Local opposition to data centers is one of the main threats to tech giants’ artificial intelligence ambitions, as residents protest rising energy bills and the potential strain on local water resources — and question whether AI is even good for society, or worth the high cost.
The mistrust is compounded by the fact that hyperscalers often hide behind shell companies with code names to negotiate permits and development agreements with local officials under NDAs — a practice they defend as a way to prevent inflated land prices.
That secrecy extends to data centers’ water consumption, which most hyperscalers don’t disclose for each location they operate in. That information gap is particularly concerning for projects proposed in water-stressed regions.
Google is trying to undo that impression. This week, the company publicized a handful of responsible stewardship practices. The practices aren’t new — Google had already announced them over the last few years. But at a time when data center pushback is reaching a fever pitch, Ben Townsend, the company’s head of infrastructure and sustainability, said he hopes they can serve as a blueprint for other hyperscalers and potentially help communities evaluate new projects in their towns.
“As we sat down with community members and policymakers and walked through the way that we were approaching water, a lot of them told us, ‘You’re checking all the boxes of things we’re concerned about, but is everyone else doing this?’ And we said absolutely not,” Townsend told Latitude Media in an interview. He added that it makes sense why communities are alarmed if data centers’ water consumption is a “black box.”
In late 2022, Google became the only major tech company to publish site-specific water consumption for its data centers. Its other stewardship practices include setting a goal to replenish more water than it consumes at its locations by 2030; investing in upgrades to public water and wastewater infrastructure; assessing local watersheds before building a new data center; and opting for air cooling or recycled water in regions at risk of water scarcity.
For years, however, Google was just as secretive about its water consumption as its peers. The shift followed The Oregonian/OregonLive’s lawsuit against the city of The Dalles, Oregon, seeking documents showing how much water Google consumed to cool its data centers in the region. The parties ultimately settled the case, with the city and Google agreeing to disclose annual water use.
The resulting documents reveal that Google quintupled its water consumption in The Dalles between 2012 and 2025, to about 550 million gallons, or nearly 40% of the city’s total water consumption last year.
In its latest environmental report, Google lists the water consumption of data centers across 36 cities in 2024, with the majority — 23 — located in the U.S. On a global scale, all of its data centers combined consumed 7.7 billion gallons of water.
But some communities still feel left in the dark about the potential impact of Google’s data centers on local water resources. Earlier this year, Virginia’s Roanoke Rambler sued the local water authority for records showing how much water a planned Google data center campus in Botetourt County would consume. Those records showed that officials agreed to provide the project with up to 2 million gallons of water daily by January 2028, with the potential to rise to 8 million gallons daily, depending on future expansion.
Townsend said Google’s development agreements are treated as proprietary because they include “infrastructure design specifications” that do not directly align with water usage. They’re designed to determine the maximum volume of water a facility could draw during peak-demand events, such as a fire or cooling during the hottest day of the year. He called it the “peak capacity fallacy.”
“It is inaccurate to interpret peak infrastructure capacity as a baseline for annual usage,” he said. “Assuming a facility consumes its peak infrastructure capacity 24/7 is equivalent to assuming a car travels at its maximum speedometer speed every second it is on the road.”
How data centers use water
Data centers primarily use water to cool racks of densely packed servers that generate a lot of heat. The most popular system has been evaporative cooling, which uses less energy than traditional air conditioning — and therefore puts less strain on the grid and can cause fewer carbon emissions — but laps up more water.
“We started doing detailed hydrological assessments at every single one of the campuses that we’re considering before a design is selected,” Townsend said. “If we found that the watershed is in stress, we didn’t use evaporative cooling.”
The AI boom is shifting things, however. The latest AI chips have much higher power densities and can be packed more tightly together, so traditional cooling technologies often aren’t sufficient to keep temperatures down. That has led developers to pursue direct-to-chip, closed-loop cooling, which uses less water than evaporative cooling but more than traditional air cooling. Cold plates are placed directly on hot chips, and a liquid coolant recirculates through sealed pipes.
Townsend said Google has used direct-to-chip approaches for more than a decade, but most of its data centers use evaporative or air cooling.
Meta, Amazon, and Microsoft have all said they are shifting toward closed-loop for AI factories, though it’s unclear how widespread it is in practice. The companies don’t disclose how much water they consume at the site level, but in 2023, Meta reported that its data centers globally consumed 776 million gallons of water, accounting for 95% of the company’s total. Each of those tech giants has promised to eventually become “water positive” across their operations — meaning they replenish more water than they consume — and invest in public infrastructure.
Even without clear numbers, it’s apparent that water use across the industry is surging. In 2023, before the AI boom was fully underway, the Lawrence Berkeley National Laboratory estimated that U.S. data centers consumed 17 billion gallons of water directly through cooling, and projected that by 2028, that amount could quadruple to nearly 33 billion gallons.
By comparison, that would be about 1% of the water that Americans use on their lawns each year. But data centers also indirectly consume water because the power plants they rely on for electricity have their own cooling needs.
Researchers estimated an indirect water footprint of 200 billion gallons in 2023, with the caveat that they were constrained “due to the unavailability of facility-level data.” These estimates also didn’t account for on-site, behind-the-meter generation, which could “significantly affect water consumption” estimates depending on the electricity source, the report said.
