Here is a number that should bother you: training a single large AI model can consume 700,000 liters of water for cooling, according to researchers at the University of California Riverside. The AI boom is turning data centers into the planet’s most wasteful water consumers. And some engineers are now asking a question that sounds insane until you look at the chemistry: can we use urine instead?
Why This Is a Real Conversation in 2026
Data centers do not run on magic. They run on electricity and water. A lot of water.
According to the International Energy Agency, data centers consumed roughly 340 terawatt hours of electricity in 2022. That number has nearly doubled since then as AI workloads exploded. But the water story gets far less attention than the power story. Most large facilities use liquid cooling because it is dramatically more efficient than blowing air over hot chips. The problem is that the biggest AI campuses are being built in places without abundant water. Arizona, Nevada, Texas. Hot places with cheap land and tax incentives. Also places with serious and worsening water shortages.
Microsoft disclosed that its data centers consumed 6.4 million cubic meters of water in 2022, according to its own sustainability report. That was up 34% from the year before. And this was before the full generative AI wave hit capacity planning. The combination of more facilities, hotter chips, and denser GPU racks is pushing water consumption into territory that regulators and water districts are starting to push back on hard.
So engineers are hunting for alternative cooling fluids. And yes, urine is on the table. Not as a punchline. As a serious engineering input.
The Chemistry Actually Works
I know what you are thinking. Stay with me because this gets interesting fast.
Urine is about 95% water. The rest includes urea, creatinine, ammonia compounds, and various salts. The part that matters for cooling is what you can extract when you process urine at scale. Urea breaks down into ammonia. And ammonia, known industrially as R-717, is one of the most effective refrigerants ever discovered.
Ammonia refrigeration has been used in industrial cooling for over 130 years. It transfers heat more efficiently than most synthetic refrigerants. It has zero global warming potential. And according to the American Society of Heating, Refrigerating and Air-Conditioning Engineers, ammonia systems can be 3 to 10% more energy efficient than equivalent hydrofluorocarbon refrigerant systems in large-scale industrial applications.
Researchers at institutions in the UK and Netherlands have been testing bioelectrochemical systems that pull ammonia directly from urine streams. The recovered ammonia feeds into absorption refrigeration systems, which use heat rather than electricity to drive a cooling cycle. For a data center that generates enormous amounts of waste heat, an absorption cooling system powered by ammonia from urine is not fiction. It is a closed loop that could cut both water consumption and electricity use for cooling at the same time.
The scale math is real too. A metropolitan area of 1 million people produces roughly 1.5 million liters of urine per day. That is ammonia precursor sitting in sewage treatment plants right now. For free. Data center operators who build near wastewater treatment infrastructure are not making a coincidental choice. They are positioning for this kind of resource play.
If you are building content around AI infrastructure shifts for investors or business owners, I have been using InVideo AI to turn dense technical stories exactly like this one into short explainer videos that actually get watched and shared. The AI handles scripting and editing in minutes, which matters when the news cycle moves fast.
What This Means For Operators and Investors
If you are an investor, start paying attention to the water angle in AI infrastructure plays. Everyone is focused on chips and power grids. The next real constraint is water. Companies that solve the cooling water problem will have a serious cost advantage in markets where water is priced correctly or starts getting rationed by regulators.
Here is what I would do. I would watch the industrial ammonia sector more closely. Ammonia producers and distributors are already benefiting from green hydrogen demand. If ammonia sourced from urine or wastewater scales into data center cooling, it adds another demand driver to an already interesting industrial commodity story.
I would also track any data center operators building campuses adjacent to large wastewater treatment facilities. That proximity is a resource strategy, not a coincidence. It means consistent ammonia feedstock and a built-in heat sink for absorption cooling systems. That is a defensible cost structure that competitors sitting in suburban office parks cannot replicate quickly.
For smaller operators managing server infrastructure and trying to track energy and cooling costs, AppSumo regularly features lifetime deals on monitoring and operations tools that would otherwise run hundreds per year in subscriptions. Worth keeping an eye on if you are managing any meaningful infrastructure yourself.
The broader point is that the AI buildout is creating resource constraints most people are not thinking about yet. Water is going to be a bigger story than electricity in several major markets. The operators who cracked the problem using unconventional inputs, including waste streams, will win on margins while everyone else fights over aquifer rights and cooling tower permits.
The Bottom Line
Cooling servers with urine sounds like a joke. It is not. The chemistry is proven. The industrial precedent with ammonia refrigeration is 130 years old. And as water becomes the binding constraint on AI infrastructure expansion in dry markets across the American Southwest and beyond, ammonia from wastewater goes from a curiosity to a competitive weapon. Someone will build this at scale. The only question is who locks in that cost structure first and how quickly the rest of the industry scrambles to catch up.
Frequently Asked Questions
Can urine actually be used to cool data centers?
Yes, the science is real and not new. Urine contains urea which breaks down into ammonia, one of the most efficient industrial refrigerants in use for over a century. Researchers are actively developing bioelectrochemical systems that extract ammonia from urine streams and feed it into absorption cooling systems suitable for large facilities including data centers.
How much water does data center cooling actually use?
According to Microsoft’s own sustainability report, its data centers used 6.4 million cubic meters of water in 2022, a 34% jump from the prior year. At the workload level, University of California Riverside research found that training a single large AI model can consume 700,000 liters of water for cooling purposes alone.
Is ammonia cooling safe enough for data centers?
Ammonia refrigeration has been used safely in industrial settings including food processing, cold storage, and manufacturing for over 130 years. Established safety protocols and engineering standards exist for containment and leak management. The main challenges for data center applications are adapting those systems to higher-density server environments, and that engineering work is already underway at several research institutions.
Why are engineers looking at urine specifically and not just recycled water?
The value in urine is the ammonia precursor it contains, not the water itself. Standard recycled water does not carry meaningful ammonia content. The urea in urine breaks down to ammonia through a well-understood chemical process, giving engineers access to a refrigerant-grade compound from a waste stream that cities produce in enormous and consistent volumes every single day.
Which companies are building data center cooling innovation around water scarcity?
Microsoft, Google, and Meta have all disclosed internal water reduction targets and are testing immersion cooling and alternative heat exchange methods at pilot facilities. No major operator has publicly committed to a urine-sourced ammonia cooling system yet, but the underlying engineering is being validated in wastewater and industrial contexts that feed directly into data center applications.
