Artificial intelligence may exist in the digital world, but the infrastructure powering it is decidedly physical. AI requires enormous data centers filled with high-performance processors, electrical equipment, cooling systems, backup generators, and miles of supporting infrastructure. And as communities across America are discovering, those facilities can require substantial amounts of water.
That raises a straightforward question: Why should we use drinking-quality water to cool computers when reclaimed wastewater can potentially do the job?
If America intends to lead the world in artificial intelligence—and it should—we need to confront the infrastructure requirements that come with that ambition. The answer should not be to restrict AI development through another layer of bureaucratic regulation. Nor should government attempt to centrally plan where technology companies can build. Instead, policymakers and industry should encourage practical solutions that allow development while protecting resources taxpayers have already paid dearly to provide.
Reclaimed water offers one such solution.
Municipal wastewater is traditionally treated to prescribed standards and discharged into rivers, lakes or other receiving waters. Depending on the treatment process and local infrastructure, some of that water can instead be reclaimed for non-potable uses. Cooling industrial facilities and data centers is an obvious application worth considering.
There is little reason that water circulating through an industrial cooling system must begin its journey at the same quality standard as the water coming from a family’s kitchen faucet.
That distinction becomes increasingly important as data-center construction accelerates.
AI computing generates tremendous heat. Modern processors designed to train and operate sophisticated AI models consume significant amounts of electricity, and removing the resulting heat is an engineering challenge. Some facilities rely heavily on air cooling, while others use evaporative, liquid or hybrid cooling technologies. The resulting water demand varies considerably from one facility to another, meaning sweeping claims about how much water “AI” consumes can be misleading.
But the larger issue remains.
Where data centers require significant water, communities should ask what kind of water they are consuming.
Fresh potable water is expensive infrastructure. Taxpayers have financed reservoirs, pumping stations, treatment facilities, water mains and distribution systems so homes and businesses can receive safe water. In regions facing drought, population growth or aging infrastructure, that capacity becomes even more valuable.
Using reclaimed wastewater could reduce pressure on those systems while creating an economic use for water that municipalities already collect and treat.
There are complications, of course.
Reclaimed water is not magically free. It may require additional treatment depending upon its intended use. Dedicated pipelines may have to be constructed between wastewater facilities and data centers. Operators must manage mineral concentrations, corrosion, biological growth and other water-quality issues. Pumping and advanced treatment consume energy. Every proposed system therefore deserves a genuine cost-benefit analysis rather than an environmental slogan masquerading as engineering.
Location matters as well.
A data center sitting beside a major wastewater-treatment facility presents a very different economic proposition from one requiring construction of a 20-mile reclaimed-water pipeline. Government should be particularly cautious about forcing taxpayers to subsidize expensive infrastructure simply to benefit one enormously profitable technology company.
That is where conservative economic principles should guide policy.
The companies creating extraordinary demand for computing infrastructure should generally bear the costs associated with serving that demand. If a hyperscale data center needs a dedicated reclaimed-water pipeline, the public should ask who benefits, who pays, who owns the infrastructure and what happens if the facility eventually closes.
The arrangement should benefit taxpayers rather than quietly socializing infrastructure costs while privatizing profits.
There is also an opportunity here.
Wastewater-treatment plants have traditionally been viewed primarily as sanitation infrastructure. Increasingly, they can be understood as resource-recovery facilities. Treated water can have economic value. Biogas generated during treatment can provide energy. Nutrients can sometimes be recovered. Waste heat and other byproducts may also become useful under the right circumstances.
AI infrastructure could provide another market for reclaimed water.
Instead of framing the debate as technology versus the environment, communities should demand technological solutions to technological problems.
America will need more data centers if it wants domestic AI companies to compete against China and other international rivals. Attempting to suppress that development would risk surrendering an enormously consequential industry. But embracing AI does not require communities to hand technology companies unlimited access to public resources.
There is a sensible middle ground.
Build the data centers. Build the electrical generation required to power them. Accelerate permitting where bureaucracy unnecessarily obstructs construction. Encourage private investment in new infrastructure.
But where practical, stop cooling computers with drinking water.
If treated wastewater can safely and economically perform the same industrial function, putting that water to work is not radical environmentalism. It is resourcefulness.
The AI revolution will demand extraordinary amounts of physical infrastructure. America should build it aggressively—but intelligently.
And sometimes intelligent infrastructure begins with a remarkably simple principle: Don’t waste clean water where dirty water will do.