Every time you ask an AI chatbot to draft an email or summarize a report, a data center somewhere is quietly using water to keep its servers cool. That reality sits behind a growing conversation about AI water usage, and the facts are often reported in confusing or exaggerated ways. This guide lays out what is actually known, using figures from peer-reviewed research and the technology companies themselves.
The short version is that a single AI request uses a very small amount of water, but the total across billions of daily requests adds up fast. Where those data centers are built matters just as much as how much water they use. Here is a clear, fact-checked look at AI water usage, why it happens, and what it means for a business that relies on these tools.
What Is AI Water Usage?
AI water usage is the freshwater consumed to run the data centers that train and operate artificial intelligence models. Researchers split it into two parts. On-site water cools the servers directly, and off-site water is used at power plants to generate the electricity those servers need.
The University of California, Riverside research team behind the widely cited study “Making AI Less Thirsty” calls these scope-1 and scope-2 water use. Most of it is what scientists call blue water, meaning it is drawn from rivers, lakes, or groundwater. A large share is clean, drinking-quality water rather than recycled or salt water.
That distinction is the heart of the concern. AI water usage is not just a big number. It is often high-quality freshwater pulled from the same supply that homes and farms depend on.
How Much Water Does AI Actually Use Per Request?
A single AI request uses a small amount of water, but the exact figure is genuinely disputed. The gap comes down to what each estimate counts.
OpenAI chief executive Sam Altman said in 2025 that an average ChatGPT query uses about 0.000085 gallons of water, roughly one-fifteenth of a teaspoon. Independent researchers point out that this figure appears to count only the water evaporated inside the data center. It leaves out the water used to generate the electricity and the water used to train the model in the first place.
The UC Riverside team estimated that a longer AI conversation can consume about one 500-milliliter bottle of water for every 10 to 50 responses, once both on-site and off-site water are counted. Their numbers shift with location and time of day, because cooling needs rise in hot, dry weather. Both estimates can be true at once. They are simply measuring different things.
Why Training an AI Model Uses So Much Water
Training is where AI water usage becomes dramatic. Building a model means running thousands of specialized chips at full power for weeks or months, and all that heat has to go somewhere.
The same UC Riverside study estimated that training the GPT-3 model in Microsoft’s US data centers directly evaporated about 700,000 liters of clean freshwater. Counting the water used to generate the electricity, the total reached roughly 5.4 million liters for a single training run. Newer, larger models require far more computing power, which is a major reason data center water consumption keeps climbing.
Why Data Centers Need Water at All
Data centers use water because evaporating it is a cheap and effective way to remove heat. Many large facilities rely on evaporative cooling, which works like sweat on skin. Warm air passes over water, the water evaporates, and the air cools down.
The tradeoff is that evaporated water leaves the local system. It is not returned to the river or reservoir it came from. Cooling with water instead of electricity-hungry air conditioning saves energy, so for years it was the practical choice. The rise of power-dense AI hardware has pushed that tradeoff into the spotlight.
The Numbers Behind Data Center Water Consumption
The large technology companies publish some of these figures, and the trend is clearly upward. These are the most reliable public numbers available.
Google reported that its data centers consumed about 6.1 billion gallons of water in 2023, an increase of roughly 14 percent from the year before, which it linked to AI growth. Microsoft reported that its global water consumption rose about 34 percent in a single year, reaching close to 6.4 million cubic meters. Both companies now report water as a core sustainability metric, which was not always the case.
Looking ahead, the UC Riverside researchers projected that global AI could withdraw between 4.2 and 6.6 billion cubic meters of water in 2027. For scale, that is several times the total annual water withdrawal of a country like Denmark. Rising AI water usage is not a rounding error at that level.
Why Location Makes AI Water Usage a Bigger Problem
Where a data center sits matters as much as how much water it uses. The same gallon of water means very different things in a rain-rich region versus a drought-prone one.
A Bloomberg News analysis found that roughly two-thirds of the data centers built or planned in the United States since 2022 are in areas already facing high water stress. Dry regions are often attractive for other reasons, including cheap land and reliable power, but they are the places least able to spare freshwater. That mismatch is why AI water usage draws local concern even when the national totals look manageable.
