• Trane and LiquidStack have launched a new 2.5-megawatt coolant distribution unit aimed at liquid-cooling dense AI server racks.
  • The unit pairs Trane’s building-cooling engineering with LiquidStack’s liquid-cooling hardware in a single, higher-capacity CDU.
  • Larger CDUs let operators add high-density, liquid-cooled rows without multiplying smaller units, turning CDU sizing into a core capacity-planning decision.

The joint coolant distribution unit is designed to support higher-density, liquid-cooled server racks as AI compute scales up.

Trane and LiquidStack have launched a new 2.5-megawatt coolant distribution unit, or CDU, according to a Data Center Dynamics report, giving data center operators a higher-capacity option for delivering liquid coolant to dense AI compute racks.

A CDU sits between a facility’s chilled-water loop and the server racks it serves, regulating the temperature, pressure and flow of coolant delivered to direct-to-chip or immersion cooling systems. Pairing Trane’s building-cooling engineering with LiquidStack’s liquid-cooling hardware points toward larger, pre-engineered cooling blocks as AI training and inference clusters push rack power densities well beyond what air cooling can handle. It is also part of a broader pattern across the cooling industry, in which established HVAC manufacturers are teaming with liquid-cooling specialists rather than building direct-to-chip expertise from scratch, betting that AI data center operators will want a single, warrantied cooling block rather than a stack of components sourced separately.

What it means for operators

As AI rack densities climb past 100kW and higher, the coolant distribution unit becomes less a commodity component and more a capacity-planning constraint of its own. A 2.5MW-class CDU gives operators more headroom to add high-density, liquid-cooled rows without a matching proliferation of smaller units, but it also concentrates more cooling capacity, and more risk, behind a single piece of hardware. Facilities moving toward direct-to-chip or immersion cooling for AI workloads should treat CDU sizing and redundancy as a first-order design decision, not an afterthought bolted onto a cooling retrofit, and should factor unit-level failure modes into their broader thermal risk planning before committing to one vendor’s cooling block at scale.

Source: Data Center Dynamics