Johnson Controls has outlined changing thermal-management requirements for AI data centres, where increasing rack densities are challenging conventional cooling approaches. The company says rack densities for AI workloads can reach 50–150 kW and often go beyond that, increasing the role of liquid and direct-to-chip cooling technologies.
Johnson Controls argues that data centre cooling should not be viewed simply as a choice between air and liquid cooling. While high-density GPU loads can require liquid cooling, networking equipment, power supplies and other loads can continue to use air cooling. The company therefore highlights hybrid cooling strategies that allow operators to adjust the balance as computing requirements change.
The company also recommends detailed resource planning as facilities scale from 20–30 MW towards 200 MW or even 1 GW. This includes assessing power, land, labour, cooling equipment and piping requirements. Johnson Controls notes that some liquid-cooling systems can operate with supply-water temperatures of 35–45°C, compared with the 7–12°C range typically associated with traditional data centre cooling.
