One proposed AI data center in Florida raises a simple but serious question: how much water should one project be allowed to demand before the public gets a clear answer?
______________________________________________
Artificial Intelligence, it lives on our screens. We type a question, the answer appears, or we prompt it to create an image, an info-graphic or edit our work. Summarize a website or an entire book. The task once took several hours, or days. Now, a few minutes.
It’s fun, its challenging, it creates jobs and in many cases giving people a new career and lifestyle.
However, somewhere, far away from the screen of our computers, there are machines running day and night. These machines are producing heat, heat that needs to be cooled. Cooling, in many cases means using water.
I was reading the proposed Stonebridge data center project paper that is in Fort Meade, Florida. In Florida Commerce Secretary Alex Kelly warned:
Further, the development will require extensive commitments of natural resources including water and energy, all of which need to be adequately and transparently addressed in the Sunshine to preserve the quality of life in the community and regain trust from your citizenry.
Yes, that is the right question to ask. It wasn’t about the popularity of AI, or whether technology benefits from it or if companies have a right to build.
The question was simply stated: How much water will this take, and has the public been told the truth clearly enough to understand what is being approved?
Look at the numbers.
According to WFSU, the proposed Stonebridge project in rural Polk County would cover about 1,300 acres and involve a 4,400,000-square-foot data center. The developer initially said the project would need about 140,000 gallons of water per day to operate. Later, Stonebridge reportedly said it would need 50,000 gallons per day for ordinary daily operations, such as bathrooms and break rooms. Kelly’s letter described the 50,000-gallon figure as the “minimum demand” and said the projected potable water demand appeared “woefully underestimated” for a project of that size.(read it here)
Let those numbers sit in your thinking for a minute.
50,000 gallons a day equals about 18.25 million gallons a year.
140,000 gallons a day equals about 51.1 million gallons a year.
WFSU reported that permits for Project Stonebridge still need to be obtained from the Southwest Florida Water Management District, and that projected water demand had not yet been included in the permit information referenced by the district.
The heart of the matter is, a community cannot honestly weigh a project’s benefits against its costs if one of the most basic costs, water, is unclear.
Why data centers need so much water?
Data centers are filled with servers and computer systems that run constantly. The more powerful the computing, the more heat is produced. AI systems, especially large AI systems, can require enormous computing power, and that heat has to go somewhere.
Some facilities rely more heavily on air cooling. Others use evaporative cooling, where water absorbs heat and much of that water is lost as vapor. The MOST Policy Initiative explains that evaporative cooling can reduce electricity used for cooling, but it creates a tradeoff by increasing water use.
The Lawrence Berkeley National Laboratory’s data center efficiency program also states plainly that data centers consume both energy and water, with direct water use generally tied to cooling needs and indirect water use tied to electricity generation.
This means the water question is not separate from the energy question, the two are connected.
If a data center uses electricity from power plants that also require water, then the total water footprint can be larger that what is seen at the data center fence line.
Stonebridge is only one project and it should concern every community.
If one proposed project can raise questions about tens of thousands or hundreds of thousands of gallons per day, then what happens when several projects arrive in the same region? What happens when one county approves one, then another county approves one, then a nearby city approves another?
A single project may be described as manageable. But ten “manageable” projects can become a regional burden. By the time people realize the combined effect, the permits may already be issued, the land may already be cleared, and the community may already be locked into the consequences.
The reason why Kelly’s statement matters. He pointed to the thing that should come first in any honest public discussion: transparency.
How big can water demand get?
The largest already operating AI-related facilities are difficult to compare because companies do not always report water use in the same way, and many do not disclose site-level water figures at all. But the reported numbers that are available show why people are worried.
In Memphis, xAI’s Colossus supercomputer facility has been described as one of the world’s largest AI supercomputing sites. Governing reported that Colossus uses water in two ways: a closed loop for cooling chips and an open loop that goes to cooling towers. According to xAI engineer Mark Carroll, the open loop at Colossus needs 3 million gallons per day.
That is about 1.095 billion gallons a year.
Governing also reported that xAI is building a water recycling plant so the facility can use treated wastewater rather than drinking water for cooling. That is a better approach than using potable water, but the number still deserves attention because it shows the scale of cooling demand for major AI infrastructure.
Quoted from their site:
- An enormous quantity of water is needed to prevent servers in data centers from overheating.
- A Memphis xAI facility, which builds Grok, is located near a wastewater treatment plant. The data center will use water from the plant instead of drinking water. (read more)
Another useful comparison comes from Google’s data centers. The MOST Policy Initiative reported that Google’s largest data center, in Council Bluffs, Iowa, withdrew an average of 3.9 million gallons of water per day and consumed 2.8 million gallons per day.
Millions of gallons per day at the largest sites.
The information is not hiding folks, it is on the Internet in black and white, read it, stay informed, stay educated.
According to the Environmental and Energy Study Institute, large data centers can consume up to 5 million gallons per day, comparable to the water use of a town of 10,000 to 50,000 people.
So when a proposed 4.4-million-square-foot facility is discussed in terms of 50,000 or 140,000 gallons per day, it is reasonable for public officials and residents to ask: is that the whole number, or only part of the number?
The use of bathroom water being used is not the cooling numbers. This is where the language matters.
A figure for bathrooms, break rooms, landscaping, office use, and employee needs is not the same as a full cooling demand for a high-powered AI facility.
When people hear “50,000 gallons per day,” they may imagine that is the project’s total water use. But according to the WFSU report, that figure was described as servicing everyday operations like bathrooms or break rooms, while Kelly called it the “minimum demand.”
For an ordinary office building, bathrooms and break rooms may be a major part of water use.
For a massive AI data center, they may be the easy part to count.
The harder questions is about the cooling.
- How will the heat be removed?
- Will the project use potable water?
- Will it use reclaimed water?
- Will it use cooling towers?
- How much will evaporate?
- How much will be withdrawn?
- How much will be returned?
- How much will be consumed and gone from immediate use?
- What happens during drought?
- What happens when other developments need water too?
These are questions that might be asked before the permits have been given and development has been started.
Many people use AI and still care deeply about the land, water, birds, farms,wetlands. rural towns, and quality of life. However, if AI is going to become part of our daily life, then we as the public have a right to know and understand what must be given up to support it.
Homeowners should know what is expected of what their well can provide, a farmer needs to know about his irrigation challenges, or a cities plan for growth.
Water should not be treated as a minor detail for data centers, it is not minor.
Kelly’s statement about addressing water and energy “in the Sunshine” is important because it speaks to public trust.
People are not wrong to feel suspicious when projects move quickly, when numbers change, when operators are not disclosed, or when basic resource demands are not plainly explained. WFSU reported that the operator for the proposed Fort Meade data center had not yet been disclosed, according to Kelly.
A community should not have to piece together the real cost from scattered sources. Those numbers should be clear and upfront.
How much water is needed? From what source? At what daily maximum? What of droughts? Paid for by whom? Monitored by whom?
___________________________________________________
Sources and Further Reading
- WFSU, “Florida’s commerce secretary comes out against a proposed AI data center,” April 20, 2026.
- Governing, “Wastewater Will Cool This Memphis Data Center,” March 16, 2026.
- MOST Policy Initiative, “Data Center Water Use,” April 8, 2026.
- Environmental and Energy Study Institute, “Data Centers and Water Consumption,” June 25, 2025.
- Lawrence Berkeley National Laboratory, Center of Expertise for Data Center Efficiency, “Water Efficiency.”