AI datacenters may gulp a New York City’s worth of water on hot days
Summary
Researchers at the University of California, Riverside warn that growth in AI datacentres could create very large peak water withdrawals during hot days, potentially requiring an extra 697 to 1,451 million gallons per day (MGD) of capacity by 2030 — roughly comparable to New York City’s daily supply.
The study explains datacentre cooling works in two stages: server-level cooling (usually closed-loop and not directly consuming water) and facility-level cooling (which can rely on evaporative water use, such as cooling towers or adiabatic systems). Evaporation-based cooling can sharply increase water demand on the hottest days, creating peak strains on public water systems that many communities may not be able to meet without major upgrades.
Key Points
- UC Riverside estimates US datacentres could need an additional 697–1,451 MGD of peak water capacity by 2030, at an estimated cost up to $58 billion.
- A 100 MW IT load may require about 0.5–2.5 MGD depending on climate and cooling design; gigawatt-scale AI campuses would demand far more.
- Public water systems are numerous and mostly small: ~50,000 community water systems exist, ~40,000 serve ≤3,300 people, and only ~708 are large systems serving >100,000 people.
- Peak withdrawals matter more than annual consumption — water taken in during peaks is unavailable to other users even if returned later.
- Many datacentre projects have already required local water infrastructure upgrades, sometimes for peak demands as low as 0.1 MGD.
- Report recommendations: operators should report peak water use (not just annual averages), partner with communities on funding upgrades, and coordinate with utilities to switch cooling modes when either water or grid capacity is strained.
Context and relevance
This study links the rapid expansion of AI-focused datacentres to a growing infrastructure challenge: meeting short-duration, extremely high water demand during heatwaves. It sits at the intersection of climate risk, urban utilities planning and tech infrastructure rollout. The finding is particularly relevant to planners, datacentre developers, utility managers and policy-makers who must balance energy, water and community resilience as AI deployments scale.
The report highlights a practical planning gap: many communities’ water systems are sized for historical peaks and may not be resilient to large new industrial peak demands. If datacentre water use is not managed or coordinated with local utilities, regions could face costly upgrades or local shortages during extreme heat events.
Why should I read this?
Because if you care about summers that don’t turn into surprise water rationing for residents or surprise bills for councils, this one matters. It explains why AI growth isn’t just an electricity problem — on hot days it could be a water one too — and what operators and communities should actually do about it.
Author note (style)
Punchy: This is a clear red flag for infrastructure planners and anyone tracking the real-world costs of scaling AI. Read the detail if you manage utilities, build datacentres, or influence planning policy — the stakes (and bills) are substantial.