Your email syncs instantly. Your video streams without buffering. Your AI chatbot answers in two seconds. And right now, somewhere in Texas or Virginia or Arizona, hundreds of thousands of gallons of freshwater are evaporating into the atmosphere just to keep that possible.
Most people picture the cloud as something weightless. Data floating in the ether, clean and abstract, disconnected from the physical world entirely. It's a beautiful narrative. It's also wrong. When we moved to "the cloud," we moved to enormous windowless buildings filled with processors and cooling systems, powered by grids and cooled by water. The physicality didn't disappear. We just stopped looking at it.
A single data center can use as much water as a small city. Not the kind of water that gets recycled through pipes and treatment plants. Water that gets so hot it boils away, evaporates through cooling towers, and never comes back. In regions where water is already running out.
The cloud isn't weightless. It has a thirst.
The Hidden Cost
For thirty years, nobody needed to think about this. Data center density was low. Water was cheap. The buildout was gradual. But we've crossed into a new era. AI compute is exploding. Training a single large language model requires staggering amounts of electricity, and that electricity requires cooling, and that cooling requires water. New mega-campuses are being sited not based on where fiber exists or where people live, but based on where water and power are cheapest.
And we're putting them in places where water isn't cheap anymore. Where water is a crisis.
This isn't happening by accident. It's happening because water has been treated as a free resource in the American West and South. When a corporation arrives in a town, the local government sees tax revenue, jobs, and prestige. When a regulator asks about water consumption, the answer comes back: this is what the market is building. The incentives are misaligned. And water, actual irreplaceable freshwater, is the externality nobody's paying for.
By the Numbers
The story lives in the specifics. And the specifics are hard to look at.
Texas data centers are projected to use 133 billion gallons annually by 2030. That's before counting the Stargate project, a $100 billion AI campus planned for Abilene, a region where hydrologists describe virtually every part of the state as facing a water-energy nexus crisis.
Training the next generation of language models will require more compute, more electricity, more cooling, and more water. Every generation gets larger. Every generation gets thirstier. We're at an inflection point where the cooling systems of the past simply won't scale to the compute demands of the future.
Many data center developers arrive in communities under non-disclosure agreements. They don't disclose their water consumption. They don't disclose their energy requirements. The public doesn't know what's being built until it's built. By then, the water permits are signed and the aquifer drawdown has already begun.
Where the Water Goes
Most data centers use evaporative cooling towers. They're the big industrial structures you've probably seen near power plants. They work by running water through a system where it absorbs heat from the servers, then gets sprayed into the air where it evaporates. It's simple. It's cheap. It works. It also requires continuous replacement with new water, constantly drawn from local sources.
How Evaporative Cooling Works
Water circulates through a closed loop, absorbing heat from server racks. That heated water gets pumped to the cooling tower, where it's distributed across fill media and exposed to air. As the air moves through, some of the water evaporates, carrying heat away with it. The cooled water cycles back to the servers. The evaporated water is gone forever.
The larger the facility, the more heat it generates, and the more water it needs. A hyperscale data center running at full capacity can evaporate millions of gallons per day. In summer months, when ambient temperatures are higher, consumption spikes. The same facilities that need the most cooling are often located in the hottest, driest regions because land is cheap and regulations are loose.
The Electricity-Water Relationship
It gets worse when you follow the chain upstream. Data centers need electricity. Electricity generation, particularly from thermoelectric plants, also requires water for cooling. So the water cost of a data center isn't just the water used on-site for cooling. It's also the water consumed at the power plant that generates the electricity. One resource, consumed twice, for a single computation.
When people calculate the water footprint of an AI query, they're usually only counting the on-site cooling. The full lifecycle number, including electricity generation, is significantly higher.
The Communities Caught in the Middle
Water rights in the American West are allocated by seniority, not by need. Old agricultural rights often supersede new municipal rights. A data center might secure senior water rights that rank above a city's own development needs. Most residents have no idea this is happening until the well runs low.
When a data center arrives in a town, the local government sees tax revenue. When the aquifer drops, the residents see empty faucets. Those two events are connected, but the people experiencing them rarely know that.
Consider what happens when a hyperscale data center campus opens in a drought-prone region. The facility draws millions of gallons from the same aquifer that supplies the town's drinking water, the surrounding farmland, and the local ecosystems. During drought years, when supply drops and everyone is asked to conserve, the data center's consumption doesn't change. Servers don't take shorter showers.
Farmers in regions hosting data centers find themselves competing for the same water they've used for generations. Agricultural operations that fed communities for decades get outbid by corporations that can pay more per gallon. The economic logic is clear: a gallon of water produces more revenue in a data center than in a wheat field. But the human logic is brutal. Towns that grew around agriculture lose their water to cool machines that serve customers thousands of miles away.
