable
- On by default. Not buried in a settings menu
To whom: to the providers themselves — OpenAI, Anthropic, Google, Meta and xAI — and to the European Commission and the European Parliament, who can write the rules. This call runs outside a formal petition process: its weight comes from numbers and attention, not procedure.
Why a law, and not a polite request: this has been tried voluntarily for three years. The AI Energy Score, a label companies can voluntarily have their models rated on, has existed since 2024. Almost no major company takes part. As long as no one has to, no one will, because the first company to open up looks worse than everyone else, alone.
A requirement removes that disadvantage. Everyone at once, or no one.
Support this? Add your name.
“But what about…”
“Isn’t AI nothing compared to flying or the meat industry?”
True, today. Data centers now use 4.4% of all power in the United States. By 2028 that’s estimated at 12%, and more than half of that goes to AI. This is the only line on the power bill tripling in four years. That’s exactly why now is the moment to make it measurable — not after the fact.
“Does a number on screen really change what people do?”
Maybe not right away. But the fridge energy label mainly changed what manufacturers built, not what buyers chose. Visibility makes comparison possible, and comparison makes efficiency worth pursuing. Right now there’s no reward for an efficient model, because no one can see the difference.
“This is a trade secret — this would hurt innovation.”
The question isn’t how a model works or how many parameters it has. Only how much power went through the meter. That’s not a trade secret, that’s a bill. Carmakers publish their fuel consumption too, without taking their engines apart.
Who’s behind this
This call comes from The AI Footprint Meter — an independent initiative by people who want AI use to be just as measurable as any other form of power consumption.
- MIT Technology Review, Power Hungry: AI and our energy future, May 2025 — technologyreview.com
- Lawrence Berkeley National Laboratory, United States Data Center Energy Usage Report, December 2024 — eta.lbl.gov
- ML.Energy leaderboard, University of Michigan — ml.energy
- AI Energy Score, Hugging Face and Salesforce — huggingface.co
Every figure on this page can be checked. Spotted something wrong? aifootprintmeter@proton.me
What happens with your signature
- Now — collecting
Goal: as many signatures as possible, shown on this page.
- Next — the companies
The call goes, with every name attached, to OpenAI, Anthropic, Google, Meta and xAI. Their responses will be posted on this page. Even if they don’t respond.
- Ongoing — press and numbers
Without a formal petition process, this call works through visibility: media, other organizations and repeated pressure on providers.
- Ongoing — the list
Every company that adopts this gets listed here.
This page will be updated. Whoever signs hears what happens.
