ISOIL Industria Spa

How much water does artificial intelligence consume?

In a previous article we addressed the topic of energy consumption of data centers and solutions for improving their energy efficiency. Another important issue that we want to delve into is the enormous environmental impact of these structures in terms of water consumption, which has grown significantly with the pervasive advent of  digital intelligence in our lives. A study by the University of California Riverside’s Research Center has estimated for the first time the water footprint of executed AI queries, revealing alarming data.


Water, a precious resource for AI

Every day, billions of users interact with AI, contributing to the growth of a market that, according to PwC forecasts, will add $15 trillion to the global economy by 2030.

Artificial intelligence represents one of the main and rapidly expanding workloads of data centers, which are busy working 24 hours a day to process calculations useful for answering the many questions posed by users to services such as OpenAI’s ChatGPT or Google’s Gemini.
This results in a significant consumption of resources, including water: most data centers use water cooling systems, which have an efficiency between 50 and 1,000 times higher than air cooling systems.

There are different methods to avoid server overheating: it is possible to spray water on a heat exchange surface and let it evaporate to lower the temperature of the air, which is then ducted (evaporative or adiabatic method); cooling towers can alternatively be used to remove excess heat through evaporation; another solution, more energy intensive, involves the use of chillers and closed circuits of pipes in which the water circulates, absorbing the heat generated by the equipment.


Alarming numbers

Asking between twenty and fifty questions on ChatGPT is equivalent to consuming half a liter of water. According to a study by the University of California Riverside’s Research Center, published in the journal Nature, if each of the 100 million weekly users of the virtual assistant wrote just one command (prompt), up to five million liters would be consumed every week.

Big tech companies have released data on their consumption: in 2022 the tech giants Google, Microsoft and Meta consumed over 2 billion cubic meters of water for server cooling and electricity use, more than double that of Denmark in a year.
The situation, in light of the incessant evolution of AI and in the scenario of growing drought and environmental problems, is destined to worsen: according to forecasts, which also take into account the water used indirectly from other sources connected to the data center such as power plant that powers it, by 2027 the demand for drinking water from the AI will be equal to 6.6 billion cubic meters.


From awareness to innovation: a collective commitment to sustainability is urgently needed

The water footprint of AI models must be addressed as a top collective priority, before it has an irreversible impact on global water supplies.
There is no shortage of virtuous companies that have implemented innovative ecological solutions to make their systems more efficient and reduce their water impact.
The big tech companies themselves are mobilizing in this sense and have communicated their commitment to replenishing a quantity of water greater than that used in their direct activities by 2030.

Having a full awareness of your water footprint is the first step in adopting efficiency solutions and it is essential to adopt measurement tools that meet high safety and reliability standards: smart flowmeters to monitor water consumption in real time, water leak detectors for quickly identify any inefficiencies and minimize waste, probes for analyzing water quality useful for avoiding potentially dangerous contamination for the safety of equipment and operators.
If you need these and other measurement technologies, do not hesitate to contact our specialists.