TEMPO.CO, Jakarta - Artificial intelligence (AI) is growing increasingly ubiquitous in everyday life: you can use it as a search engine, generate images, or even vent your feelings. This technology works by relying on large numbers of computers stored in data centers.
Behind this convenience, there's one less-known environmental impact of AI: water consumption.
Is AI depleting the amount of water on Earth?
Simply put, AI doesn't simply cause water on Earth to disappear. Water continues to follow the hydrological cycle: evaporation, cloud formation, precipitation, flowing into rivers and oceans, and back to evaporation.
The problem isn't the loss of water from our planet as a whole, but rather depleting fresh water available for human use at certain locations and times.
AI requires computers with high processing power. These computers generate significant amounts of heat when used, especially when training or running large AI models.
To prevent devices from overheating, data centers require cooling systems. Some facilities use water to help dissipate this heat. In other words, the greater the AI computing needs, the greater the energy and cooling requirements, potentially increasing water usage.
Certain cooling systems use water that can then evaporate to carry heat away from the facility, which won't be readily available to nearby communities.
Research on AI's water footprint shows that water use can come from two main sources: direct use for data center cooling and indirect use from power plants that supply energy to the data center.
Meaning, AI's water footprint comes not only from inside the data center, but away from the building, according to Antara.
Electricity is also linked to water use
AI requires great amounts of electricity, since data centers running the services must operate around the clock and require high computing power. The problem emerges when water is used indirectly in the electricity supply chain.
Data from the International Energy Agency (IEA), cited in a study on data center water footprints, shows that indirect water consumption from power generation can be a larger share than direct use for cooling. By 2023, total global data center water consumption is estimated to be around 560 billion liters, with approximately 373 billion liters coming from indirect use for electricity supply and approximately 140 billion liters used directly in the facilities.
This figure includes data centers as a whole, not just those used for AI. However, the rapid development of AI is linked to the growth in computing and data center demand.
So, does each AI query consume a lot of water?
The answer isn't that simple.
Several studies have attempted to calculate the water footprint of AI activities, but results can vary, depending on the AI model, the length of the query, the location of the data center, the weather, the cooling system, and the power source used.
A study led by researchers from the University of California, Riverside, for example, estimated that training a large AI model like GPT-3 in a typical data center could directly evaporate approximately 700,000 liters of fresh water. The study also estimated that global AI demand could reach 4.2 to 6.6 billion cubic meters of water by 2027 if growth continues.
However, these figures are estimates based on specific methods and assumptions. Therefore, the aforementioned water figures for a single AI query or a conversation with a chatbot cannot be considered definitive figures for all AI services.
Different region, different impact
The biggest issue isn't how much water AI uses globally, but where.
Using large amounts of water in areas with abundant water resources will certainly pose different consequences than using the same amount in areas experiencing drought or water stress.
Recent research on AI's water footprint also shows that many data centers are located, or could potentially be, built in areas facing water resource pressure. Therefore, the location of data center construction is a critical factor in assessing the environmental impact of AI.
Thus, a single data center may have a relatively small impact on global water availability, but could pose a serious problem for local communities.
Is it possible for AI to become more water efficient?
The good news is that AI usage doesn't have to be synonymous with excessive water use.
Data center managers can use more efficient cooling technologies, including air-based cooling systems, liquid cooling, recycled water use, and closed cooling systems that allow water reuse.
Some modern data center designs are even developed with systems that reduce or eliminate the need for water for direct cooling. But these choices lead to other consequences, such as increased electricity demand for certain cooling systems.
Therefore, simply reducing water usage is insufficient to achieve a more environmentally friendly AI system. Power sources, device efficiency, the data center location, and cooling technology all need to be taken into account.
Read: Are Data Centers Really Bad for the Planet?
Click here to get the latest news updates from Tempo on Google News
