Why Everyone’s Talking About Data Centers (and Missing the Point)
 

My feed has been full of hot takes on data centers and water. At first, I was excited. A national conversation about water infrastructure! We never get that kind of attention unless it’s a story about a catastrophic failure. Surely, I thought, all the money and momentum behind the AI boom could help rebuild some of our nation’s aging systems.
Of course, it never works out the way you’d like.

The first wave of articles from the mainstream media grabbed attention with dramatic headlines and sometimes shaky research. The New York Times, a favorite punching bag for opinion writers, became a focal point. Their article, "Their Water Taps Ran Dry When Meta Built Next Door," seemed to insinuate that the operation of a data center stole water from a Georgia family. But when you read the piece, it describes a possibility that data center construction caused well water problems. This turned out to be a horrible example to anchor the story on. The mismatch between potential construction issues and the headline's insinuation completely overwhelmed the article's valid points about water consumption.
Then came the backlash. Most notable was probably Matthew Yglesias’s piece, "There's plenty of water for data centers," and since the blogosphere is a copycat league, a torrent of similar articles followed. The general thrust was that the NYT was full of it, agriculture uses a ton of water, and data centers represent a tiny percentage of national use.

And here’s the thing: no one is entirely wrong. If you look past the flawed well water example, the Times correctly identifies real concerns: data center construction is expanding at a breakneck pace, future rate hikes are almost certain, and the risk of localized water shortages is growing. Yglesias is also right: energy infrastructure will likely be a bigger bottleneck than water, agriculture uses the lion's share of water and could be far more efficient, and in aggregate, the U.S. is a water-rich country.
But here’s a key piece both sides miss. Not all water is the same. Agriculture often uses untreated irrigation water, think alfalfa fields or golf courses. Data centers, by contrast, need potable water or reclaimed water, either way requiring treatment and distribution through infrastructure designed for that purpose. You can’t just close golf courses and redirect their irrigation supply to a data center; the logistics and treatment requirements make that impossible. Water systems are built around intended use, and reallocating flows is neither simple nor cheap.
And this matters because data centers aren’t chasing water, they’re chasing cheap power. Water is relatively inexpensive and often treated as a secondary concern. That means the burden of ensuring treated supply, whether potable or reclaimed, falls squarely on the public sector. Utilities must process it, distribute it, and upgrade systems to meet sudden new demand. Meanwhile, the companies driving this demand are some of the wealthiest and most politically powerful in the world.

The problem, then, is not that either wave of commentary is completely wrong. It’s that both miss the structural issue. The AI boom didn’t create America’s water infrastructure problems, it exposed them.

Hyperscale data centers, with their immense, clustered, and round-the-clock water demand, act like a massive, unscheduled stress test. A single large facility can consume anywhere from 300,000 to 4.5 million gallons per day^[1], a demand comparable to that of a small town.

When you inject that kind of load onto an already strained system, the hidden weaknesses are exposed.

In that sense, data centers aren't the cause of the crisis. They are the accelerant that forces us to confront it. And while the current conversation may be messy, I have to say it clearly: I'm glad water is finally part of the tech debate. We desperately need this focus. But we also need the right framing, because the real story isn't about blame; it's about a fundamental, structural mismatch. Data moves at the speed of venture capital; water moves at the speed of municipal bonds.

Take, for example, Northern Virginia's Loudoun County, the world's highest concentration of data centers. It hosts over 25 million square feet of data center space, processing an estimated 70% of global internet traffic. Data center water use there tripled between 2019 and 2023, reaching nearly 900 million gallons annually just from potable supplies. This happened despite the county's award-winning reclaimed water program, which was specifically designed to supply data center cooling. The timing gap is stark: data centers were built and brought online faster than the reclaimed water distribution network could expand to serve them all.2

Note that data centers quadrupled over 6 years (projected for 8x over 10 years) while water supply grew by ~ 25% over 20 years)^[2]

The Loudoun story illustrates what I call the infrastructure timing gap; our world now runs on two fundamentally different clocks. A data center can go from site selection to full operation in 18 to 36 months. The water treatment upgrades and reuse networks they depend on take 5 to 10 years to plan, fund, and build. The transmission lines that carry their power can take 7 to 15 years. The math simply doesn't add up.

