Pilots flying into Columbia Gorge Regional Airport on cold winter mornings are sometimes unable to land. The runway is less than a mile from Google's data center campus in The Dalles, Oregon, and on still days when temperatures drop, the cooling towers produce steam plumes that settle into the Columbia River corridor and form a fog bank thick enough to zero out visibility. Flights divert to Portland. Locals have a name for the phenomenon. They call it the "Google Cloud."
No scientific study has confirmed a causal link between the data center's evaporative cooling exhaust and the fog events. But Chuck Covert, a pilot and former manager of the airport, has spoken publicly about the pattern. The fog forms downwind of the cooling towers. It appears on cold, calm mornings when atmospheric conditions prevent the plume from dispersing vertically. And it tracks the river, which acts as a natural channel for the dense, moisture-laden air. The correlation is visible from Interstate 84 in winter. Anyone driving through the gorge can see it.
The physics here are straightforward. Evaporative cooling towers reject heat by passing warm water over fill media while drawing ambient air upward through the structure. A portion of the water evaporates. That phase change absorbs thermal energy from the remaining water, cooling it. The evaporated water exits the tower as warm, saturated air carrying significant moisture content.
When that saturated exhaust meets cold ambient air, the moisture condenses instantly, driven by the same thermodynamic process that forms natural clouds over mountains or lakes. Warm air holds more water vapor than cold air. Push warm, wet air into a cold environment and the excess moisture precipitates out as visible droplets. On a warm afternoon with a breeze, the plume rises and dissipates quickly. On a cold, still morning with a temperature inversion trapping air near the surface, the plume has nowhere to go and spreads laterally along the lowest available terrain. In The Dalles, that terrain is the Columbia River gorge, a natural channel that funnels the moisture-laden air right past the airport.
A single large cooling tower on a data center campus can evaporate tens of thousands of gallons of water per hour during peak operation. Google's The Dalles campus, which encompasses over 350,000 square feet and has received $2.4 billion in investment since it first went online in 2006, runs multiple cooling towers continuously. The cumulative moisture output is enormous. On the wrong morning, it produces weather.
Google consumed approximately 550 million gallons of water in The Dalles in a single year for evaporative cooling. That figure represents roughly 40 percent of the city's total water consumption. For context, The Dalles is a town of about 15,000 people. One company. Four out of every ten gallons.
Every large evaporative cooling deployment at hyperscale produces numbers like these. A modern data center running at 30 to 50 megawatts of IT load with a water-cooled system will consume roughly 10 to 15 million gallons of water per month, depending on climate, PUE, and cooling tower efficiency. Google's The Dalles campus, built over two decades of expansion, almost certainly draws more than 50 megawatts. The water bill scales linearly with heat rejection load, so each jump in compute capacity compounds the evaporation rate and the size of the plume that follows.
The broader water crisis facing the data center industry is well documented. The industry collectively draws billions of gallons per year from municipal water systems, aquifers, and surface water sources. What makes The Dalles different is the visibility. In most locations, the water disappears into the atmosphere invisibly. In The Dalles, the geography and climate conspire to make the exhaust collect, condense, and sit at ground level where everyone can see it. Where pilots have to fly through it.
PUE, WUE, total cost of ownership. The water-power tradeoff is a known variable. Operators choose between air-cooled systems that consume more electricity and water-cooled systems that consume more water but run at lower power. Every model, every measurement, every optimization lives within the four walls of the facility.
What happens outside the property line is a different question. Offsite water consumption effects are rarely accounted for in site selection or environmental review. The steam plume from an evaporative cooling tower is treated as a thermodynamic exhaust product, not as a local weather modification event. There is no FAA review process for cooling tower placement relative to airport approach paths. There is no EPA standard for plume dispersion modeling at data center scale. The externality exists in a regulatory gap.
Neither Google nor the FAA have publicly responded to the aviation concerns raised by local pilots. Google employs approximately 200 people at The Dalles campus. The company pays taxes and water fees. The economic relationship between Google and the city is straightforward. The atmospheric relationship is not.
The Dalles is a preview. As data center campuses scale to hundreds of megawatts across the American interior and the buildout accelerates in water-constrained regions, the externalities of evaporative cooling will become harder to ignore. Community opposition is already blocking or delaying tens of billions of dollars in data center projects across the country. Most of that opposition centers on noise, water, and power consumption. Aviation safety is a new vector, but it follows the same logic. The facility's operating envelope overlaps with something the community depends on, and nobody negotiated the boundary.
Noise from air-cooled systems has driven community pushback in multiple states. The heat island effect around large campuses is measurable. Now, steam plumes grounding aircraft. The pattern is consistent. Open-loop evaporative cooling systems externalize costs that don't appear on any operator's balance sheet but land squarely on the community next door.
Closed-loop cooling systems, dry coolers, and hybrid configurations can eliminate visible plume entirely. Zero-water cooling pilots are already running in Phoenix and Mt. Pleasant, demonstrating that high-density data center cooling can be achieved without municipal water consumption. Seawater cooling offers another path for coastal facilities. These systems cost more upfront. They typically consume more electricity per unit of heat rejected. The tradeoff is real.
The cheapest cooling system is only cheapest if you ignore what it does to everything around it. Diverted flights cost airlines money. Reduced airport utility costs a rural community economic access. Municipal water competition between a hyperscaler and 15,000 residents creates political risk that can stall future expansion.
Google built its first custom data center in The Dalles in 2006 because the Columbia River offered cheap hydroelectric power and abundant cold water. Twenty years later, the facility consumes 40 percent of the city's water and may be creating weather events that interfere with aviation. That is the trajectory of evaporative cooling at hyperscale in a confined geography. The water goes in clean and comes out as a cloud.
The cooling industry needs to price these externalities into the design phase, not discover them after a pilot reports fog on approach. Plume dispersion modeling should be standard in environmental review for any facility within five miles of an airport. Water consumption agreements should include atmospheric impact assessments. And operators choosing open-loop evaporative systems in river valleys, basins, or inversion-prone climates should be required to demonstrate that their exhaust will not degrade local air conditions.
The Dalles is a 15,000-person town along the Columbia River. Google arrived with $2.4 billion and 200 jobs. The town received a tax base and the airport received fog, and that is what an unpriced externality looks like when it meets geography, thermodynamics, and twenty years of compounding compute demand.