From company reports to think tanks, there’s a lot of discussion about the potential of AI for climate protection. And some well-known Big Tech figures effectively believe we’ve already halted climate change, at least for the immediate future. There’s just one catch: there is currently no real evidence of AI’s enormous potential. This is the conclusion of a new report commissioned by a consortium of environmental organisations, including Beyond Fossil Fuels, Climate Action Against Disinformation (CAAD), Friends of the Earth U.S., Green Screen Coalition, Green Web Foundation and Stand.earth, and authored by Ketan Joshi, an independent climate and energy analyst.
The message could not be clearer: 74 percent of claims about the climate benefits of AI are unsubstantiated and primarily serve the profits of the tech and fossil fuel industries. Meanwhile, the blending of traditional AI and generative AI (GenAI) serves to downplay the significant climate damage caused by large language models. The report “The AI Climate Hoax” calls this a new form of greenwashing. It’s therefore high time for a nuanced assessment of the potential of AI.
There is little reliable evidence to support the climate benefits of AI
“Essentially, we have found that the evidence for the claimed climate benefits of AI comes from unreliable sources—such as corporate websites—or is non-existent,” Ketan Joshi reports in an interview with RESET.
The study examined 154 statements from companies such as Google and Microsoft, as well as from institutions such as the International Energy Agency (IEA), which claim that AI ultimately has a positive impact on the climate. The underlying assumption is consistent across these studies. They assume that the positive contribution of AI offsets the increased demand for fossil fuels required to power data centres. However, according to Joshi’s findings, only 26 percent of the claims were backed by scientific research. 36 percent provided no evidence at all. “In our view, the models showing that global emissions will fall by gigatonnes thanks to the use of AI are overrated,” said Joshi.
The dangerous confusion between traditional AI and GenAI
The report also highlights a serious confusion between different types of AI. This is because most of the ‘benefits’ for the climate that have actually been demonstrated are linked to older, simpler machine learning models. The report refers to these as ‘traditional AI’. The greatest harm, however, is generative AI (GenAI), according to the report.
We at RESET also repeatedly emphasise this important distinction. Traditional AI comprises small models, each designed to solve specific tasks. Their training is mostly based on manageable datasets, which keeps energy requirements low. Such traditional AI applications can, for example, help detect forest fires at an early stage or assist with wildlife monitoring. You can find many more examples of AI applications here: Artificial Intelligence in Environmental and Climate Protection.
GenAI, on the other hand, includes large language models such as ChatGPT, Gemini or Copilot. They are designed to handle every conceivable task, which is why their training requires as much data as possible. Consequently, the technology has enormous energy requirements, as evidenced by the absurd boom in data centres. Big Tech companies are investing vast sums in building new data centres. In many cases, the local grid cannot supply the necessary energy. When the data centre needs more energy than the grid can supply, gas turbines fill the gap. This has huge consequences for both local residents and the climate.
Those who downplay the negative impacts of GenAI stand to benefit most from the conflation of different AI models. “The different types of AI are being conflated, so that the small, potentially beneficial applications are used to justify the large, harmful ones,” says Joshi.
During his research, Ketan Joshi could not find a single example where generative AI had led to actual emissions reductions or other benefits for the climate and the environment. However, the report’s findings are similarly sobering for traditional AI. “We were unable to find any reliable figures on the energy consumption of traditional AI. But we assume that it is not massive.” So there are traditional AI technologies with positive environmental benefits. But here too, we must assume that reports overestimate the actual emission reductions.
Climate benefit claims obscure the irreversible damage to communities and society
The report makes it clear: the evidence that AI will lead to large-scale climate benefits is weak. In contrast, we have strong evidence that GenAI causes immediate and significant damage to the climate and the environment.
“The promises of a planet-saving technology remain hollow as long as AI data centres keep coal and gas alive every day,” notes Ketan Joshi. To tout AI in general as a technology that will lead us out of the climate crisis is therefore a misleading bait-and-switch tactic. The authors describe this as a new form of greenwashing.
We are still reliant on estimates
One reason this greenwashing is possible is that we still have very little concrete data on the resource consumption of data centres. We know that in many hotspots, network connections are no longer available. Satellite and thermal images show the gas turbines that are increasingly powering data centres. And there are increasing reports from residents who live near datacentres. They describe how their local microclimate is changing, and water resources are becoming scarcer. Nevertheless, there are no objective and independent measurements of the water and energy consumption of individual data centres. “All data must be pieced together from company figures or, in some regions, from government monitoring of energy and emissions,” confirms Joshi.
Demand solid evidence to back up strong claims
However, we should not let these unfounded claims set the tone, says Joshi, especially as we are also at a critical juncture “where the future is being decided”. We should therefore demand solid evidence to back up strong claims.
With this new report, the consortium aims above all to counter the continued expansion of fossil fuels and greenwashing narratives. Central to this is that governments demand fundamental transparency from the AI industry. “Only then can communities and scientists know how much energy is being consumed by this technology,” says Michael Khoo, co-chair for policy at Climate Action Against Disinformation and program director at Friends of the Earth U.S.
Furthermore, we must clearly distinguish the various types of AI from one another in order to realistically assess their environmental impact. This is a task that only companies, organisations and governments must tackle, but one that we as journalists should also take seriously.
But what does this mean for us as end-users—should we avoid GenAI? Ketan Joshi’s answer is clear: “Yes, avoid GenAI wherever you can. It’s good for your brain and for the planet. Of course, you won’t be able to avoid it entirely. But the most important thing is: support artists, journalists, writers, activists and creatives who do real human work. They need your support far more than we need someone who boycotts chatbots.”
Concerns about CO2 emissions from GenAI? Here’s how to use language models more efficiently
AI chatbots like ChatGPT, Perplexity or Gemini enhance your daily life, but are you concerned about sustainability? There are ways to reduce your digital carbon footprint when using GenAI. We’ve put together some guidelines to help you use language models more sustainably.
