A new analysis shows that most planned AI data centers in the US are being built in drought-stricken regions, creating an insurmountable challenge.
The AI industry has been obsessed with one resource- compute.
Who has the most GPUs? Who can build data centers the fastest? Who can secure enough power to stay ahead in the AI race?
But a new analysis from The Guardian suggests the industry may have overlooked another resource that is becoming just as important: water. About two-thirds of planned AI datacenters in the US will be built in regions already experiencing severe drought conditions, even as demand for water-intensive cooling continues to rise.
That creates an uncomfortable contradiction.
The AI industry talks about intelligence as if it exists in the cloud. In reality, every AI model ultimately runs on physical infrastructure. Servers need electricity. Chips generate heat. Heat needs cooling. And cooling requires enormous amounts of water in several cases. Some large data centers consume millions of gallons daily, and overall, this water use could rise dramatically over the next few years.
That is where the story becomes larger than sustainability.
It becomes an investment story.
Investors evaluated AI companies based on model capabilities, adoption rates, and infrastructure scale. The assumption was straightforward: more infrastructure meant a stronger competitive position.
Now, a new variable is entering the equation.
Resource constraints.
The challenge is building a data center that communities, regulators, and local ecosystems tolerate. Opposition to new projects is already growing, with concerns over water use helping drive political pushback and even proposals to pause large data center developments in some regions.
That creates a risk many investors aren’t accustomed to pricing.
The AI boom is largely valued as a software revolution. Increasingly, it looks like an infrastructure business. And infrastructure businesses are constrained by land, energy, permitting, and natural resources.
The industry understands the problem. Companies are experimenting with closed-loop cooling systems, water-recycling initiatives, and alternative datacenter designs to reduce consumption. Microsoft recently claimed some of its newest facilities use less water than older designs.
But efficiency alone may not solve the issue.
The problem isn’t that individual data centers are becoming less efficient. It’s that demand is growing faster than efficiency gains can offset. Every improvement seems to unlock another wave of construction.
That may be the most important takeaway for tech investors.
The AI race is often framed as a battle for chips and talent. Yet some of the biggest winners over the next decade may be companies that solve a much less glamorous problem: how to scale AI without exhausting the resources that support it.
The future of AI will now be determined by who can find efficient ways to instill regular water inflow.
