The Green Web Foundation is a Berlin-based campaigning organisation that has recently found itself on the front line of debates around hyperscale data centres. It has been a valued source of data and myth-busting around the environmental impact of digital services.
The Green Web Foundation didn’t originally set out to be part of the AI debate – we just got sucked into it.
To be honest, we weren’t really an AI-first organisation. We’ve existed in some form since the early 2010s. We were all ‘webby’ people, working for the likes of Creative Commons, Mozilla, involved with the WordPress Community. And there had been an environmental angle to our work. We’d worked on things like carbon calculation or procurement as a carbon lever, because we’d seen the role government has played as a kick-starter in all kinds of sectors like clean energy.
But in the mid 2020s, we were still focusing primarily on helping people understand how the internet and digital services actually have an environmental footprint, and some of the policy levers available. Mostly around clean energy – there’s a bigger discussion about embodied environmental impacts, but we were mostly talking about energy.
We got involved in AI mainly because people kept asking us about it. We were trying to stay in our lane, but then the conversation expanded and we got sucked into the debate. So in 2024, we published a report about how to think about the environmental impact of AI – the things you have control over and the things you don’t. That was quite well received, and from there, we’ve pushed on.
Our theory of change: narrative shift, applied digital sustainability, and open data ecosystem.
For three years we’ve had this theory of change articulated around three concepts. The first was ‘narrative shift’, which is essentially moving to a richer conversation about the environmental footprint of digital services. We have a magazine called Branch, we have reports like our recent ‘State of the Fossil Free Internet’ report, we work with creatives to put forward alternative visions of how digital services, AI or the internet could work. These are intended to introduce new ideas and expand the discourse.
The second one we call ‘applied digital sustainability’. In other words, taking some of these ideas and diffusing them into the professional class or the mainstream. For example, how the idea of carbon-aware infrastructure – which responds to conditions on the grid – could apply to data centres. Also, working on defining standards: one of the projects I am leading is called Software Carbon Intensity For Web, and there’s also an AI-specific version of that standard.
The third concept is to create an ‘open data ecosystem’. Because if you want to have a data-informed discussion about reducing the harms of digital services, you need meaningful data to be published. And what we’ve seen is that some of the largest companies are among the worst when it comes to sharing meaningful usable data for responsible professionals. And in one case recently, we saw the EU’s Corporate Sustainability Reporting Directive – which was due to mandate companies’ disclosure of how much money they were making from their use of AI for oil and gas extraction – being changed at the last moment under very heavy lobbying.
So we campaign at the policy level and at the technical level to build prototypes and demonstrate the new ideas can work. We’ve ended up talking about AI because it’s the fastest growing source of energy consumption and environmental impact we see at present.
“Transformational AI doesn’t require a massive data centre build-out to actually deliver.”
– Chris Adams, Director at the Green Web Foundation
Framing the debate as being for or against AI is unhelpful – the issue is rarely the technology, it’s the forces behind it.
I’m not sure it’s about being pro-technology or anti-, or even neutral about it. That’s not a rich way to talk about it. There’s a range of views and nuances across our organisation – because we’re technically aware, but we’re also technically wary of what this technology can actually be used for. It’s more about what is this for? Who’s funding it, and what are they expecting to get back? Whose needs are we trying to satisfy first? That colours how we think about technology these days.
We initially came to this conclusion when we were looking at open-source code. There’s a set of values embedded in that, which are very much about avoiding being dependent on single suppliers or concentrations of power in one place. That’s the angle we come from. If you think about the kind of AI where bosses track people at work, for example, this is the kind of thing we’re definitely against. But that’s less about the tech and more about the people behind it. With Big Tech the problem is often with the ‘big’, rather than the ‘tech’.
We’ve found many helpful applications of AI for ourselves.
There are absolutely applications of the technology which are useful. I’ll give you one example. We did something recently using OpenCode [the open source equivalent to Claude Code or Codex]. We figured out how to build energy measuring inside it.
So when you’re using it, you can actually see the energy consumed in an agent coding session, because we figured these are the values we bring to this. We’re thinking: well of course you’re going to use these tools, and you want to know what its footprint is, so you can at least do something responsible about it. So we built prototypes of this tool, and we’re now speaking to professionals to draft standards that could be adopted by their organisations.
The ‘AI arms-race’ narrative doesn’t serve society.
I find the whole ‘arms race’ rhetoric really unhelpful. It tends to put the interests of a small number of very large companies way ahead of the needs of everyone else. And since you have a number of large companies who have considerable access to governments, and are able to spin a good yarn about being a source of growth for the government, many politicians are thinking this is going to save them. The offer is very attractive and offers a quick fix. And that’s deeply problematic. So this arms race narrative is not very helpful. It’s doing more for companies and it’s not useful for society.
There absolutely are risks which are being backgrounded right now. I feel like the discussion is basically: How do we make sure the AI we’re building is safe for US billionaires? Then you ask: How do we make sure it’s safe for US millionaires? And then finally, there’s this after-thought of everyone else. If you think about conversations we hear in the west: the fact we assume they should be allowed to make loads of people unemployed; that we shouldn’t take a stance about the labour that goes into building an LLM like, say, data workers in Kenya; that we don’t include their needs and don’t talk about safety in those terms. This says a lot.
