Schneider’s Bold Statement on AI Datacenter Cooling Needs
The cooling configuration of a datacenter plays a crucial role in its consumption of water and energy, according to Schneider Electric, who is urging stakeholders to consider this aspect even before finalizing the design plans. In a white paper released today, they have analyzed four 100 MW datacenter designs, which range from standard air cooling to advanced liquid-cooling systems. Their publication, “Water Usage at AI Scale: Insights from a 100 MW Comparative Analysis,” suggests that replacing air with liquid cooling enhances energy efficiency and reduces water consumption in areas facing resource constraints. However, they’re not merely acting out of benevolence; they have a vested interest in liquid cooling, having acquired a 75 percent stake in Motivair in 2024 and aiming for full ownership by 2028.
Paris vs. Dallas: A Climate Clash
Schneider has been experimenting with the impacts of different climates by applying their four designs to the weather conditions in Paris and Dallas. The findings indicate that the typical air-cooled setup in Dallas consumes the most water. Transitioning to liquid cooling with coolant set at a warm 45°C (113°F) reportedly reduces water consumption by over 50 percent. This exercise emphasizes how location, temperature adjustments, and cooling methods affect water usage, even as AI systems continue to demand more resources.
Water Consumption of Cooling Equipment
Your water usage primarily relies on the method used to expel heat from the data center, they argue. For an equally sized setup, Schneider claims that cooling towers can consume five to twenty times more water than dry coolers. The first scenario represents standard air cooling. The second utilizes liquid cooling at a temperate 32°C (90°F)—a nod to the early days of AI datacenters, according to their assertions. The third raises the temperature to 45°C (113°F) with identical equipment. The fourth scenario optimizes the equipment for 45°C operation, ostensibly eliminating unnecessary components and reducing capital expenses.
Increased Temperatures, Greater Savings
Using hotter coolant reportedly conserves substantial amounts of water, indicates the report, as it allows for more circumstances where cooler outdoor air can dissipate heat without mechanical intervention. The paper stresses the importance of addressing water consumption early in the design phase and exploring alternative sources to alleviate the impact on local water supplies. This is a significant issue in the US, where datacenter water usage has surged, causing considerable concern among local populations. In the UK, residents are frustrated that the government is constructing more AI server facilities without considering water requirements.
Energy, Water, and the Big Picture
Schneider is waving a cautionary flag, emphasizing that one cannot solely focus on energy or water use but should evaluate both simultaneously. The optimal configuration depends on how power usage effectiveness (PUE) is balanced against water consumption, as well as current local energy and water prices.
Oops, Wrong Faucet!
So there you have it, Gadgets and Gizmos: if you’re neglecting to investigate your cooling choices early on, you could end up with a datacenter that’s thirstier than party-goers on a Friday night.
