Skip to main content
Live
Main content

Google defends evaporative cooling as Microsoft, OpenAI retreat from it

Seven in ten Americans now oppose new data centers, and water is the top concern. The industry is splitting on how to answer.

Jaeden Schafer
Editor in Chief · · 5 min read
Google defends evaporative cooling as Microsoft, OpenAI retreat from it

The AI industry is splitting publicly over how to cool its data centers. Google on Wednesday rolled out a set of water commitments defending evaporative cooling in regions it deems water-abundant, while Microsoft, OpenAI, and Oracle have signaled in recent months they are moving away from the technique entirely. The split lands as a Gallup poll found 7 in 10 Americans oppose new data center development, with water scarcity ranking as the top resource concern.

The stakes showed up in a securities filing Monday. SpaceX amended its IPO paperwork to warn that water scarcity, drought, and water regulation could constrain its data center plans — language that until recently would have been unusual in a tech S-1. The water question is now material enough to disclose to investors.

Evaporative cooling uses fresh water to absorb heat from server racks, which is then released through cooling towers. It saves power and reduces emissions versus pump-heavy alternatives, but the water footprint is real. Google's facility in Council Bluffs, Iowa consumed more than 1 billion gallons in 2024. Lawrence Berkeley National Laboratory projects hyperscale data centers could draw up to 33 billion gallons annually by 2030 if the industry leans heavily on evaporation.

Key facts

  • 01A Gallup poll found 7 in 10 Americans oppose new data center development, with water scarcity ranking as the top resource concern.
  • 02Google's Council Bluffs, Iowa facility consumed more than 1 billion gallons of water in 2024 using evaporative cooling.
  • 03Lawrence Berkeley National Laboratory projects hyperscale data centers could consume up to 33 billion gallons of water annually by 2030.
  • 04Microsoft, OpenAI, and Oracle have signaled a move away from evaporative cooling, including in Stargate's Texas buildout.
  • 05UC Riverside research found evaporative cooling at all US data centers during peak demand could free 10 to 30 gigawatts of power.

That total is roughly on par with — or below — agriculture and oil and gas. A single fracked well uses 1.5 to 16 million gallons. But the comparison hides the local problem: summer peaks in data center cooling overlap with summer peaks in municipal demand, and many of the regions courting AI infrastructure are already water-stressed. Texas, where OpenAI and Oracle are expanding their Stargate footprint, is the obvious case.

Microsoft, OpenAI, and Oracle have publicly committed to designs that avoid evaporative cooling on new builds. Google is taking the opposite tack. The company's Wednesday announcement included pledges to replenish more freshwater than it consumes, to scale up reclaimed and recycled water, and to disclose annual water use per data center. It also committed to a data-driven framework for matching cooling design to local watersheds.

Ben Townsend, Google's global head of infrastructure and sustainability, said the company has run detailed hydrologic assessments of its sites for the past four years. In April, Google argued to the European Union that evaporative cooling remains necessary for sustainable data centers in regions with abundant water.

There is research behind Google's position. Shaolei Ren and his team at UC Riverside found that if all US data centers adopted evaporative cooling during peak demand, the grid would free up an additional 10 to 30 gigawatts of power. In regions where grids are stressed but watersheds are not, that headroom matters to utilities trying to balance summer load.

The alternative has its own bill. Avoiding evaporative cooling means burning more electricity on mechanical chillers, which on most US grids means more emissions. Carbon footprints across the hyperscalers have already climbed sharply through the AI buildout. Cutting water can mean adding emissions, or pushing the water consumption upstream to wherever the electricity is generated.

Related · from this week
Nvidia, Google, and Microsoft push 800 VDC power standard for AI factories
Jaeden Schafer · 5 min read →

The track record on follow-through is uneven. The New York Times reported in February that Microsoft's internal records project sharply rising water use despite the company's stated shift away from evaporation. Google halted a planned data center outside Santiago, Chile in 2024 after a court partially revoked permits granted in 2020 over water concerns. The company funded a lawsuit in Oregon in 2021 to keep an Oregon town from disclosing how much water an expansion would use, and only began publishing per-site water figures in 2023. Priscilla Johnson, who served as Microsoft's director of water strategy from 2017 to 2020, said public pushback and regulation are what force better designs.

The industry has to be challenged to design smarter and simplify things.
Priscilla Johnson, Independent consultant and former Microsoft director of water strategy

Ren's own framing captures the bind: cut water in summer and the power bill — and the grid stress — goes up. There is no clean answer that works in every county, which is exactly the part of the story the marketing pledges tend to flatten.

The water question is becoming a real constraint on where AI capacity gets built, not a sustainability footnote. SpaceX's disclosure language, the Chile permit reversal, and the seven-in-ten polling number all point the same direction: site selection for the next wave of US and EU data centers will be decided as much by hydrology and local politics as by power availability and fiber. Hyperscalers that can credibly match cooling design to watershed — and disclose the numbers — will keep getting permits. The ones that can't will keep losing them, and the Stargate-scale buildouts will route around the losses.

ShareXLinkedInEmail
AI Box

Every AI model. One chat.

The latest models from ChatGPT, Claude, Gemini, Sora, ElevenLabs — 80+ models in a single chat. Compare answers side by side. Pick the best one every time.

  • ChatGPT, Claude, Gemini, Grok, DeepSeek — in one chat
  • Generate images & video with Sora, Veo, Ideogram
  • Compare any two models side by side
  • From $8.99/mo · 80+ models, all included
Try AI Boxaibox.ai
Trusted by 3,000+ teams
Got a tip?

Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.

Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.

AI Box Daily briefingFree · Daily · No fluff

Stay ahead of everyone in AI.

The tightly edited AI news email engineers, founders, and investors actually open. One email. Every weekday. Five minutes to finish.

Loved by 10,000+ AI professionals
Free forever. Unsubscribe with one click.

The briefing read inside teams at

Keep reading

More from Business

Nvidia logo
Business

Nvidia, Google, and Microsoft push 800 VDC power standard for AI factories

An OCP-backed direct-current architecture aims to unblock rack density as Wood Mackenzie projects $9 trillion in AI infrastructure spend through 2040.

Jaeden Schafer5 min read
Google logo
Business

Google lifts AI capex to $205B, rattling Wall Street on data-center returns

A $15B jump from last quarter's projection lands as Nvidia, Oracle, and OpenAI deals stretch the AI build-out's financing further.

Jaeden Schafer4 min read
Amazon says its data centers used 2.5 billion gallons of water in 2025
Business

Amazon says its data centers used 2.5 billion gallons of water in 2025

Amazon's first public water disclosure pegs usage at 0.12 liters per kilowatt-hour, down 2% from 2024 even as operations expanded.

Jaeden Schafer4 min read