What Companies Are Doing to Reduce AI Water Usage
The industry is responding, and some of the fixes are meaningful. The direction of travel is toward cooling that uses far less freshwater, or none at all.
Microsoft announced a zero-water-evaporation cooling design in 2024 for new data centers, which it says can avoid more than 125 million liters of water per facility each year by recirculating the same fluid in a closed loop. Other common approaches include using recycled or non-drinkable water for cooling, shifting heavy workloads to cooler hours, and building in climates where air cooling is practical for more of the year. None of these erase AI water usage, but together they can bend the curve.
What AI Water Usage Means for Your Business
For most small businesses, the takeaway is awareness rather than alarm. The AI tools you use every day carry a real resource cost, and that cost is worth understanding as these tools become central to how you work.
If sustainability is part of your brand, AI water usage is a fair thing to factor into vendor choices and public commitments. You can favor providers that report their water and energy footprints, practice responsible AI use, and be accurate when you talk about it with customers. The facts here are strong enough that you never need to exaggerate them, and getting them right protects your credibility.
AI Water Usage: Frequently Asked Questions
How much water does one ChatGPT question use?
Estimates vary by what they count. OpenAI puts an average query at about one-fifteenth of a teaspoon, counting only on-site cooling. Independent researchers estimate closer to a 500-milliliter bottle per 10 to 50 responses once electricity generation is included. Both can be accurate for their scope.
Why do AI data centers use water instead of just electricity?
Evaporating water is a cheap, energy-efficient way to remove the intense heat that AI chips produce. Water cooling uses less electricity than air conditioning, so it became the standard choice. The downside is that evaporated water leaves the local supply, which is why the practice now gets more scrutiny.
Is AI really causing water shortages?
AI is one of several growing pressures on freshwater, not the sole cause of shortages. The concern is concentration. When large data centers cluster in already dry regions, their local demand can strain communities even though the national total is a small share of overall water use.
How much water did training GPT-3 use?
UC Riverside researchers estimated that training GPT-3 in Microsoft’s US data centers directly evaporated about 700,000 liters of clean freshwater. Including the water used to generate the electricity, the total for one training run reached roughly 5.4 million liters. Larger modern models generally require more.
Can AI water usage be reduced?
Yes, and it already is being reduced in places. Closed-loop and zero-evaporation cooling, recycled water, and smarter siting all cut freshwater demand. Microsoft says its newer zero-water cooling design can save more than 125 million liters per data center each year compared with older evaporative systems.
Should my business stop using AI tools to save water?
Stopping is rarely necessary or practical. A more useful approach is to use AI deliberately, choose providers that disclose their environmental footprint, and represent the facts honestly in your marketing. Awareness and accuracy matter more here than avoidance.
Stay Ahead of AI’s Real Costs and Benefits
Understanding the full picture of the tools you use is part of running a modern business well. If you want help using AI effectively while keeping your data, claims, and reputation clean, our team can guide the strategy. Explore our SEO and analytics services to build content that earns trust, and if the environmental and legal side of AI feels overwhelming, reach out to Demur Design for a straight conversation. To keep up with clear, fact-checked AI coverage, subscribe to the Demur Design newsletter in the footer below.
Sources
This explainer is researched and drafted with AI, then reviewed, fact-checked, and published by Demur Design.
- Li, Yang, Islam, and Ren, “Making AI Less Thirsty,” University of California, Riverside (arXiv): https://arxiv.org/abs/2304.03271
- UC Riverside News, “AI programs consume large volumes of scarce water”: https://news.ucr.edu/articles/2023/04/28/ai-programs-consume-large-volumes-scarce-water
- Data Center Dynamics on Sam Altman’s per-query water and energy figures: Data Center Dynamics
- Bloomberg News analysis, “The AI Boom Is Draining Water From the Areas That Need It Most”: Bloomberg
- UK Government report, “Water use in AI and Data Centres”: GOV.UK