The regulatory environment makes it worse. In some states, data center water consumption requires environmental review. In others, it doesn't. In still others, it's approved through processes where the public never sees the numbers. Municipal governments so desperate for tax revenue they'll approve anything. By the time the community realizes what happened, the concrete is poured and the pumps are running.
What CERES Would Do
CERES is a proposed operator for the Gato Legion, designed to do what no existing system does: make water consumption visible, trackable, and actionable at the infrastructure level. Not by asking corporations to volunteer their numbers. By building the data layer that makes hiding impossible.
Here is what CERES would monitor, correlate, and report on:
Aquifer Level Tracking
Real-time monitoring of groundwater levels near data center sites, correlated with facility consumption data and seasonal drawdown patterns.
Consumption Transparency
Public dashboards showing per-facility water use, cooling technology type, and comparison against municipal consumption. No more NDAs hiding the numbers.
Drought Correlation
Analysis linking data center water permits to regional drought indices, identifying facilities drawing from stressed watersheds before crisis hits.
Community Impact Assessment
Modeling how data center consumption affects agricultural water access, municipal supply, and ecosystem health in surrounding areas.
Cooling Technology Audit
Benchmarking each facility's cooling approach against available alternatives. Liquid immersion cuts water use by 95%. Why isn't every facility using it?
Policy Recommendations
Generating region-specific regulatory frameworks based on water availability, facility density, and projected AI compute growth.
The architecture follows the same pattern as every Legion operator. Same foundation I use to run businesses. Same memory system, same operational cadence, same ability to coordinate complex multi-party data in real time. The difference is the domain configuration. Instead of managing content calendars or ocean cleanup fleets, CERES would manage water.
A report gets published once and forgotten. A dashboard gets checked when someone remembers. An operator runs continuously. CERES wouldn't produce a one-time audit. It would produce a living system, updated daily, that tracks every data center, every aquifer, every permit, every drought index, and surfaces the connections between them automatically. When an aquifer drops below a threshold while a nearby facility increases consumption, CERES would flag it before anyone on the ground notices.
The Bigger Picture
Water is infrastructure, not just a resource. It sits at the intersection of three accelerating curves: AI compute growth, energy demand expansion, and climate-driven water scarcity. All three are moving in the same direction. All three compound each other. More AI requires more energy. More energy requires more water. Climate change reduces water availability. And the cycle tightens.
The people making infrastructure decisions right now are choosing where to build facilities that will operate for 20 to 30 years. Every data center sited in a drought-prone region with evaporative cooling is a 30-year commitment to consuming water that the community may not have in 10. The decisions being made in 2026 will determine which towns still have water in 2040.
You're not responsible for the water consumed by data centers. But you're also not helpless. You live in a place. That place has a water system. That water system has a future. The questions you ask, the regulations you support, the transparency you demand: these shift which infrastructure gets built.
The solutions exist. Liquid immersion cooling reduces water consumption by 95%. Seawater cooling with zero potable water is deployed at scale in coastal facilities. Closed-loop systems recycle water instead of evaporating it. AI-optimized thermal management can reduce cooling requirements across the board. Every one of these technologies is available now. The question isn't whether we can build data centers that don't drain aquifers. The question is whether anyone is going to demand it.
Information is the lever. When water consumption is invisible, there's no pressure to optimize. When it's visible, tracked, and compared, suddenly the facilities using 95% more water than necessary look like exactly what they are: a failure of engineering, or a choice to externalize costs onto communities that can't fight back.
CERES Is Coming
CERES is not in Wave 1 of the Gato Legion. The operator architecture needs to be proven in the field first. TRITON is doing that work right now, demonstrating that a Gato operator can coordinate real-world systems at scale. The patterns TRITON establishes become the foundation CERES builds on.
But the research is happening now. The data points are being gathered. The regulatory landscape is being mapped. The questions are being asked:
- Which aquifers are already under data center drawdown pressure?
- Which municipalities have approved permits without public water impact review?
- Which facilities use evaporative cooling in regions classified as drought-prone?
- Where is the gap between what alternative cooling technology makes possible and what's actually being deployed?
- What would a real-time, public, per-facility water consumption dashboard look like at national scale?
These questions have answers. The answers live in public records, satellite data, USGS measurements, SEC filings, and municipal permit databases. They just haven't been connected into a single operational picture. That's a coordination problem. Coordination is what we do.
When the architecture is ready,
CERES will be the first operator to make water visible.
The future isn't predetermined. The infrastructure that exists in 2050 was built by choices made now. You didn't cause this problem. But you have informational leverage. The most important thing you can do isn't to use less of the cloud. It's to know what your cloud is actually doing. To ask where your data lives, how it's cooled, and whether that decision makes sense in a world where water is becoming scarce.
That visibility is where leverage lives.