Artificial intelligence (AI) is an indispensable tool for new generations, used to create reports, find information and even generate Studio Ghibli -style images. However, behind this technology a serious environmental risk is hidden that threatens human survival, directly affecting one of the most vital resources for its existence: water.
Social networks have been filled with images generated by artificial intelligence, whose ingenuity to transform a photograph into a cartoon format has turned it into a trend. However, this popularity could be depleting drinking water reserves, since data centers require constant cooling to keep servers at an adequate temperature and avoid overheating.
A study by the University of Colorado Riverside and the University of Texas Arlington details that the elaboration of a single image generated by AI represents the consumption of 2 to 5 liters of water, depending on the energy used by the servers, which, when overloading, require more power and, therefore, more cooling.
The study “making the AI less thirsty” emphasizes that, to develop a Ghibli Studio -style image based on a photograph, the server can consume up to 3.45 liters of water on average, while the images generated from scratch consume an even greater percentage.
Although it could be thought that only the generation of images requires this type of resources, the reality is that the creation of text also represents a considerable expense of energy and water. The study highlights that the generation of a 100 words text through chatgpt or other artificial intelligence tool consumes at least 519 milliliters of drinking water, that is, half a liter.
A latent environmental crisis
Environmental concern increases by understanding that the use of AI exponentially increases drinking water consumption. Although the planet is composed of 70 % water, only 3 % is suitable for human consumption. More than two thirds of this water are stored in glaciers, which are another essential resource to regulate the temperature of the earth.
The United Nations Environmental Report details that almost two thirds of the world’s population suffer a serious water shortage at least one month a year, which could make drastically worse in the coming years if water consumption continues to increase by AI developed companies and technological advances.
The projection for 2030 warns that about half of the world’s population will face a serious water stress, further aggravating the gap in access to this vital resource, which is already scarce in some regions of the world at this time.
Why is salt water used to cool servers?
Although the possibility of using salt water in these systems has been raised, this option is not viable due to the serious damage it could cause. Salta water causes corrosion in materials such as steel, aluminum and copper, which are commonly used in cooling systems. This corrosion deteriorates internal pipes, pumps and components, which could seriously compromise the infrastructure of data centers.
Experts detail that salt accelerates the oxidation process, quickly damaging the equipment. In addition, this type of liquid reduces its ability to absorb and transfer heat efficiently, which prevents its use in server cooling systems. For these reasons, only drinking water can be used in these processes, thus increasing the demand for this vital resource.
The Washington Post He explains that, every time IA is used, either chatgpt or other artificial intelligence tool, its servers execute thousands of calculations to collect information and develop, in a small time, a coherent text. This process implies the use of water to cool the equipment and keep them in operation, since the servers generate heat when overloading.
This process makes the heat generated on the servers addressed to the cooling towers, where it is expelled through steam.
Studies developed to analyze this problem estimate that, worldwide, AI could generate an extraction of between 4,200 and 6.6 billion cubic meters of water by 2027, an amount that is consumed annually by a country like the United Kingdom.
Another problem that AI has generated is the consumption of electricity. According to data from the US Department of Energy, these centers require large amounts of energy to develop their answers and store their data, so it is estimated that, by 2028, electricity increases 300 %.