Artificial intelligence in 2025 consumed record amounts of water and electricity

Artificial intelligence automates work, makes life easier and consumes valuable resources. Their scale is difficult to estimate. A Dutch researcher has calculated, however, that annually AI may require as much electricity as the entire city of New York, and the amount of water consumed corresponds to the global annual consumption of bottled water. This is more than previously assumed.

How does artificial intelligence affect water and electricity consumption?

So far, the primary source of information on water and electricity consumption by artificial intelligence has been data published by the International Energy Agency (IEA). According to its estimates, AI accounted for 15 to 20 percent of energy use in data centers in 2024. The IEA also estimated that CO2 emissions from data centers in 2024 amounted to 182 million tons, and water consumption in 2023 reached 560 billion liters, without clearly indicating what share was attributable to AI.

Alex de Vries-Gao, a doctoral candidate at the University of Amsterdam, decided to examine the issue more closely. The conclusions of his calculations are as follows: CO2 emissions resulting from energy use by artificial intelligence may amount to between 32 and 79.7 million tons annually. Estimates of water consumption are also surprisingly high, ranging from 312.5 to 764.6 billion liters per year, more than estimated by the IEA.

Difficulties in precise estimation of consumption

Calculating the actual resource consumption of artificial intelligence is difficult. Evidence of this is the imprecise data published by Google in August this year. According to calculations by the technology giant, Gemini, its flagship AI product, each user-entered prompt consumes five drops of water and as much electricity as needed to watch television for nine seconds.

Scientists questioned this information. Alex de Vries-Gao commented on Google’s claims as follows: This is only the tip of the iceberg. In its calculations, the company considered only water consumption for cooling data centers. It did not include indirect consumption, for example for electricity generation. In addition, consumption during AI model training was omitted, a long, repetitive process that requires large amounts of resources.

Call for greater corporate transparency

During his research, Alex de Vries-Gao faced a lack of access to complete data. In his calculations, he considered direct water consumption for cooling data centers as well as indirect consumption for electricity generation. He also used information from corporate reports showing CO2 emissions. In each of these cases, however, there was no clear indication of what portion of consumption was attributable to artificial intelligence. He therefore made estimates based on publicly available information, for example on company revenues.

The researcher notes that CO2 emissions and resource use depend to a large extent on the location of data centers and on where the energy that powers them comes from, whether renewable energy sources or conventional ones.

The author of the study emphasizes that even big tech companies, the largest purchasers of water and electricity for artificial intelligence, do not publish detailed data. In the report, he calls on companies to become more involved in collecting and analyzing data, as well as to be transparent and publish it. Only in this way will it be possible to realistically assess the impact of AI on the environment and resource consumption.