The AI Diary: AI Needs Drinking Water

Nov 19, 2025

Every time you type a question into ChatGPT, you’re using water. Sam Altman claims it’s about 1/15th of a teaspoon per query. But when you consider that ChatGPT alone handles over a billion messages daily, and add in Gemini, Claude, DeepSeek, and others, we’re talking about a serious resource drain.​​

Here’s what makes this controversial: some researchers say Altman’s numbers are way off. They estimate that 10-50 queries could consume around 500 milliliters of water when you factor in cooling and electricity generation. That’s a huge difference, and OpenAI won’t share their methodology.

Why AI Is So Thirsty

AI queries require massive computational power. Those calculations happen on specialized chips — the most powerful processors in the world — housed in enormous data centers. These chips generate intense heat. Without proper cooling, they’d overheat and break down.​

Traditional air cooling doesn’t cut it anymore. The industry has shifted to liquid cooling systems, which require clean drinking water to prevent bacterial growth and system corrosion.​​

Here’s how it works: coolant flows over the processing chips, absorbing heat. Water then cools that coolant before it recirculates. The heated water goes to cooling towers where fans and evaporation dissipate the heat. The problem? Up to 80% of that water evaporates — it’s gone from the water cycle permanently.​​

A typical 100-megawatt data center uses about 2 million liters of water daily. That’s equivalent to 6,500 households. More than 160 new AI data centers have been built in the U.S. in just three years (2022–2025) — a 70% increase.

It’s Not Just Cooling

Water consumption goes beyond what happens inside data centers. Generating electricity for these facilities accounts for roughly 60% of total water use. Coal, gas, and nuclear power plants heat water to create steam that drives turbines.​​

The International Energy Agency projects electricity demand for AI data centers will jump 400% by 2030 — matching the UK’s entire annual consumption. Nearly half of U.S. data center servers run on power from water-hungry plants in water-stressed regions.​​

Even manufacturing the chips requires significant water. Water is embedded throughout AI’s entire lifecycle — from raw materials to production to operations.

If you’re wondering how much fresh drinking water there is on Earth, it accounts for less than 1% of the planet’s total water.

Communities Are Fighting Back

People around the world are pushing back against data centers. Protests have erupted in Spain, India, Chile, Uruguay, the Netherlands, and parts of the U.S.​​

In central Mexico, residents near a Microsoft data center reported increased power outages and water shortages lasting weeks. Schools closed. Kids got sick with gastrointestinal illnesses. A local health clinic operator treated patients by flashlight and watched a 54-year-old farmer get denied oxygen during a power failure.​

Querétaro, Mexico experienced its worst drought in a century in 2024 while data center development accelerated. Local activist Teresa Roldán put it bluntly: “Water is what’s needed for the people, not for these industries”.​

In Chile, the government relaxed environmental standards for data centers, blocking communities from accessing information about environmental impacts. When residents asked for details, they hit walls of “corporate confidentiality”.​

Ireland now dedicates over 20% of its national electricity to data centers. Brazilian journalist Laís Martins described it as “data colonialism” that “repeats a colonial legacy of resource extractivism”.

Corporate Promises

Major tech companies have pledged to become “water positive” by 2030 — meaning they’ll replenish more water than they consume. Microsoft, Meta, Google, and Amazon have all made commitments.​

Their strategies include reducing water intensity and funding restoration projects. Facebook invested in projects that replenish over 850 million gallons annually. Google pledged to replenish 120% of water consumed.​

But experts are skeptical. “There is a long way to go to get to those kind of numbers”. As of 2023, nearly 80% of water consumption by Google’s AI data centers in the U.S. came from drinking water sources.

What’s Next?

The industry is experimenting with solutions. Companies are testing cooling systems that don’t evaporate water at all, and capturing heat from data centers to warm homes. Advanced techniques include direct-to-chip cooling and immersion cooling where entire servers are submerged in liquid.​​

The most radical idea? Moving data centers to space. In November 2025, startup Starcloud launched a satellite with an NVIDIA H100 GPU — 100 times more powerful than any computer previously operated in space. The vision: massive GPU clusters powered by constant solar energy, radiating heat into deep space, running with zero fresh water.​​

Google announced “Project Suncatcher” to launch prototype satellites by early 2027. China launched 12 satellites in May 2025 for a space-based computing constellation.​

Skeptics point to numerous hurdles: thermal management in vacuum, radiation tolerance, high-bandwidth communications, system reliability.

Dominance

Remember my previous article, ​“The AI Diary: World War III Will Be Fought with Code”​? Here’s the gist: AI dominance — the nation with greater computational power — translates into geopolitical dominance determining the next world leader.

That said, few will care how much water AI data centers consume. If AI needs water, it will get water — regardless of the cost — because the higher-stakes cost is falling behind in the race to build the most powerful AI models, which confer competitive advantage and global dominance.

This leads me to conclude that developing and deploying effective cooling systems is part of this race.

Think: where in the world is there abundant drinking water? Who owns those resources? Are the current owners aligned with major geopolitical players? Might those players seek control over these water supplies?

Bottom line: nobody cares how much water AI “drinks”. The real question is how to gain control over essential drinking‑water resources.