Skip links

The Environmental Costs of Generative AI

Along with the hype and enthusiasm regarding generative AI, few people stop to wonder about the environmental costs of such a technology. However, awareness and responsible decision making is needed to avoid that AI contributes negatively to the climate crisis.

When considering sustainability problems, few people consider the environmental costs of running generative AIs such as ChatGPT. Maybe it is because the costs of streaming, cloud storage, and ChatGPT seem to be abstract to us, compared to when we go on an airplane or let the water run for too long. Nevertheless, running generative AIs consumes a lot of electricity and fresh water – a lot more than letting the tap run for too long.

In January 2024, at the World Economic Forum’s annual meeting in Davos, Sam Altman the chief executive OpenAI said that the AI industry is going to consume vastly more power than people have anticipated. Already now, it is estimated that one search facilitated by generative AI consumes four to five times the amount of energy, compared to a normal web search. With the current development of AI an energy breakthrough is needed to keep up with the increasing demands of energy to avoid catastrophic effects on the environment.

A suggestion could be to slow or haul the explosive development of AI, until we have a better understanding of the ethical consequences and how to sustainably run these energy demanding systems. But Altman suggested developing more climate-friendly energy sources, such as nuclear fusion or cheap solar power.

Altman is also personally invested in fusion energy, as he in 2021 started a fusion company called Helion Energy. The company estimates that in 2029 it will be able to produce energy equivalent to the consumption of 40,000 average US households. However, one assessment of the energy consumption of ChatGPT suggests that it is currently consuming the energy of 33,000 homes and this demand will only increase in the years to come, potentially bringing the energy cost to those of small nations.

As mentioned above, generative AI also demands enormous amounts of fresh water to generate electricity and cool processors. In 2022, a lawsuit filed by local residents West Des Moines, Iowa, revealed that the OpenAI’s data center that is running ChatGPT-4 was consuming around 6% of the district’s water. Similarly, when Google and Microsoft developed their large language models, both companies had significant spikes in their water consumption, which increased by 20% and 34%, respectively, within one year.

These choking numbers have finally made regulators notice this growing problem. In the US, Senator Ed Markey of Massachusetts introduced the Artificial Intelligence Environmental Impacts Act of 2024. The legislation states that the positive and negative environmental impacts of AI should be investigated and measured. Similarly, the European Parliament has published the Artificial Intelligence Act: deal on comprehensive rules for trustworthy AI that demands that companies report on their energy efficiency.

However, more is needed than investigations into the energy consumption of large AI systems. Kate Crawford, a professor at the University of Southern California Annenberg, presents following suggestions:

  • Companies should follow sustainable practices and report on their energy and water use, as well as prioritise developing energy-efficient software and hardware.
  • Legislators could incentivise the adoption of renewable energy, make benchmarks for water and energy use, and demand environmental reporting.
  • Independent entities could conduct environmental audits to support transparency and uphold sustainability standards.
  • AI researchers could optimise AI architectures to be more sustainable as well as collaborate with environmental scientists to ensure that technical designs take ecological sustainability into account.

The first step to ensure a future where AI is sustainably powered, is to create awareness of the massive energy consumption of huge AI systems. Hereafter, legislators, researchers, and developers must collaborate and act responsibly, if AI is not to contribute negatively to the climate crisis.



NATURE: (last assesssed on Mar. 1, 2024)

REUTERS: (last assessed on Mar. 1, 2024)

Also read where has been focusing on the topic as a partner for year