Summary
- CDU 2.X provides coolant flow of up to 3,750lpm at 3.5 bar.
- The platform supports end-of-row or rack-adjacent configurations, dual feeds, and different redundancy arrangements.
- Pre-orders are open, with first shipments scheduled for the second quarter of 2027.
LiquidStack has launched a coolant distribution unit capable of delivering up to 3,750 litres per minute at 3.5 bar as data centre operators prepare liquid-cooling infrastructure for another increase in AI rack densities.
CDU 2.X, announced by Trane Technologies, which acquired LiquidStack in March 2026, is intended to provide a common coolant-distribution architecture across current and future accelerator platforms instead of requiring a new CDU layout whenever server hardware changes.
The system can be deployed at the end of a row or adjacent to racks and supports configurable control valves, power feeds, and redundancy arrangements. Dual A/B power feeds and automatic transfer-switch options are available where operators require higher electrical resilience.
Trane is taking pre-orders now, with shipments due to begin in the second quarter of 2027. The company says the system is designed to support emerging GPU systems including Nvidia Vera Rubin and later platforms.
Cooling design moves upstream
Increasing CDU flow is not simply a component specification. As direct-to-chip liquid cooling handles a greater proportion of server heat, the distribution system between the facility water loop and IT equipment becomes part of the critical path for capacity.
A CDU must maintain flow, pressure, water quality, heat exchange, controls, and redundancy while fitting within a data hall that may have been designed around much lower-density equipment. Large units can also introduce additional electrical requirements and harmonic loads of their own.
LiquidStack says CDU 2.X uses an ultra-low-harmonics variable-frequency-drive architecture intended to address distortion at source and reduce the amount of additional power-quality equipment needed around the system. That claim will ultimately be tested in deployed installations, but it addresses a real design problem as mechanical equipment scales alongside IT power.
The platform also accepts facility inlet temperatures of up to 45°C. Higher water temperatures can improve the economics of heat rejection because chillers may operate less frequently, or dry coolers can reject heat more efficiently under suitable outdoor conditions.
The achievable result depends heavily on climate, server requirements, approach temperatures, and the design of the wider cooling loop. A CDU capable of accepting warmer facility water does not by itself guarantee low cooling energy consumption across an entire site.
Trane has positioned configurability as the principal benefit. The aim is to avoid building a facility around one fixed CDU arrangement when GPU platforms, manifold layouts, rack densities, and coolant requirements may change repeatedly over a data centre’s operating life.
That becomes particularly relevant for retrofit projects. Traditional halls often have limited floor space, fixed electrical distribution, and mechanical systems that were not designed to place high-capacity liquid-cooling hardware beside IT rows. Being able to locate CDUs either at row ends or closer to racks gives designers more options, although pipework routes and maintenance access remain site-specific constraints.
The 3,750lpm figure is substantial, but flow alone is an incomplete measure of cooling capacity. Delivered thermal performance depends on the coolant temperature difference, fluid properties, pressure, system resistance, and operating conditions across the primary and secondary loops.
Those details will matter when shipments start next year. For operators planning AI capacity today, however, the launch is another indication that liquid cooling is moving from a rack-level engineering choice into the facility’s long-term mechanical and electrical architecture.
