Does AI Use a Lot of Water

Does AI Use a Lot of Water? The Hidden Water Crisis Behind Every Prompt (2026)

You've probably heard that AI uses a lot of energy. But water? That one catches most people off guard.

The truth is that every time you type a prompt into ChatGPT, generate an image with Midjourney, or use an AI-powered search, somewhere in the world a data center is consuming fresh water to cool the servers, making that possible. And the scale of that consumption in 2026 is staggering — and growing fast.

This article breaks down exactly how and why AI uses water, how much it's consuming globally, and what it means for the future of freshwater supplies.

Part of our AI & Environment series. For the full picture, read our pillar guide: How Does AI Affect the Environment?

Why Does AI Need Water at All?

Here's the thing: most people don't connect that AI runs on hardware, and hardware generates enormous heat.

Data centers — the massive warehouse-like buildings filled with thousands of servers — need to stay cool 24 hours a day, 7 days a week. If servers overheat, they fail. So data centers use water-based cooling systems to absorb that heat and carry it away.

There are two main ways this happens:

1. On-site cooling towers — Water absorbs heat from the server rooms, then gets pumped to cooling towers where much of it evaporates into the air. This evaporated water is lost — it doesn't return to the source.

2. Direct-to-chip liquid cooling — More advanced systems pipe liquid coolant directly to the GPUs and CPUs, doing the actual computing. This is more efficient but still consumes water.

On average, data centers use about 1.9 liters of water for every kilowatt-hour of energy consumed. Some older or less efficient facilities use up to 2.4 liters per kWh.

And since AI workloads use far more energy than regular computing tasks — MIT researchers found that a generative AI cluster uses seven to eight times more energy than a standard computing workload — the water consumption multiplies accordingly.

How Much Water Does AI Actually Use?

Let's get into the real numbers.

Per Query

• A single 100-word AI prompt uses approximately 519 milliliters of water — roughly one standard water bottle — according to researchers at the University of California, Riverside
• Generating a 100-word email with GPT-4 uses about three 500ml bottles of water
• Even a basic Google AI search uses around 10ml per query
• OpenAI CEO Sam Altman stated in 2025 that the average ChatGPT query uses about 0.32ml of water — though this likely reflects only on-site cooling and not the full chain, including electricity generation

Per Training Run

• Training GPT-3 required approximately 700,000 liters of water for cooling
• As AI models grow larger with each generation, training water costs are rising 50% year over year

At Data Center Scale

• In 2022, Google, Microsoft, and Meta together used an estimated 580 billion gallons of water to power and cool their data centers — enough to meet the annual water needs of 15 million households
• Microsoft's total water use surged 34% in a single year (2023), rising to 6.4 billion gallons annually, largely due to ChatGPT infrastructure
• Google's Iowa data center alone uses 4 million gallons of water every day to cool AI models
• An average mid-sized data center now consumes approximately 1.4 million liters (370,000 gallons) of water per day
• Data centers in Texas are projected to use 399 billion gallons of water by 2030

Global Projections

• U.S. data center water consumption could grow from 200 billion gallons (2021) to 1 trillion gallons by 2030
• Global AI water use could reach 4.2–6.6 billion cubic meters by 2027, equivalent to the total annual water consumption of Denmark
• By 2040, AI could consume up to 1.5% of the world's total freshwater supply

Where Is This Water Coming From?

This is where the problem gets serious — and personal for many communities.

Data centers aren't being built in the ocean. They're being built in cities, suburbs, and rural areas — often drawing from the same municipal water supplies, aquifers, and rivers that local residents, farmers, and ecosystems depend on.

Drought-Prone Locations

Many of the world's biggest AI data centers are located in areas that are already water-stressed:

• Phoenix, Arizona — one of the driest cities in the US — has seen a projected 32% increase in water stress due to data center expansion
• The Dalles, Oregon — Google's data centers there make up 25% of the entire city's water use, and consumption has nearly tripled since 2017
• Northern Virginia — the largest data center hub in the world — is drawing heavily on Potomac River water
• 30% of EU data center locations are in areas already facing water stress

One study found that by the 2050s, nearly 45% of data center facilities globally may face high exposure to water stress if current expansion trends continue.

The Chip Manufacturing Hidden Cost

It's not just the cooling. Before a server even reaches a data center, the chips inside it have already consumed enormous amounts of water.

Manufacturing semiconductors requires "ultrapure" water to rinse off silicon residue without damaging the chips. A typical chip factory uses about 10 million gallons of ultrapure water per day — as much as 33,000 U.S. households. It takes approximately 1.5 gallons of tap water to produce just one gallon of ultrapure water.

