I see the benefits of using AI, but I’m very aware of its environmental impact.
So, I searched to see if there was a possible way to dramatically reduce my impact when using AI, and it turns out there’s quite a few ways to minimize your impact with AI. Let’s talk about the main factors :
PS : In this post I focus on the environmental impact of AI, but, of course, there are a lot of other problems like harmful bias, stereotype propagation…
Do you need AI?
The most effective way is to not use AI, of course. I know that you’re like “Dude, of course! But I NEED AI”. Fair enough, but from my experience, once we start using AI, it very quickly becomes something you use for EVERYTHING.
Need to do a calculation? Let me check on ChatGPT.
Need to search for new software? I’m gonna to see what ChatGPT will tell me.
Need to find the movie you’ll watch tonight? My ChatGPT BFF will find it.
You get the point. We use AI for everything. Sometimes our use of AI is relevant, other times not so much.
If you need to do a simple calculation, you can use your calculator.
If a simple Duckduckgo search will do the trick, then do it.
If a friend can give you a good recommendation, maybe ask him/her.
Choosing the right model size
Not every model has the same size.
Fable is bigger than Opus which is bigger than Sonnet.
The bigger the model is, the more it consumes. Even for the same task.
If you ask Fable and Opus: How much 2+2 is equal to? Fable will consume more than Opus for this same task.
So, if you need to do a simple task, like writing a post or correcting a spreadsheet, don’t use a big model, it’s way overkill, and it will not produce better results (in fact, there’s some evidence that using bigger models for simple tasks creates less quality output than using smaller models)
Choosing the location where your model run
A big factor in how much your AI usage impacts the environment is if your AI is powered by petrol, gas, nuclear or renewable energy.
Renewable/Nuclear are WAY cleaner options to produce electricity than petrol or gas.
So if your AI chatbot runs a data center which is powered by fossil fuel, it has a MUCH BIGGER environmental impact than the same model that runs in a data center powered by nuclear energy, which is low-carbon emissions.
How do you know if your AI runs in a clean or dirty data center ? Well, if you use the major AI chatbots (ChatGPT, Claude, Gemma, Grok…), assume that it runs in dirty data centers.
These companies are so in need to accommodate their spectacular growth that they build data centers everywhere even where there’s no option to produce clean electricity. In fact, it’s almost always the case.
But do not worry, you can choose other AI providers or directly choose where your model runs.
Choosing cleaner AI providers
A note about open weight models
You’ll see that almost the entire list of tools below use open-weight models. So what are they?
At their core they are models, like Opus, GPT-5… BUT unlike proprietary models, you can download, install them and run them on your computer/server. They are kind of similar to open source software in the way that you can download them and install them on your computer/server but they’re also not really open source because they don’t let you inspect everything, like with which data they have been trained onto and stuff like that.
Still, it’s nice to be able to run on your own infrastructure a model, because you can make it run on an infrastructure powered by renewable energy or nuclear power. So you greatly minimize the environnemental impact of your AI.
Another nice thing is that their are plenty of small/medium models that are open-weight. So you can have a small model running on a server powered by renewable energy, which dimunish further the negative environnemental impact.
Maybe you’re thinking? But, I’m not a developer, I don’t know how to host a model! Don’t worry, because you don’t have to! Some companies do that for you! Check the tools I recommend below:
Euria
Euria is proposed by Infomaniak, a Swiss hosting companies that also propose an AI Chatbot and AI models accessible via their API. Their datacenters are based in Switzerland and run on renewable energies. It’s one of the leading sustainable hosting companies out there.
Euria/Infomaniak are not developing their own models but rather use models like Qwen and other open-weight models. Depending on what you ask Euria will use a model rather than another. Which means that it can use the right model for the right task, to not overconsume.
Another benefit : Euria/Infomaniak do not use or sell your queries/data. If you need confidentiality, it’s a good choice!
Honestly, if you have basic use of AI chatbot, this one can totally replace your ChatGPT, Claude…
If you need something more advanced, with MCP connection, for example, then take a look at Mistral or Hugging Face that I talk about below.
Thaura
Have a broader mission that the environnemental one, they are also trying to create an ethical AI that do not negativaly impact people (regarding opression, privacy), you can read their ethical framework here.
On the environemental level, Like Euria, they use an open-weight model (Qwen) and they run it on their own GPUs powered by renewable energy in Finlande.
Unlike Euria, the tool is more complete with possibility of doing coding, creating interactive reports… I never tested it, but I seriously need to!
GreenPT
Never tested this one, but it seem like a good option. Like Euria or Thaura, they’re not developping their own models, instead they run open-weight models like GreenL (from Mistral) and Gemma (from Google).
The models runs in datacenters located in Paris, powered by renewable energy. It’s hosted on Scaleway : not sure how much this hosting is environnemental friendly, I need to do more check on them, but still, I believe, using GreenPT is vastly better than using ChatGPT or any major AI chatbot out there.
One thing that seem really cool about GreenPT is that they let you see your monthly consumption, which is awesome to have a scale of your AI usage impact.
Mistral
French AI provider with all of its datacenter in Europe (mostly France, soon in Sweden), so they all run on nuclear power.
Unlike Euria and other options above, Mistral do train and develop their own models.
The interface is great, the models are capable. The one thing I find frustrating is that they do not let you pick your model. Maybe your query is automatically matched with the best model or it automatically use the big model all the time. I don’t know. Regardless, even when using a big model it’s still better than using the lighter model of Claude as Mistral infrastructure runs on nuclear power.
Otherwise the tool is fully featured with possibility to do coding and other advanced stuff.
HuggingChat
Hugging Face is the home of open-weight models, and they have a chat that let you use these models. Either you can let the Chat automatically pick the best model deping on your task or you can choose manually between plenty of large, medium or small models. You can also choose were they run, because Hugging Face is not running these models, it’s only an interface between you and an hosting company that run the model. So not all the hosting you’ll find here are environnemental friendly, but I believe that some are and having the ability to chose a small model is still interesting in itself.
Beaware: it’s not a traditionnal AI chatbot. You are paying an hosting to run the models on their server, so you’ll pay depending on your usage (aka: you pay your token consumption). Which means that the more you chat, the more you’re paying, so it will usually be more expensive than a ChatGPT subscription. It’s a good way to be more thoughful of your usage tho 😛
Other great resources
Bearing
A tool that tell you which is the best model for you depending on how you use AI. Maybe you don’t need a large model?
Photoroom
Generating images is so resource intensive! Don’t do it unless you really need to 😉 AND if you need to, Photoroom seem a good option, because they use models that are specialized on image processing, which greatly minimize power consumption compared to a generalist model like ChatGPT, Claude…
Photoroom have written about their sustainable approach in this article and about their broader impact on this page.
Better Tech
It’s not a tool, but a blog about AI sustainability. I learned a lot from it and I discovered some of the tools mentionned here on this blog. So if you want to dig deeper into this topic, I highly recommend to check it out. You can start with this article.
Leave a Reply