An argument you can check.
The point is the discussion
The purpose here is to spark discussion about the resources AI consumes — not as a method for controversy or shaming, but as a conscious, analytical and validated effort. Nobody makes a better decision because they were made to feel guilty. They make a better decision when they can see the trade-off, in numbers, with the workings attached.
That is also why this tool never gives a verdict. It reports a range and shows you where the range came from. What to do about it is your call, in your context.
Who it is for
The goal is to help professionals, academia and government identify the real impact of the solutions we deploy using artificial intelligence — with the focus on generative AI, given that it is the area with the highest growth of the last four years.
Those are the rooms where a number gets quoted once and then repeated for a year. It should be a number somebody can check.
Grounded in papers, not guesses
The data is supported by published papers and studies, not by raw estimation. You can verify how each estimate was made: how we calculate sets out the system boundary, the scaling, the assumptions, and — as plainly as we can manage — what is not counted at all.
Where this is still weak
A good deal of it rests on educated assumptions, because for some models the information is simply not published. That is not hidden in a footnote: it is what the confidence tiers are for, and a closed model can never be labelled measured here — the build fails if anyone tries.
A note from the person who built it
As an independent professional, I often find myself in conversations where the information is scarce, or misunderstood by the leaders and managers at the moment of taking a decision. So I built this tool as a reference — something to talk from, with arguments, about different scenarios, and to guide better decisions: technical, financial, and environmental too.
As stated, a lot of the information is based on educated assumptions, because for some models it is not available. But the goal remains understanding how our decisions about deploying AI go beyond an office or a data centre, and can impact people's lives.
Feel free to reach out if you have any comment or suggestion.
Best,
Danilo (Dan)
Corrections and comments welcome
All recommendations or comments are welcome. A correction that comes with a citation is the single most useful thing you can send — it is what moves the dataset out of seed status.
Write to
aicontentfootprint [at] origo dot ec
Written out rather than linked, so it stays readable to people and tedious for bots.