Some AI advocates claim that bots hold the secret to mitigating climate change. But research shows that the reality is far different, as new datacenters cause power utilities to burn even more fossil fuels to meet their insatiable demand for energy.
It has been a common refrain of late that, as the tech industry’s single-minded focus hones in on AI, abandoning climate pledges in the process and embracing quick-to-build fossil fuel power plants to drive its datacenters, the AI being spun up in new mega facilities would simply solve the problem for them.
Former Google CEO Eric Schmidt has pushed that line of reasoning, as has Microsoft cofounder Bill Gates, both arguing that AI’s long-term climate benefits could outweigh, or help offset, the emissions associated with the growing power demands of datacenters. Others, including the International Energy Agency, claim to have support for the same.
With funding from climate action groups, energy analyst Ketan Joshi, however, is suggesting that AI firms and research from groups that support their plans, like the IEA, aren’t just wrong, but may be intentionally greenwashing the problem.
Joshi’s report was published this week and funded by groups including Beyond Fossil Fuels, Climate Action Against Disinformation, and Friends of the Earth U.S. He reached two central conclusions after looking at 154 claims of AI climate benefits arising from eight sources (including the IEA, climate researchers, Microsoft, Google, and others).
Which AI are we talking about?
First off, there’s a serious problem with the conflation of traditional AI, like predictive models and computer vision, with generative AI tools that chat, create images, and make music.
“This analysis found that the overwhelming majority of AI climate benefit claims relate to ‘traditional’ forms of AI rather than generative AI,” Joshi wrote in the report.
Across the eight sources and 154 claims examined as part of the research, only four of them in any way related to generative AI systems and their potential to help the environment. In the other 150 cases, the language used in the data pointed to the deployment of traditional AI models as a potential source of climate solutions.
“At no point did this search or analysis uncover examples where consumer generative systems such as ChatGPT, Gemini or Copilot led to a verifiable and substantial level of emissions reductions,” Joshi wrote. “However, much of the projected energy consumption of AI will stem specifically from ‘generative AI,’ rather than more traditional forms of machine learning.”
By generalizing their use of AI, companies claiming their systems would benefit the world they’re polluting are pulling the wool over people’s eyes, the report concludes, citing the coupling of traditional and generative AI in this context to be “illogical” and “false.”
“The benefits and harms exist in discrete technological domains; rendering the core ‘net climate benefit’ defence of AI growth utterly implausible,” the report concluded.
The evidence that isn’t
Along with conflating the potential climate benefits of using traditional AI with the far more energy-intensive and frequent deployments of generative AI datacenters, the report concludes that most of the evidence that even traditional AI would help with the climate crisis lacks support.
Of the 154 climate benefit claims that were examined in the study, only 26 percent cited published academic papers, and 36 percent lacked any citations for their claims at all. The remaining papers that included citations largely relied on corporate publications (29 percent), while the rest cited media reports, NGOs, institutions, and unpublished research papers.
Unsurprisingly, those corporate sources mostly “did not include any primary assessable evidence or peer-reviewed and published academic work to support their claims,” Joshi found.
All the while, the sources reviewed for the report (all of which are linked in the report for those curious) paint a picture of a future in which AI will solve the climate problems it is exacerbating.
“Even the narrower, older forms of AI may be seeing an exaggeration and overstatement of their climate benefits, considering the lack of strong, peer-reviewed and verifiable evidence of their deployment in reducing greenhouse gas emissions in the real world,” Joshi noted. “The narrative of a gigatonne-scale shift in global emissions that ‘offsets’ the harm of generative AI is not supported by the results of this analysis.”
Paint it green
If it’s not claims that AI datacenters are going to be powered by next-gen nuclear followed immediately by admissions that all that current buildout is simply being powered by fossil fuels, it’s just another form of greenwashing from the AI industry, it seems.
Joshi told The Register in an email that he was motivated to undertake this research after hearing Google persistently claim that its AI deployments would reduce emissions by 5 to 10 percent based on what he characterized as “mind-bogglingly weak evidence” that he considers “one of the worst greenwashing claims I’ve seen in a while.”
As to what could be done to combat such widespread misinformation about the climate effects/benefits of AI, Joshi suggested those fighting to force honesty on the AI industry need to play the waiting game, to a degree.
“The entire software class is leveraged on hype rather than effectiveness or usefulness and is arguably doomed either way,” Joshi told us, echoing claims that AI is a bubble likely to burst at any time.
“There is widespread public skepticism for what all of this is actually for, skepticism of the tech industry and bipartisan hostility towards destructive data centre developments,” Joshi added. “I would not underestimate how much public opinion is stacked on the side of us climate advocates and analysts.”
Joshi said that the job of climate advocates in the current moment is to ensure that there’s as little fossil fuel infrastructure built, and locked in to produce years of emissions, as possible.
He’s not advocating action out of far left field, either: A growing number of US states are trying to pass moratoriums on new datacenter projects as activist demands pile up at the state and federal level. Elon Musk’s xAI, which has earned public ire and a lawsuit over gas turbines being used to power a massive AI datacenter in Memphis, Tennessee, is now being looked at by state regulators unhappy with the situation.
In short, even with the AI industry claiming it’s helping to solve a climate crisis of its own exacerbation, public opinion may be starting to turn.
We reached out to the IEA, Microsoft, and Google for comment on this story given how heavily they were featured in the report. The IEA didn’t respond, Microsoft had no comment, and Google stood by its claims.
“We stand by our methodology, which is grounded in the best available science. And we are transparent in sharing the principles and methodology that guide it,” The Chocolate Factory told us. ®