Back to Blog
•6 min read

Is AI Really Draining the Planet’s Water? Myth, Reality, and the Part Nobody Mentions

Amna Razzaq

November 1, 2026

5
Is AI Really Draining the Planet’s Water? Myth, Reality, and the Part Nobody Mentions

You have probably seen the headlines. AI is “drinking” our rivers. Every chatbot question wastes a bottle of water. Data centers are turning towns into deserts.

Some of that is exaggerated. Some of it is real. The honest answer is somewhere in the middle, and it depends on where you look.

Where the famous claim came from

The most repeated line is that every ChatGPT conversation uses a bottle of water. That figure traces back to a single academic paper from UC Riverside researchers. It was a calculated estimate built from public data, not a meter reading from a real data center.

Real measurements from companies point to a much smaller number. In August 2025, Google reported that a median text prompt to Gemini used about 0.26 ml of on-site water, roughly five drops. Other estimates are higher. One 2026 explainer puts a single LLM response at 10 to 50 milliliters. The gap exists because studies measure different things. There is no single water footprint for AI, since studies define water use and system boundaries differently.

So the viral “bottle per chat” claim is shaky. But that does not mean AI’s water use is zero.

How data centers actually use water

AI chips run hot. Many facilities cool them by circulating water that absorbs the heat and then partly evaporates in cooling towers, which needs constant fresh water. But this is not the only way to cool a building. Evaporative cooling is just one of several approaches, and liquid cooling should not be confused with water consumption.

The newer designs are changing the picture. The newest data centers use closed loop designs, immersion cooling and reclaimed water agreements. Microsoft has unveiled designs optimized for AI hardware that use zero water for cooling, even in desert climates.

The hidden water: electricity

Here is the part that gets less attention. A data center’s biggest water use may not happen inside the building at all. A Lawrence Berkeley National Laboratory report estimated that US data centers directly consumed about 17 billion gallons of water for cooling in 2023, and roughly 211 billion gallons indirectly through the water used to generate their electricity.

That means the electricity question and the water question are really one question. Power plants use large amounts of water for their own cooling, so a data center on a fossil heavy grid has a bigger hidden footprint than one running on wind or solar.

The big picture: small nationally, serious locally

Compared with the whole country, data centers are not a major water user. Their direct annual use works out to about 0.3% of the US public water supply. One policy group that supports data center growth argues the concern is overblown, saying total US data center use is under 0.5% of American freshwater use. That group takes a strong pro industry stance, so treat its framing with care. The 0.3% figure above comes from a source that sells cooling technology, so it has its own angle too. The numbers agree, though: nationally, this is a small slice.

But national averages hide local problems. A single facility can matter a great deal to a single town.

  • Google disclosed that one of its Texas data centers used about 10,000 gallons in a year, roughly two months of a single household’s use, while its largest water consumer, in Iowa, used around a billion gallons.

  • Much of the installed base was designed when water efficiency was not a priority, and communities have had real cause for complaint in several documented cases. Reports from Arizona describe data centers competing with agriculture for limited water. That article comes from a cooling equipment company, so check local reporting for specific claims.

Water quality is also part of the story. Water quality and wastewater are the “elephant in the room,” because the debate focuses on gallons while fewer people ask what is in the water and where it goes.

Water is not the only environmental question

If you care about nature, electricity may be the bigger issue than water.

  • The IEA says global electricity demand from data centres is set to more than double over the next five years, using as much electricity by 2030 as all of Japan does today.

  • Data centres made up about 1.5% of global electricity use in 2024, and that share has grown by 12% a year over the past five years.

  • Data center emissions are expected to rise from 180 million tonnes of CO2 today to 300 million tonnes by 2035.

Context matters here too. That is a minimal share of the 41.6 billion tonnes of global emissions estimated for 2024. The IEA says the emissions increase would be small and could be offset if AI is widely adopted to cut emissions elsewhere. That offset is a possibility, not a guarantee. The same report also flags rising demand for the critical minerals used in data center equipment, which has its own mining and habitat costs.

The real problem: we can’t see the full picture

Perhaps the most important finding is about transparency. A 2026 peer reviewed comparison of the largest AI buyers found that no company reports any AI specific metrics for water or carbon, even though Google, Meta and Microsoft link rising electricity and water use to AI growth in their own disclosures.

Until companies report AI’s water and energy use separately, the public is left arguing over estimates.

Myth or reality? The verdict

Myth: AI is draining the world’s water supply. Nationally, data centers are a small user, and the “bottle of water per chat” claim does not hold up.

Reality: AI’s footprint is real, it is growing, and it can be serious in water stressed places and on dirty power grids. Its water use is real, rising and locally significant, even though per query use is small and other industries use far more.

What matters most: Where a data center is built, how it is cooled, and what powers it. The same AI model can have a very different impact depending on those three choices.

What can be done

  • Build data centers away from water stressed regions, or use reclaimed water.

  • Use closed loop and liquid cooling designs.

  • Power them with clean energy, which cuts both emissions and indirect water use.

  • Require companies to publish AI specific water and energy data.