Artificial intelligence is everywhere. We use it to write, translate, work, study, create images, optimize industrial processes and even tackle the climate crisis. But behind this apparently “immaterial” technology lies an enormous, often ignored, environmental cost. A new scientific study, published in a peer-reviewed journal, turns the spotlight on a sobering fact: in 2025 alone, artificial intelligence systems may have consumed an amount of water comparable to the entire global annual consumption of bottled water.
How much water did artificial intelligence consume in 2025?
According to the research, AI would require between 312.5 and 764.6 billion liters of water in 2025. Numbers that have nothing abstract about them and that suddenly become concrete if we think that we are talking about a volume similar to that used every year, throughout the world, by the packaged water industry.
The main reason for this consumption is not the training of the models itself, but the cooling of data centers, the huge infrastructures that host the servers that run systems such as advanced language models. The calculations required to make these technologies work generate an impressive amount of heat and, to avoid overheating, evaporative cooling systems, which use large volumes of water, are often used.
Added to this is a less obvious but even more impactful aspect: the water used indirectly to produce the electricity that powers the data centers. The authors of the study explain that water consumption related to electricity production can be three to four times higher than water used directly to cool servers.
The carbon footprint of AI is comparable to that of a large metropolis
Water consumption isn’t the only red flag. The same research estimates that CO₂ emissions generated by artificial intelligence in 2025 will be between 32.6 and 79.7 million tonnes. To give an order of magnitude, these are values comparable to the annual emissions of a city like New York, considering the energy and transport sectors.
A fact that highlights an increasingly evident paradox: a technology also designed to help us solve environmental problems is significantly contributing to making them worse. The adoption of AI is growing at exponential rates, but the infrastructure that supports it does not always follow the same pace in terms of sustainability.
The hidden side of data centers between corporate silence and the request for transparency
One of the most critical aspects highlighted by experts concerns the lack of transparency. Many large tech companies do not make public detailed data on how much water and energy is actually used for AI services. This makes it difficult to have precise estimates and, above all, evaluate the real impact of individual platforms.
Scholars point out that, without clear data, it is also complex for governments and citizens to understand how much AI is impacting natural resources, particularly in areas already affected by water stress. This is not a minor detail: many data centers are located in regions where water is already a critical resource, increasing competition between industrial uses and the needs of local communities.
The good news is that there is a way out. Experts agree on some key points: data centers powered by renewable sources, more efficient and less water-intensive cooling systems, recovery of the heat produced by servers and, above all, more stringent regulation that imposes clear environmental standards.
Without these interventions, the risk is that the expansion of artificial intelligence will continue to take its toll on water, climate and ecosystems, transforming a promise of progress into a new environmental problem.
AI can be part of the solution, but only if we stop considering it immaterial and also start looking at its physical foundations, made of servers, energy and millions of liters of water.
You might also be interested in:
