This is a real-world example of how measurement and attribution can shape how a problem is understood—and how easily we might end up solving the wrong one.
A Note on How to Think About This
Water use, like many issues, can become emotionally charged—especially in regions where the effects are visible and immediate. The concern is not misplaced.
When a topic becomes widely discussed, the way numbers are presented matters. Large, aggregated figures can create a strong impression when they combine multiple systems operating at different scales and locations.
You can care deeply about an issue and still question how it’s being measured.
What Data Centers Do—and Don’t
The term data center sounds simple— a place where data is stored. That’s incomplete. A data center is a facility that stores, processes, and moves data continuously. A data center performs three functions:
- Storage: Photos, files, backups, and databases are stored across distributed systems, replicated in multiple locations.
- Processing (Computes): Applications, transactions, software, AI model training and deployment.
- Movement (Network): Data moving between users, applications, and systems.
What a Data Center Does Not Do
A data center does not drive the system - it responds to it. It does not decide what data gets created, stored, deleted, or replicated. It does not control how efficiently software runs or how electricity is generated. Those decisions are made by users, product designs, business models, and energy systems.
The Claim
Data centers are using “massive amounts of water.” Numbers such as:
- 1–5 million gallons per day
- Equivalent to a town of 10,000–50,000 people
- AI will increase water use by 870%
Before reacting, ask a basic question:
What is being measured?
What Does “Water Use” Mean?
Most discussions blur water use from three different sources. When multiple systems are combined into one metric, the result isn’t clarity—it’s misattribution.
- Water Withdrawal: Water taken from a source and later returned.
- Water Consumption: Water is not returned due to evaporation or discharge elsewhere.
- Discharge: Water returns that are often ignored, creating a one-direction narrative. This is treated as invisible in the published numbers.
What the Largest Number Is Measuring
Much of the largest amount of water attributed to a data center is not measuring what happens at the building. It is measuring water used to generate the electricity the building consumes.
A data center cooling system is one water system. The power plant supplying electricity is another. Manufacturing is a third. When these are combined into a single number, they create the impression that the visible facility is responsible for all of it. In many cases, it isn’t.
The number is no longer describing the building.
It is describing the electricity system.
That changes the question from:
“How do we reduce water at the building?” to questions such as:
- How is electricity generated?
- How much water is required to generate electricity?
How much electricity does a data center demand?
Where the Water is Used
A “data center’s water footprint,” is reported as a single number. It’s the sum of three different systems operating in different places and times. Water attributed to data centers comes from:
- On-site cooling (at the facility)
- Power generation (off-site)
- Hardware manufacturing (upstream)
Data centers draw electricity from the grid. Power is almost never generated on-site. That means most energy-related water use occurs elsewhere.
Semiconductor manufacturing (hardware) uses significant amounts of water. That water is included in figures attributed to data centers. A large data center may use up to a few million gallons per day for cooling, while a semiconductor fabrication facility can use several million to tens of millions per day for manufacturing. These occur in different systems, at different times, and require different solutions.
All systems use water somewhere in their lifecycle. Some use water continuously during operation, primarily to manage heat. Others use water during manufacturing, with little ongoing use. A system that uses water once is not the same as one that uses it continuously. When these are combined into a single number, it becomes difficult to understand what is being measured.
Cooling Towers Are Not What People Think
Cooling towers are often misunderstood. They are efficient systems that reuse water. The primary loss is evaporation—not continuous withdrawal. Water is used. That is not the same as continuously extracting new water from the ground. Using a cooling tower and chillers inside a building is more efficient than running standard air-conditioning, which requires a greater amount of electricity.
Are Data Centers Unusual?
Large buildings—offices, hospitals, universities, airports—use similar cooling systems. Per unit of heat removed, these systems are similarly efficient. The difference is the load. Data centers operate continuously, generate more heat, and require constant cooling.
A data center doesn’t use more water because it’s a data center. It uses more water because it is a building that is always on.
Are Data Centers “Using” That Much Water?
Not in the way implied. Data centers use water for cooling. They depend on electricity. The water attributed to them is used in generating that electricity. The issue is attribution. Water is used in one place—at power plants—and attributed to another—the data center.
Why Per-Query Water Claims Seem Misleading
Statements like “one AI query uses X amount of water” sound precise. They describe averages layered onto a continuously operating system. A data center does not turn on and off per request. It has baseline power and cooling demand. Individual queries add incremental load.
It’s as saying every mile you drive uses a fixed amount of fuel, without acknowledging that the engine might be idling part time. A car idling still uses fuel. Driving faster adds more. It’s incremental, not starting from zero each time.
What the Numbers Point To
What the numbers are often measuring is not data centers, rather they are measuring electricity. Most of the water attributed to data centers is tied to power generation. That water is used at power plants—not at the data center.
Electricity is electricity—whether it powers a data center, a home, or a factory.
Electricity Usage
Data centers may feel big. They are a small slice of total electricity use. Data centers use approximately 4–5% of total U.S. electricity. That means roughly 95% is used elsewhere—homes, industry, transportation, and other systems. If water use is tied to electricity generation, then most water use follows total electricity demand—not data centers.
The numbers attributed to data centers are pointing to system-wide electricity use.
- Data centers (~3%) → small slice of total electricity
- Personal devices at home (~10–15%) → relatively small
- Everything else (~80–85%) → dominant driver (HVAC, industry, etc.)
If Electricity Were Unlimited
If electricity were abundant and not water-intensive to produce, the picture changes. Data centers would still use water for cooling, not fundamentally different from other large buildings operating continuously.
This is not just a data center question. It is a system-wide question about how we generate electricity, how much we consume, and what we choose to use that power for.
Focusing on data centers isolates one visible node.
It does not explain the system that drives it.
Water Allocation vs Water Use
The deeper concern is not usage. It is allocation. Water is needed for drinking, basic human needs, and agriculture, and ranching. The relevant question becomes:
Is data center water use competing with those ?
The answer depends on the source. If water used for cooling or power generation comes from non-potable sources, is recycled, or is not part of municipal supply, then the framing changes.
Not all water use draws from the same system.
The Framing Risk
When all water use is combined into a single number, it creates an impression: data centers are taking water directly from the same pool used for essential human needs. That is not always the case. Without separating source, system, and location, the conclusion becomes easy and incomplete.
Why the Story Focuses on Water
Electricity is abstract and expected.
“Megawatts” means little to most.
Water is tangible and local.
“Millions of gallons of water” is understood.
Research on this issue began with system-level estimates simplified into single numbers. The number got attached to the most visible part of the system.
The driver is electricity.
The story is being told through water.
We are focusing on the visible variable.
We’re focusing on what’s easy to see, not the problem.
Final Thoughts
This is not about data centers.
It is about what happens when:
- a large number is presented without context
- multiple parts of a problem are collapsed into a single variable
- solutions target what is visible
When I first heard that data centers use large amounts of water, I pictured them pumping water out of the ground faster than it could be replenished. Then I thought about how cooling towers work, and that picture didn’t hold. Those systems are designed to be efficient. They use less energy than alternatives like air conditioning.
That mismatch led me to ask a different question:
What is actually being measured and reported?
Data centers use water.
Much of the water attributed to them comes from how electricity is generated, not from what is used at the facility.
That raises another question:
If it were true that water is the constraint, what ways to generate power that do not rely on water?
If those options are limited, is the real constraint water or electricity?
The question isn’t whether data centers use water.
The question is whether we understand what is driving that use.
Most debates about data centers and water aren’t about water.
The number, as presented, is not real.
The number combines water use from multiple systems—power generation, cooling, and manufacturing—and assigns it to a data center itself.
The result is a number that exists in aggregate.
It does not describe what is happening at the facility.
The question is what that number is measuring—and whether that’s the real problem we should be solving.
Solve the right problem.