The AI energy debate is loud. The AI water debate is quiet — and arguably more constraining in the places data centres actually get built. In early July 2026, Vancouver startup Wafr Technologies raised $100 million (of a targeted $300 million) to commercialise a water-efficient cooling system for AI data centres, claiming it can cut water use by up to 95% and cooling power by up to 80% versus conventional evaporative systems. When investors write nine-figure cheques for cooling technology, it's a signal that heat and water have become first-order infrastructure problems.
Why AI data centres are so thirsty
Densely packed AI servers generate enormous heat, and that heat has to go somewhere. Many large facilities use evaporative cooling — essentially water evaporating to carry heat away — which is energy-efficient but water-hungry. According to a 2025 International Energy Agency analysis, a single 100-megawatt AI data centre in the US can consume roughly 2 million litres of water per day — about the daily usage of 6,500 households — and cooling can account for 30–45% of a facility's total electricity load. As AI drives ever-denser, hotter server configurations, both the heat density and the water demand climb together.
The friction is geographic. Data centres want to be near cheap power, good connectivity, and their users — but those locations are often already water-stressed, putting hyperscalers in direct competition with communities and agriculture for a shared, finite resource. That's a regulatory and reputational problem as much as a technical one.
The technologies racing to fix it
The cooling industry is innovating fast, along several tracks:
- Water-efficient chilling, like Wafr's, aimed at slashing consumption while managing rising heat density.
- Liquid cooling — piping coolant directly to the chips rather than cooling the whole room, far more effective for dense AI hardware and increasingly the default for new builds.
- Immersion cooling — submerging servers in non-conductive fluid, the most aggressive approach for the highest heat densities.
- Efficient heat-transfer units — newer air-conditioning designs delivering meaningfully better efficiency for high-heat equipment rooms.
The direction of travel is clear: from cooling air in a room to cooling the silicon directly, and from consuming water to recirculating or avoiding it.
Why it reaches your cloud bill and strategy
This isn't only an operator concern. Cooling is a major slice of a data centre's operating cost, so cooling efficiency feeds into the price of the cloud you rent. Water availability increasingly dictates where new capacity can be built, shaping which regions have headroom. And sustainability reporting is tightening — for many organisations, the water and energy footprint of their cloud usage is becoming something they have to measure and disclose, which makes a provider's efficiency a procurement criterion, not a footnote.
The takeaway
Power grabs the headlines, but water and heat are the quieter limits on AI's physical expansion — and they're already steering where data centres can be built and how much they cost to run. Expect cooling efficiency and water stewardship to become standard factors in how cloud capacity is priced, sited, and chosen.
Where Educatifu fits
We help companies design cloud architecture that's efficient by default — lower footprint, lower cost, and resilient to the capacity and sustainability pressures reshaping the market. If you want infrastructure built for where things are heading, get in touch.