How Does AI 'Gulp' Down Our Drinking Water

The use of Artificial Intelligence (AI) is growing rapidly. This technology requires a large amount of electricity to work and water is constantly used to keep data centers cool.

According to the United Nations, half of the world’s population is already facing water shortage. Due to climate change and increasing demand for water, this crisis may deepen in the coming times. In such a situation, will such a fast pace of AI increase this water shortage further?

Table of Contents

How much water does AI use?

OpenAI’s Chief Executive Officer (CEO) Sam Altman says that it takes about one-fifteenth of a teaspoon of water to answer a question asked from ChatGPT.

But a study conducted in California and Texas in the US found that the company’s GPT-3 model uses about half a liter of water to answer 10 to 50 questions. That is, about 2 to 10 teaspoons of water is consumed for each answer.

The estimate of water use depends on many things, such as what kind of question it is, how long the answer is, where the answer is being processed and what things have been included in the calculation.

The estimate of 500 milliliters made in this study by American researchers also counts the water that is used to produce electricity, such as coal, gas or to produce steam to run turbines in nuclear power plants.

This figure given by Sam Altman may not have been included. When the BBC asked OpenAI about this, the company did not reveal its method of calculation.

OpenAI says that ChatGPT answers one billion questions every day and ChatGPT is not the only AI bot.

This US study estimates that by 2027, the AI industry will use four to six times more water each year than an entire country like Denmark.

“The more we use AI, the more water we will use,” says Shaolei Ren, a professor at the University of California, Riverside and one of the study’s authors.

How does AI use water?

Everything we do online, whether it’s sending emails, watching videos or creating deepfakes, is processed by thousands of computer servers in huge data centers. Some of these data centers are as big as several football fields.

When these servers are constantly running on electricity, they get very hot. Water, mostly clean fresh water, is vital to keep them cool. Cooling methods vary, but in some systems up to 80% of the water used evaporates.

AI tasks, such as creating images, videos or creating complex content, require much more computing power than normal online tasks (such as shopping or searching online), and therefore consume more electricity.

It is difficult to estimate the exact difference. But the International Energy Agency (IEA) estimates that a query made on ChatGPT uses about 10 times more electricity than a search made on Google.

And when more electricity is used, more heat is also generated, so more water is needed for cooling.

How fast is water use for AI growing?

Big AI tech companies do not disclose how much water their AI-related activities use, but their total water use figures are constantly increasing.

Google, Meta and Microsoft, which are big investors and shareholders in OpenAI, have seen a sharp increase in their water use since 2020, according to their environmental reports. Google’s water use has almost doubled during this period. Amazon Web Services (AWS) has not given any figures.

As the demand for AI grows, the International Energy Agency (IEA) estimates that water use by data centers will almost double by 2030. This includes the water used to generate electricity and make computer chips.

Google says its data centers took 37 billion liters of water from water sources in 2024, of which 29 billion liters were “consumed” i.e. most of it evaporated.

Is this a lot? It depends on who you compare it to. According to the United Nations, this amount of water can meet the drinking and use needs of 1.6 million people for a year, i.e. 50 liters per day. Or according to Google, this much water is equivalent to irrigating 51 golf courses in the southwestern US for a year.

Why are data centers built in drought-prone areas?

Over the past few years, opposition to data centers has been in the news in many drought-affected areas of the world, such as Europe, Latin America and the US state of Arizona.

In Spain, an environmental group called “Your cloud is drying up my river” has formed to protest the expansion of data centers.

In Chile and Uruguay, where severe droughts are occurring, Google has halted or changed plans for its data centers after protests over water use.

NTT Data has more than 150 data centers around the world. The company’s CEO Abhijeet Dubey says that “there is a growing interest in building data centers in hot and dry areas.” He says that the availability of land, power infrastructure, renewable energy sources like solar and wind, and easy regulations make these areas attractive for companies.

