Takeaways, trends and notes from Green IO Paris 2025
I just got back to Berlin tonight after being at Green IO, a conference in Paris in its third year, that is dedicated to the fields of digital sustainability and Green IT. Before I forget, I figure it’s worth sharing a few takeaways from sifting through about a bajillion pics of slides, and all notes scribbled down over the last three days, for others and my future self. Off we go.
There was a lot of content at Green IO.
Day one was a full workshop day, which to be honest, felt more like a set of 25 minute talks aimed primarily at people already with some prior background in the intersection of AI and technology.
Day two had various sessions, but for people working in nonprofits dedicated to sustainability in tech there was an unconference that followed a variant of the OMG Climate playbook I used with some friends in 2019 for series of events in London and Berlin. That’s now the third time the initial work I had put it in to design an unconference back in 2018 has turned out to be useful. Result!
The final day was a more conventional single-track style conference, which was aimed at a wider audience, but still managed to have some extremely information dense presentations.
Anyway. Themes from the conference. Off we go.
Theme 1: More progress on getting actually usable standards and labels for Green IT / Digital
Barton Finn, at TCO Certified, presenting their new standard TCO Certified CloudIf you work in a digital team, and you’ve followed this field and tried to get anything changed where you work in terms of sustainability, then you’ll know how hard it can be. One of the challenges is that if you want to talk about how the environmental performance of a digital service is good or bad, it really helps to have a consistent way to measure something, in a way that other people will accept, and for that it really helps to have standards. Generally speaking, it feels like there is some progress here, and the tools or services using them are getting more sophisticated and usable.
I’ll intro a few specific projects or ideas worth your attention that were presented at the conference, and why I think they’re worth it.
Energy Score AI
Energy Score AI is an open source project intended to provide a consistent, understandable way to compare the performance of different AI models from different providers, in terms of how well they do against a set of predefined benchmark, but also in terms of how energy efficient they are.
So, if you are building a service that uses AI models somewhere, and you want to see what your options in terms of models for carrying out a specific task, you can do so, and understand the tradeoffs you might make based on cost, energy consumption and how well each model actually carries out the task.
The team working on it, led by Boris Gamazychikov and Sasha Luccioni, launched v2 recently. If you are a buyer of AI services I think it’s the best chance we have of having any kind of meaningful comparability across providers, which is good for you as a customer. The project leaderboard is comprised almost entirely of open source / open weights models , but there are ways to use it with proprietary models too. They have sample language for using it in your contracts if you want to require it in commercial agreements.
TCO Certified for cloud
Another new initiative was a new eco-label from TCO Certified, a certification body based in Sweden, who have been creating various kinds of eco-labels for the past 30 years.
Given that this is the tech sector, and the role procurement plays in sustainability, you might assume TCO here stands for Total Cost of Ownership – in this case it doesn’t, but instead comes the name of a workers union that was the original parent organisation Tjänstemännens Centralorganisation (or literally, White-collar workers’ Central Organisation).
But I digress. This is a new eco-label standard for cloud services, and it covers lots of the criteria you’d want to care about as a buyer of cloud services, like meeting minimum levels of quality for carbon intensity, embodied carbon, transparency around resource usage, water usage, utilisation and so on.
There is an immense amount of jargon in Green IT, so having a meaningful multi-criteria label for cloud services is a useful thing to have in the public domain. The actual criteria used by the label come from an existing set of standards for cloud services and data centre IT hosting services that have been established via a process led by ADEME, the French government Agency for Ecological Transition.
ITEE and ITEU – meaningful metrics for understanding how much compute you’re really getting when you buy servers
Remember when I said there was a lot of jargon in Green IT? This here is one example. I’ve been in and around Green IT for a few years now, and I only found out about ITEE and ITEU in November this year. Moreover, it took me watching almost the same talk, by the same presenter, Rich Kenny twice to finally get my head around it.
But the short version is that ITEE – refers to IT Energy Efficiency, which you can largely think of as a normalised score for how much computation a server can do per unit of energy. A high score means a server is able to do a lot of work for the power it uses, so if you have a set amount of computation you need to do, if you have servers with a high ITEE, you can do it with fewer servers and less total energy consumed.
You might ask how is this a new insight? Don’t you get this with new kit anyway?
You’d be forgiven asking this, but because relative improvements have decreased with each generation of server over the last few generations, it turns out that in a lot of cases, you can get a comparable ITEE improvements just by upgrading components in a server like RAM and CPU, instead of having to replace the whole thing.