We need to redefine the concept of safety across the whole supply chain.
What I’ve found refreshing recently is some of the work by Rachel Coldicutt from Careful Industries. She’s been trying to reclaim this idea of safety. She talks about safety in terms of the entire supply chain, for example. Whose safety are you thinking about here? Is it being deployed safely? And who are we harming? There’s a whole ethics discussion there.
We need these types of definitions of safety harms. They kick in way before we talk about p(doom) and the existential AI safety risks that take up so much of the oxygen. That’s one of the things I find really useful. It’s interesting because it’s not come from discussions with ‘buzzy’ companies, but from work by companies like Lloyds of London, a 300-year old insurance company which is used to dealing with risk. When organisations like theirs bring new ideas to the table, it is actually really helpful.
The policy debate is really thin right now. There is an opportunity to have a wider discussion that takes account of others’ needs. We should be looking outside of this Anglo-Western approach because there are ideas in other parts of the world which expand what we understand to be possible. There was a good example a couple of months ago, where a Chinese court ruled you’re not allowed to simply use AI as an excuse for laying off workers now. The idea that somewhere out there, a country is saying you can’t just lay people off, AI isn’t for that. That feels like an expansion of the discussion that we need but don’t have right now.
The transformative implementations of AI don’t require huge data centres.
Let’s look at the benefits from a climate point of view for a minute. For predictive AI, you might end up with a much more efficient grid, or better use of clean energy, or predictive maintenance. Unlike the frontier labs’ LLMs, these things don’t require a massive data centre build-out to actually deliver. You do need people to upskill and develop some new skills, but these are not particularly new ideas.
There’s good research from Schneider Electric on this as well as a number of other sources, and we’ve also published our own research. When you look at the claims about the promises of what AI can deliver, many times it’s relatively low energy-consuming traditional AI. But when you look at the conversations around the UK or Germany trying to woo Microsoft, Google or Amazon, you hear about these huge data centres. Because they’re being used to build generative and agentic kinds of tools.
And look, I enjoy the vibe-coding and agent coding sessions as much as the next person. But let’s not kid ourselves that that’s what is transformative. Policymakers need to decouple these two ideas [of growth being linked to compute].
These narratives are about creating dependency.
In the same way that we’ve seen the US using geopolitics to get people to buy lots of fossil fuels as opposed to getting it from Russia, you can see the same thing being done with these data centres. So you’re using, say, Microsoft’s tools or OpenAI or Claude – these proprietary models which you can only get from a handful of companies.
It’s the same kind of thing we might have done 150 years ago. Trying to create dependencies in other parts of the world for things that we want to keep selling. You see lots of parallels between energy dependency and compute dependency. If you can be the sole provider of a commodity, then you have a guaranteed source of revenue or a guaranteed source of political influence. I see the discussions about frontier models as somewhat similar to the discussions about getting people to use American fossil fuel or buying lots of American exports. There’s a geopolitics angle.
AI sovereignty and open source are trying to address the same issue: dependency.
I think the notion of sovereignty is overdue. So, when I spoke about open source, that feels somewhat comparable to sovereignty. The idea you’re not dependent on a single supplier, or a small list of suppliers. That’s what some of the sovereignty discussions have been about, and I think they’ve really come to a head.
We’ve seen this recently with the stories about the Trump administration pressuring Microsoft to cut access to the International Criminal Court president’s email. And even if it’s been denied by Microsoft, it speaks to the mood in Europe. It is not in our national interests to be so dependent on someone else like this. So I think the sovereignty discussion is actually quite helpful.
But it can end up feeling a bit like nationalism and protectionism, and I don’t think that’s very helpful. Because there are some cases where US tech is really good, and totally worth having. But that’s not the same as saying this technology is really cool, therefore we should be entirely dependent on that. The discussion about building your own capability is actually long overdue. And you can see this shift in France and Germany – particularly France – who are saying: we’re going to look at what we’re dependent on and we’re going to build our own tech on top.
There is value in sovereignty, but it also can be seen as a way to support less competitive organisations. You can absolutely accept that some things we do are going to be more expensive right now because it is in our economic interest to build these skills. Take nuclear in the UK for example. It is horrifyingly expensive, but there is an economic interest in maintaining that skill set. Because if you don’t build your own nuclear power stations, then eventually you don’t have those skills anymore and the option is no longer on the table at all in future.
Narratives around existential risk are a distraction from real, immediate problems.
When it comes to the discussions about existential risk, I find this really problematic. Because I feel there’s a lack of curiosity in some of the existing challenges around us already. That you have to go looking for all these far out problems because you don’t seem to be interested in problems that are happening right now, affecting real people in the here and now.
There’s an unwillingness to deal with the fact that real life is messy. It’s like the discussions about solving the AI data centre problem by putting them all in space, so that way we don’t have to think about community acceptance. We don’t have to think about the role people play in deciding whether we get to do it or not. But that is a very blinkered way of talking about it.