Schneider says hotter coolant can make AI datacenters less thirsty

A datacenter's cooling architecture can have a decisive effect on its water and energy use, according to Schneider Electric, which argues that operators should evaluate it early in the design process. In a white paper published today, the energy management specialist compares four modeled 100 MW datacenter designs, ranging from traditional air cooling to optimized liquid-cooling systems. The paper, "Water Usage at AI Scale: Insights from a 100 MW Comparative Analysis," concludes that moving from air to liquid cooling consistently improves energy efficiency and can reduce water consumption in constrained locations. Schneider has a commercial interest in that conclusion. It acquired a 75 percent controlling stake in liquid-cooling specialist Motivair in 2024 and said it intended to buy the remainder by 2028. The assessment also models the effect of location by applying the four designs to the annual weather profiles of Paris, France, and Dallas, Texas. Schneider's model puts the highest onsite water consumption at a traditionally air-cooled facility in Dallas. Switching to liquid cooling with coolant supplied at 45°C (113°F) reduces water use by at least 50 percent, it claims. The firm says the results demonstrate how location, operating temperature, and cooling architecture determine a facility's water requirements, even as rack densities and the demands of power-hungry AI systems increase. Water consumption is determined largely by the equipment used to reject heat outside the data hall, the report says. For a facility of equivalent size, Schneider claims cooling towers can consume five to 20 times as much water as dry coolers. The first scenario uses air cooling. The second uses liquid cooling supplied at 32°C (about 90°F), which Schneider says is representative of early AI datacenters. The third retains the same cooling equipment but raises the supply set point to 45°C (113°F). The fourth design reduces the equipment count by sizing the cooling system specifically for operation at 45°C. Schneider says this avoids excess capacity and lowers capital expenditure. Higher coolant temperatures produce "substantial additional water savings," the report says, because they extend the range of conditions in which cooler outside air can reject heat without mechanical chilling. The paper recommends assessing water consumption early in the design process and considering alternative sources to reduce demand on supplies used by local communities. The issue is particularly contentious in the US, where datacenters' water consumption has tripled in a decade and local opposition to new projects is growing. In the UK, meanwhile, the government has been accused of overlooking water demand while promoting the construction of more AI server farms. Schneider cautions that water and energy efficiency must be evaluated together. The best design will depend on the relative importance of power usage effectiveness (PUE) and water consumption, as well as local energy and water costs. ®