Study exposes AI climate promises as misleading

Artificial intelligence consumes enormous amounts of energy and computing power. Nevertheless, the big tech companies are increasingly presenting AI as a climate savior. However, a new study now shows that there is little reliable evidence for most of these promises.

Experts assume that the global electricity demand of data centers could more than double by 2030. According to a forecast by the International Energy Agency (IEA), the energy demand of so-called “accelerated” servers will increase by 225 percent between 2025 and 2030.

These high-performance servers also provide the basis for data centers, without which AI tools such as ChatGPT, Gemini and Co. would be unthinkable. For comparison: According to the IEA forecast, the power consumption of conventional servers is only expected to increase by 52 percent over the same period.

Nevertheless, there are numerous claims from the ranks of the tech giants that artificial intelligence could bring a net benefit for the climate. A new study has now examined these claims about AI climate benefits and comes to the conclusion that much of it is unproven or has little scientific support.

Greenwashing? This is what’s behind Big Tech’s AI claims

For his study, climate and energy analyst Ketan Joshi examined a total of 154 statements that promise a net benefit for the climate through AI. This also included statements from tech giants such as Google and Microsoft.

The report, which was commissioned by the German NGO Beyond Fossil Fuels, among others, comes to a sobering conclusion. Accordingly, a large part of the claimed climate benefits of artificial intelligence are bogus.

Overall, 74 percent of industry claims about AI climate benefits are unproven. Only 26 percent of the claims cited published academic papers. For 36 percent there was no source at all.

“It appears that technology companies are using the lack of clarity about what is happening in energy-intensive data centers to gloss over environmentally harmful expansion,” explains study author and climate and energy analyst Ketan Joshi. “This has trickled down to organizations like the International Energy Agency.”

Differences between generative and traditional AI are becoming blurred

One of the central problems is the difference between generative AI and “traditional” AI, as the report states. Generative AI causes significant environmental costs, while traditional AI results in significantly lower energy and environmental impact.

The analysis did not find a single example that generative AI used by ChatGPT, Gemini or Copilot can lead to a significant, verifiable and substantial reduction in emissions. In contrast to this is traditional AI, which is used for machine learning to predict wind patterns, among other things.

“The promises of climate-friendly technologies remain empty words, while AI data centers consume coal and gas every day,” explains Joshi. Exaggerating the climate potential of AI distracts from the real costs of energy and water-intensive data centers.

Also interesting: