Is AI Depleting Our Water Supply? Facts, Research, and Expert Opinion

Is AI Depleting Our Water Supply? Facts, Research, and Expert Opinion

In recent years, discussions about AI technology have focused not only on intelligence, productivity, or automation, but also on its impact on the environment. One persistent issue is the question: is AI really depleting our water supply?

The answer: AI does need water, but the context is much more complex than simply “AI drinks our water.”

This article explores scientific facts, expert opinions, and the latest research to help you understand how much AI water consumption actually is, where it comes from, and how to solve it.

Why Does AI Need Water?

Water is needed primarily for cooling data centers, which are facilities where thousands of servers work to train and run AI models such as GPT, Llama, and others.

AI servers process data on a massive scale. The hotter the servers run, the greater the cooling requirements.
This is where the water comes in.

Generally water is used in:

• Cooling Tower / Evaporative Cooling: water is evaporated to release heat
• Chillers: cooling systems that still use water
• Liquid Cooling: some modern cooling systems use special water-based fluids.

According to the World Economic Forum (WEF), some data centers can use tens of millions of liters of water per year, depending on their size and cooling technology.

How Much Water Does AI Use? Expert Opinion & Scientific Studies

1. “Making AI Less Thirsty” Study (Li et al., 2023 – 2024)

The study found that training large AI models can “consume” hundreds of thousands of liters of water through the cooling process & electricity supply chain.

Giant model = hotter server load ⇒ more cooling ⇒ greater water evaporation.

2. Latest scientific review (2025): Water Use of Data Center Workloads

The study confirms that data center water consumption can vary by up to 10,000× between one AI workload and another, meaning that heavy AI such as large language models require more water than typical computing tasks.

Main factors:

• type of cooling
• computing intensity
• location (hot climates require more water)
• energy sources

3. AI Water Efficiency Dataset (2024)

Researchers created a dataset to measure water consumption per AI task.
Examples of findings:

Writing a 10-page report with a large model like GPT-4 or Llama-3-70B can be equivalent to 0.5–1 liter of water, depending on the data center location.

Smaller models = lower consumption

Large model = much higher consumption

This shows: the everyday use of AI still has an impact, even if it is relatively small compared to other activities such as agriculture, industry, etc.

4. WeForum 2024 – “Circular Water for Data Centers” Report

WEF warns:

• AI power & cooling demand is soaring rapidly
• Less than 1/3 of data centers track water consumption (WUE – Water Usage Effectiveness)
• Without innovation, global data center water consumption could increase dramatically, especially in drought-prone areas.

5. Water-cooled cooling system research (2024, ScienceDirect)

Field research shows that cooling consumes a significant portion of energy and water. System efficiency significantly determines a data center's water consumption.

What Impact Will This Have on Our Water Supply?

Experts agree: AI won't immediately deplete the public's drinking water supply, but the greatest risk lies in certain locations.

The greatest impact occurs if:

• Data center built in drought-prone area
• Using fresh water that is suitable for drinking (potable water)
• Hyperscale data center scale
• Cooling still relies on evaporative cooling

Environmental experts assert that data center water consumption can:

adding pressure on local water supplies

creates conflict of needs (industry vs society)

worsening “water stress” in areas already short of water

Several cities in the US, Europe, and South Africa have already complained about increased water use by digital facilities.

So, Is AI a Problem or a Solution? Expert Opinions

The majority of researchers agree that:

AI is not the primary cause of the water crisis, but it can exacerbate conditions in already water-stressed areas if:

• Data centers extract large amounts of water
• There is no regulation
• Inefficient cooling technology

But on the positive side, AI could be part of the solution:

• Optimizing water management
• Predicting community water leaks & consumption
• Improving industrial energy efficiency
• Assisting irrigation & agricultural management

Experts' Recommended Solutions

According to environmental research & data center infrastructure experts:

✔ 1. Switch to a waterless cooling system

For example:

• Direct-air cooling
• Liquid-to-chip cooling with a closed system
• Immersion cooling

✔ 2. Using recycled water (greywater)

Not public drinking water.

✔ 3. Transparency of WUE metrics

Companies are required to report their water consumption.

✔ 4. Data center development regulations

Determine which areas are safe in terms of water supply.

✔ 5. Combination of clean energy + server efficiency

Clean energy = less water from the electricity supply chain.

Conclusion

Is AI wasting our water?
Not globally, but it can drain local water if not managed properly.

The truth is:

• AI and data centers do require large amounts of water.
• The amount depends on the cooling technology & location.
• The impact is very real in dry areas or areas with limited water supplies.
• But AI can also help reduce water use in other sectors.
• The solution is available, just need to be implemented