This isn’t about bad actors or poor intentions. Public infrastructure is deliberately methodical, layered with environmental reviews, bond funding approvals, and civic oversight designed to ensure it serves communities reliably for decades. Digital infrastructure is deliberately rapid, driven by fierce competition and the relentless pace of technological change. When these two systems collide, communities get caught in the middle, watching their resources strain under loads that arrived faster than anyone could prepare for.

Here’s why this timing gap is so dangerous: its effects remain invisible until it’s too late. The strain builds quietly, hiding until a shock tips the system over.

Water professionals know this well, but we’re often left out of the broader conversation. The public distrusts bond measures, and when utilities compromise with half-measures, those partial fixes often fail, eroding trust even further. Most people haven't felt the impact yet because infrastructure stress reveals itself slowly, in ways that don't make headlines until it’s too late. Utility rates are politically sensitive, so they lag behind actual costs by years. A single wet winter can mask a decade of overuse from reservoirs and aquifers, creating false confidence about water security. The cracks typically show up only during seasonal stress—summer heatwaves when both air conditioning and data center cooling peak simultaneously, or during droughts when every gallon counts.

Meanwhile, the infrastructure itself degrades quietly. More frequent main breaks that get fixed overnight before most people notice. Emergency repairs that prevent outages but don't address underlying capacity constraints. Power grids that hit new peak demand records every summer, operating closer to their limits with less margin for error. These are the warning signs of a system under increasing pressure, but they’re easy to dismiss as isolated incidents rather than symptoms of a deeper structural problem.

The absence of an immediate, visible crisis is precisely what makes the risk so dangerous. It builds quietly, invisibly, until a shock finally tips the system over the edge.

None of this means we should slow down data center development. On the contrary, the digital infrastructure powering AI will be absolutely crucial for the future of water management. We will rely on it for everything from optimizing the energy use of treatment processes and predicting pipe failures before they happen to modeling complex climate impacts on our water sources. The issue isn't the growth itself; it's the profound mismatch in how we plan for it.

What we're seeing with data centers and water isn't a unique crisis. It is the clearest and most urgent manifestation of a larger pattern that now defines the 21st century. We see this exact same timing gap playing out in the housing crisis^[3], in our efforts to modernize the power grid for the energy transition, and in our struggle to build resilient infrastructure for climate adaptation. In every case, private sector innovation and demand move at an exponential pace, while the civic infrastructure and regulatory frameworks that support them plod along at a linear one.

This is not just a problem of concrete and steel; it is a human capital crisis. The infrastructure gap isn't just about pipes and wires—it’s about people. We desperately need a new generation of skilled workers who can build, maintain, and protect these increasingly complex systems. Yet our workforce development, from trade school programs to university engineering curricula, operates on the same slow, linear civic timeline, unable to keep pace while the demand for these critical skills surges ahead.

Data centers are simply the most visible stress test of this broader misalignment. They are the canary in the coal mine for a systemic, society-wide challenge. The real question, therefore, isn’t “Are data centers bad for water?” It’s a much deeper and more urgent one: How do we build regulatory frameworks, planning processes, and educational systems that can finally keep pace with the reality of exponential technological change


 

1. ^{^}

   The primary direct consumptive use of water by data centers is evaporative cooling. However, power generation for these facilities creates significant consumptive use as well. There are many variables that come into play so we have a large range here. For this article I’m combining direct use and power generation.

2. ^{^}

   “With approximately 200 data centers built and 117 in the development pipeline. There has not been a single day in 14 years when a data center was not under construction in Loudoun County” Meanwhile over the last 25 years the only meaningful increase in supply has been its 11mgd reclamation plant, which add 28% total capacity.

3. ^{^}

   See a similar mismatch in housing, Housing costs too much and the supply is too low.