Key takeaways
- I measure my daily AI footprint in energy, CO₂ and water instead of ignoring it.
- Around 50 kg of CO₂ in five weeks, roughly the same as 210 km by car.
- Transparency is no free pass, but it is the honest first step.
- The measuring tool, AI-Foodprint, runs locally and is open source.
I donate to climate projects. I am involved in the Klimakino, because I believe stories move people faster than statistics do. And every morning I start the AI and let it run all day, because it makes my work for clients many times faster. Both are true. Holding both at once sometimes feels like the splits.
The contradiction I won't talk away
It would be convenient to say AI barely uses anything. That just isn't quite true. Every request pulls electricity, cools data centers with water, and hangs off a grid that is far from clean. Use AI all day like I do and it adds up.
Downplaying that would be exactly the kind of greenwashing I criticize in others. So I don't downplay it. I measure it.
Why I still use AI every day
Because it makes the difference for my clients. What used to take days is ready in hours. Efficiency is itself a form of saving resources: a result in a fraction of the time means less idling, less duplicate work, fewer trips.
But that is only an honest justification if I also know the other side of the ledger. Otherwise it's an excuse.
What I do about it: measure instead of spin
So I built myself a tool that records every day what my AI use costs. Not in euros, but in energy, CO₂ and water. It is called AI-Foodprint, it runs locally, and the code is open.
Every session is measured, the tokens are converted into kilowatt-hours, grams of CO₂ and liters of water. Abstract numbers turn into tangible comparisons: kilometers by car, phone charges, days a tree needs to sequester it again.
The numbers I don't round in my favor
Over roughly five weeks I had racked up a good 50 kilograms of CO₂, about 135 kilowatt-hours of electricity and almost 500 liters of water. That was about 210 kilometers in a combustion car, when I wrote this article.
The dashboard has been running ever since. As of July 2026, the measured period, since I started tracking Claude (late April 2026), adds up to around 66 kg of CO₂, just under 170 kWh and just under 590 liters of water, equivalent to roughly 265 km by car. The numbers keep growing daily and are embedded live below.
My deal with myself
Transparency is no substitute for engagement. The dashboard doesn't make my footprint smaller, it makes it visible. My deal: I use AI where it creates real value, not for play. I stay engaged for nature and the environment without buying my way out. And I put the numbers in the open, even when they're uncomfortable.
Anyone rolling out AI in their company should ask the same question: do you actually know what your AI use draws in resources? If not, measuring it is the first honest step.
My AI footprint, live
Open the full view or work out your own footprint with the open code on GitHub. Claude Code and local models (Ollama) are measured; the web tools (Gemini, Notion, Wispr Flow) are rough estimates I add by hand each month.
