The Closed-Loop Illusion: Water, Energy, and the Hidden Costs of AI Infrastructure

As AI has driven a historic data center buildout, water has become one of the industry's biggest constraints, consumed by the millions of gallons for on-site cooling. Hyperscalers' answer has been a new generation of closed-loop cooled data centers. On the surface, it looks like the problem is solved. But these same AI data centers are also extraordinarily power-intensive, packed with compute-dense hardware that demands enormous electricity just to run. Generating that electricity, increasingly from the natural gas and nuclear plants hyperscalers are racing to bring online, consumes water of its own. It is projected that AI's total water demand will grow 129% by 2050, with 54% of that growth coming from power generation alone, against only 4% from the on-site cooling the industry is currently celebrating. Water hasn't left the AI story, it's simply moved from the cooling tower to the power plant. This panel exists to trace that shift and hold hyperscalers accountable for the water footprint their closed-loop cooling claims leave out.

This session is for:

• ICCR members engaging hyperscalers on environmental disclosures who want to move beyond on-site water metrics to total footprint accountability

• Institutional investors in data center REITs, hyperscalers, and utilities whose operations intersect with AI power demand

• Shareholder advocates developing water-related proxy resolutions or preparing co-filer coalitions ahead of the 2027 proxy season

• Members with portfolio exposure to nuclear, natural gas infrastructure, or water utilities who need to understand how AI’s race for power is reshaping watershed demand

• Anyone who attended the Spring 2025 “Deep Dive into AI’s Thirsty Growth” panel and wants to understand what has and has not changed since

1. Closed-loop cooling has become the industry's standard answer to data center water concerns. This panel names the blind spot in that answer the closed-loop illusion.