Previously, there was something of a default assumption that the cost of energy consumption improvements for a fleet of servers would come in the form of having spending lots of money on new servers, which has it own significant environmental footprint, this way you can still get the energy consumption savings.
The other metric ITEU, IT Equipment Utilisation, is a measure of how much of a server’s actual capacity is being used. There are various flawed ways to track how much capacity of a machine is really being used, and my takeaway from the talk was that this was a more reliable composite metric that manages to avoid some of the pitfalls they have.
Combined with ITEE, it gives you two useful metrics to see how you might change the total fleet energy consumption for the amount of computation you need to support, at the lowest cost and lowest embodied carbon, by:
- repurposing underused high ITEE servers to do more work (increase the ITEU to make use of the the already high ITEE),
- upgrading medium ITEE servers (i.e. increasing their ITEE at the lowest environmental and financial costs)
- taking older ones that don’t have much prospect of increasing ITEE out of service (i.e. reallocating their ITEU to the faster, more efficient servers)
The downside – hard to find examples
I had a chance to ask a question after Rich’s talk about who is using this, and how you can use these metrics. At present, I don’t think any cloud providers actually expose these numbers for their own cloud servers (definitely not the big three, AWS, GCP or Microsoft), but there is some good news. It’s now written into UK government procurement standards, and it’s also in the European Energy Efficiency Directive, so assuming people follow the law, this is a metric that ought to be disclosed. However, I’m not aware of any orgs helping calculate these metrics other than Interact the organisation Rich Kenny runs. If there are, I’d love to know.
Here’s the link to Rich Kenny’s deck, and this here is the link to Interact’s library of papers going into more detail about ITEE and ITEU (search for “Maximising Server IT Energy Efficiency through optimal Hardware and Software configuration”).
Theme 2: AI – “There’s always a button for yes, but no button for no”
Anaëlle Beignon presenting her talk, “Imposing AI: Deceptive design patterns against sustainability” showing AI features being boosted in 4 different places on the same screen
There were a fair few AI-skeptic talks at Green IO. One of the most interesting for me was Anaëlle Beignon’s presentation detailing how AI is being incorporated into existing services, and how it’s changing their own user experience. It’s really helpful to have a a UX specialist explicitly take people through the various techniques used to drive usage and adoption. I’ll be coming back to it a fair few times in future I think.
Elsewhere on the AI front, James Martin, formerly of Scaleway also had an entertaining, punchy but above all practical talk about how to respond to FOMO driven edicts to adopt GenAI – if you’re in a situation where usage is unavoidable, there are definitely strategies you can use to minimise the direct environmental impacts of doing so. His presentation came in the form of a set of tactics to use to debunk the main myths used to justify any form of remedial action, and he was the source of the “there’s no button for no” quote.
Theme 3: Positive impacts of AI – Oh wow, that’s how they calculate it?
Theophile Lenoir, presenting “Limits to the evidence on AI’s climate benefits and their implications” at Green IO
After working on various kinds of analysis at work about the direct impacts of AI, I had seen various critiques of the claims around AI for Good / AI for Climate, but the closing talk by Théophile Lenoir of the Observatory on the Environmental Footprint of AI, was one thing that really blew my mind. He took us through how the positive claims are created and what kind of bases are used to model various kinds of savings, and well…
… I had some idea of how vibes based they were, but when he spelled it out on a huge screen, it really hits home how flimsy the evidence base really is.
Here’s his deck. I understand there’s some new research coming from the organisation in the first quarter of 2026. He’s worth following on Linkedin to find out when it’s published if this caught your eye.
Theme 4: Green IT NGOs in Europe in the same room for the first time
A pic from Nathalie Otte’s post about the European Green IT NGO unconference at Green IO
The final theme was something I have to confess that I hadn’t really appreciated while at the event. If Nathalie’s post is correct, this is likely the first event explicitly set up to get Green IT NGOs in Europe together, talking and collaborating in recent years.
Given how various sustainability and transparency laws in Europe are being dismantled right now, this can’t come soon enough, and there were a good few follow-on projects coming out of the event that I’m keenly following.
Wrapping up
If you’re interested in getting involved in the Green IT scene, I really suggest keeping up with what Green IO is doing. What began as a one-off event has grown into a something spanning multiple countries and continents, and I can’t think of any similar event, where you have so much access to people working on these problems, that you can get to for free if you time it well.
The easiest way to do so is head over to greenio.tech, and sign up for the newsletter, and subscribe to the podcast. There you are – an easy new year’s resolution, right there.
#AI #events #greenio