Experts also say that where there is more moisture, the problem of corrosion increases and it takes more energy to keep the building cool. For this reason, setting up data centers in dry areas is considered beneficial. Google, Microsoft and Meta admit in their latest environmental reports that they are taking water from dry areas. According to the companies’ latest environmental reports, Google says 14 percent of its water comes from areas with a “high” risk of water scarcity and 14 percent from areas with a “moderate” risk.

Microsoft says 46 percent of its water comes from areas with “water stress.”

Meta says 26 percent of its water comes from areas with “high” or “extreme” water stress. AWS (Amazon Web Services) has not provided any figures.

Are there other cooling options?

Professor Ren says dry or air cooling systems can be used, but they consume more electricity than water.

Microsoft, Meta and Amazon say they are developing “closed loop” systems, in which water or some other liquid continuously circulates in the system and does not have to be blown or replaced frequently.

Abhijeet Dubey believes that such systems will be more needed in dry areas in the future, but he says the industry is still in the “very early” stages of adopting them.

Schemes are also underway or being made in which the heat from data centers is used in nearby homes. This work is being done in countries like Germany, Finland and Denmark.

Experts say that companies generally prefer to use clean and fresh water, the same water that is used for drinking because it reduces the risk of bacteria and rust.

However, some companies are now trying to use sea water or factory wastewater (which is not potable).

Are the benefits worth the environmental impact?

AI is already being used to help reduce pressure on the planet, such as by finding leaks of the dangerous greenhouse gas methane or by redirecting traffic to routes that are less fuel efficient.

“AI could make a big difference for children around the world in education, health and perhaps even climate change,” says Thomas Devin, global director of the innovation office at UNICEF, an umbrella organization.

But he says he would like to see companies compete to see who can do things more “efficiently and transparently,” rather than just “who can build the most powerful and advanced model.”

He also wants companies to open source their models, making them available to anyone so anyone can use and modify them to suit their needs.

Thomas Devin says that if companies open their models to everyone, then the need to train large models will be reduced. Training means putting a lot of data into the system and then preparing answers based on that and this process consumes a lot of electricity and water.

But Lorena Jaume-Palaci, who is an independent researcher and has advised many European governments, European Union and United Nations institutions, and runs a network called Ethical Tech Society, says that “it is impossible to make AI environmentally sustainable at the speed it is expanding.”

She says, “We can make it efficient, but as we make it more efficient, its use will increase.”

She says, “In the long term, we do not have enough raw materials to continue this race to create bigger and faster AI systems.”

What are tech companies saying?

Google, Microsoft, AWS and Meta say they choose cooling technology based on local conditions.

All of these companies have set a goal of becoming “water positive” by 2030, meaning that on average they will try to return more water than they use.

To this end, the companies run water conservation and replenishment projects in the areas where they have data centers, such as restoring forests or wetlands, finding leaks in plumbing or adopting better irrigation methods.

AWS says it has reached 41 percent of its goal, Microsoft says it is “on track,” and data released by Google and Meta show an acceleration in water replenishment. But Thomas Devlin of UNICEF says overall there is “a long way to go” to reach the goal.

OpenAI says it is “working hard” on saving water and energy, and says it is “critical to make the most of our computing power.”

But Prof Ren says companies’ reporting on water use needs to be more transparent and consistent. “If we can’t measure it, we can’t manage it,” he says.

Read Also:

1484700cookie-checkHow Does AI ‘Gulp’ Down Our Drinking Water

Name	Domain	Purpose	Expiry	Type
_ga	altechbloggers.com	Google Universal Analytics long-time unique user tracking identifier.	2 years	HTTP
IDE	doubleclick.net	Google advertising cookie used for user tracking and ad targeting purposes.	2 years	HTTP

Name	Domain	Purpose	Expiry	Type
_ga_GSZLHBGB35	altechbloggers.com	---	2 years	---
jetpack_post_subscribe_modal_dismissed	www.altechbloggers.com	---	1 days	---