Every AI chip has a water footprint that starts long before it ever processes a single prompt.

The Fossil Fuel Multiplier

About half of the electricity used by U.S. data centers currently comes from fossil fuel power plants — and those plants also consume water. Fossil fuel-based electricity generation is significantly more water-intensive than renewable alternatives, meaning that every AI query powered by coal or gas carries an additional, hidden water cost.

Is the Water Returned After Use?

Mostly no — and that's a critical part of the problem.

When data center cooling towers evaporate water, that water is gone from the local watershed. It enters the atmosphere as vapor rather than returning to the river, lake, or aquifer it came from.

For example, Google-owned data centers discharge only 20% of withdrawn water to wastewater treatment plants. The remaining 80% is evaporated or otherwise consumed without being returned.

This matters most in drought-prone regions. When a data center evaporates millions of gallons of water from a river system in Arizona or California, that water is not coming back to local farms, homes, or ecosystems.

The Transparency Problem

One of the biggest obstacles to understanding AI's water impact is that most tech companies don't fully disclose their water use.

Many data center operators use non-disclosure agreements to keep even basic water consumption figures from the public. Reports that do exist often combine AI and non-AI workloads, making it impossible to isolate the AI-specific impact.

The figures we do have come largely from occasional corporate sustainability reports, university research, and investigative journalism — not from standardized, mandatory disclosure. The EU is pushing for change, projecting a 15% reduction in AI water intensity by 2030 through new regulations — but enforcement and transparency remain works in progress.

Is There a Solution?

Yes — and the tech industry is aware of the problem, even if action has been slow.

Better Cooling Technologies

• Immersion cooling — submerging servers directly in non-conductive liquid — can reduce water use by up to 90% compared to evaporative cooling systems
• Direct-to-chip liquid cooling reduces water waste compared to open-loop evaporative systems
• Dry cooling systems (air-based, no water evaporation) can cut water use by around 30%, though they require more energy

Smarter Locations

Building data centers in cooler climates — Iceland, Scandinavia, coastal Canada — dramatically reduces cooling needs and water use. Some companies are already doing this, though the US and EU remain the dominant locations for political and infrastructure reasons.

Renewable Energy

Switching to solar and wind power reduces both carbon emissions and indirect water use, since fossil fuel power plants are far more water-intensive than renewables.

Efficiency Improvements

The water intensity of AI computing is projected to drop from around 2 liters per kWh today to approximately 0.5 liters per kWh by 2030 as cooling technology improves — though this gain may be offset by sheer volume growth.

What Does This Mean for Water Bills?

It's not just an environmental problem — it's starting to hit wallets too.

As AI water consumption spikes, water costs are rising in affected regions. When billions more gallons are needed annually, prices follow basic economics upward. Industries from manufacturing to food processing are already seeing higher water costs in areas with dense data center clusters — and residential water bills in those regions are feeling the pressure too.

Quick Stats Summary

FAQs: Does AI Use a Lot of Water?

Q1: Does AI really use that much water? Yes. The figures are large, real, and verified by multiple independent researchers, university studies, and corporate sustainability reports. The main debate is around the exact per-query figure — which varies by data center location, cooling system, and energy source — not whether AI uses significant water.

Q: Does ChatGPT use water? Yes. For a detailed breakdown of ChatGPT's specific water footprint, read our full article: How Damaging Is ChatGPT to the Environment?

Q: Is AI water use worse than agriculture? Global agriculture uses vastly more water than AI data centers. However, the key difference is trajectory — AI water use is growing at an explosive rate, while agricultural use has been relatively stable. In certain local areas (like Phoenix or Oregon's The Dalles), data centers are already competing directly with farms for water.

Q: Which AI company uses the most water? Microsoft and Google are the largest consumers based on available data. Microsoft's water use surged 34% in 2023 to 6.4 billion gallons annually. Google uses approximately 4 million gallons per day at its Iowa facility alone.

Q: Can AI companies reduce their water use? Yes. Immersion cooling, dry cooling, smarter location choices, and renewable energy all reduce water consumption significantly. Some companies are investing in these technologies, but adoption has been slower than the growth of AI demand.

Q: Will AI water use affect my local water supply? It depends on where you live. If you're in a region with a growing concentration of data centers — particularly in drought-prone areas of the US Southwest, parts of Europe, or rapidly developing tech hubs — the competition for local water resources is already real and measurable.