The rapid expansion of artificial intelligence has far-reaching consequences for the environment. A new analysis shows how much the increasing energy requirements of AI systems not only increase CO2 emissions, but also significantly increase the consumption of water and land.
With the introduction of ChatGPT at the end of 2022, artificial intelligence could be directly experienced by millions of people for the first time. What was initially perceived as a technological innovation has developed rapidly within just a few years. Today, AI has arrived in many areas of everyday life.
Daily use ranges from searching for information to writing texts to creating photos or videos. In 2025 alone, ChatGPT’s number of monthly active users worldwide will have more than doubled. While there were 358 million in January, the number rose to 810 million by December.
However, parallel to the growing use of powerful AI applications, the need for computing power is also increasing and with it the question of the ecological consequences of the AI boom. Researchers at the United Nations University (UNU) looked into these and evaluated in a report the significant consequences for climate, water resources and land use worldwide due to the rapid expansion of AI systems.
Electricity, water, space: What AI data centers will use by 2030
AI systems require enormous amounts of computing power both during training and in daily operations. In order for AI to recognize patterns, understand connections and respond to queries, huge amounts of data must be processed.
These processes take place in huge data centers that usually work around the clock. This not only increases the electricity consumption required for this. Because of the waste heat generated, data centers also need complex cooling.
Due to the increase in the number of users of AI systems, the need for energy and infrastructure is also increasing. According to UN University research, power consumption by AI data centers worldwide will increase to 945 terawatt hours by 2030. This corresponds to almost twice Germany’s annual electricity consumption. For comparison, data centers worldwide will have consumed an estimated 448 terawatt hours of electricity in 2025.
Water consumption for cooling data centers increases enormously to 9.3 trillion liters. The area required for AI infrastructure could grow to around 14,500 square kilometers by 2030 – an area almost the size of Schleswig-Holstein.
Why CO2 alone does not reflect the AI footprint
The researchers are convinced that not only CO2 emissions can be used when considering the ecological footprint of AI systems. “What surprised us most was how often the most environmentally friendly options from a carbon perspective end up being worse for water or land,” explains lead author Miriam Aczel from UNU. “If we continue to judge the sustainability of AI based solely on carbon emissions, we might think that renewable energy makes AI infrastructure cleaner, but in doing so we are solving one problem while creating others, often in places that did not create those problems.”
An additional problem is that models are becoming more cost-effective due to their increasing efficiency and thus usage is increasing. “Many people believe that the environmental footprint of AI will decrease as technology improves and processes become more efficient. But that is only part of the overall problem,” said UNU professor and co-author Kaveh Madani. “More efficient and cost-effective AI and energy means more AI consumption, making the overall footprint far larger than what we save through efficiency gains.”
Six principles for a more sustainable AI future
The researchers are therefore calling for a responsible AI ecosystem. This must be based on the six principles of transparency, efficiency through design, equity and environmental justice, responsibility across the entire life cycle, global cooperation and sustainable use.
To do this, governments, among other things, would have to integrate the AI infrastructure into energy planning, water management and land use approval. But users could also do their part to limit the ecological footprint of AI.
Because many questions that could be answered with a simple search query via classic search engines are now asked directly to AI systems. Since every query has to be processed by powerful data centers, the energy consumption is usually significantly higher than with a conventional web search. A conscious and targeted use of AI could therefore help to avoid unnecessary consumption of resources without having to forego the advantages of the technology.
Also interesting:
