Why water scarcity is the new choke point for AI data centers
Subscribe to our free newsletter today to keep up to date with the latest renewable energy news.
Data centers are often cast as high-energy infrastructure, with their electricity demand well documented and frequently criticized. But as the AI boom accelerates and hardware intensity increases, another resource has emerged as a critical constraint: water.
Cooling systems, particularly evaporative cooling, are central to data center operations. They dissipate the heat generated by thousands of GPUs and CPUs working at full capacity, especially in AI training clusters. While the energy side of this equation has drawn public attention, the water required to run these systems has quietly scaled alongside.
A mid-sized data center can consume about 300,000 gallons (roughly 1.4 million liters) of water per day. That’s enough to supply a small town. For hyperscalers deploying AI infrastructure at scale, the aggregate demand becomes significant. The OECD estimates that AI-specific data center demand could reach 4.2 to 6.6 billion cubic meters annually by 2027, nearly half the UK’s total annual water use.
Unlike energy, water has no substitute. A power outage can be mitigated with generators. When water becomes scarce or restricted, cooling infrastructure is compromised, risking throttled workloads, overheating, or shutdown. The problem is most acute in parts of Europe, where climate volatility, infrastructure growth, and tightening environmental policy collide.
A perfect storm: Europe’s climate, AI surge and tightening rules
Europe’s climate is becoming less reliable and more extreme. The summer of 2025 brought another wave of droughts across Southern and Central Europe, worsening aquifer depletion and reducing surface water availability. In parallel, data infrastructure has been expanding rapidly, much of it to support large AI models and storage-heavy cloud platforms.
The water burden of this growth is only beginning to be quantified. Market estimates show Europe’s data center water consumption could double between 2025 and 2030, from 0.82 trillion to 1.58 trillion liters. Heatwaves increase ambient temperatures, which in turn raise cooling demand, creating a feedback loop.
For regulators, water scarcity adds pressure to climate and sustainability frameworks. The EU-backed Climate Neutral Data Centre Pact mandates that newly built data centers in water-stressed areas meet a Water Usage Effectiveness (WUE) target of 0.4 liters per kilowatt-hour starting in 2025. Older facilities replacing cooling infrastructure must meet the same threshold by 2040.
Industry groups like CISPE warn that such constraints could redirect investment away from the EU toward less regulated regions. Builders and operators are now forced to consider water availability alongside power and connectivity.
The technologies racing to close the tap on water waste
In response, data center operators are adopting alternatives to traditional water-intensive systems. Closed-loop liquid cooling systems, where coolant circulates in sealed circuits without evaporation, are gaining traction. These systems significantly reduce water withdrawal while providing consistent thermal performance.
Microsoft has led efforts to redesign water use. By late 2024, it began implementing data center designs that eliminated evaporative cooling entirely. Each of these sites is projected to save more than 125 million liters of water annually. Across its global footprint, the company reduced its average WUE from about 0.49 liters per kilowatt-hour in 2021 to 0.30 in 2024.
Other strategies include using reclaimed water, rainwater, or greywater for non-potable cooling. In cooler regions, operators are designing systems that rely on ambient air or draw from nearby natural cooling sources, such as lakes or rivers. Environmental safeguards are required to prevent thermal pollution and ecosystem disruption.
Some facilities are also raising server inlet temperatures to reduce cooling needs. Others are using machine learning to manage cooling cycles more efficiently.
Still, adoption varies. Larger companies can invest in these solutions. Smaller operators often rely on older infrastructure and may lack the resources to transition away from evaporative cooling. In many cases, facilities still do not measure or report water use, limiting visibility.
The growing awareness of water risk is reshaping operational decisions and site selection strategies. Where location choices once prioritized fiber access, tax incentives, and grid capacity, water availability is becoming a core concern.
Yet many operators do not publicly disclose water usage. Without reliable metrics, it is difficult for regulators, investors, or local communities to understand the impact. WUE remains a voluntary benchmark in many regions, although that is changing.
Community opposition is rising. In the Netherlands, local resistance has delayed or canceled proposed data center projects. Similar debates are unfolding in Spain and parts of Germany, especially where drought conditions persist.
Long term, water scarcity is not just a compliance issue. It is a limit on how and where data infrastructure can grow. AI workloads will continue to scale. But if cooling systems cannot evolve in tandem, the growth of the digital economy could be slowed by one of its least visible resource dependencies. The future of cloud computing may hinge not just on processing power, but on sustainable access to the water that keeps it all running.