Summary
- Clivet says its high-capacity CDU can deliver up to 3MW at a 3K approach temperature or 5MW at 5K.
- The portfolio combines cooling sources, CDUs, fan walls, terminal equipment, and magnetic-levitation compressor technology.
- The specification illustrates how AI rack densities are pushing thermal infrastructure towards multi-megawatt liquid-cooling systems.
Clivet has introduced a data centre cooling portfolio including a coolant distribution unit rated at up to 5MW, as vendors scale thermal infrastructure for increasingly dense AI computing deployments.
The Italian HVAC manufacturer presented the equipment at the AI Infra Summit in Santa Clara, alongside a new magnetic-levitation compressor architecture and a wider range spanning cooling sources, liquid distribution, fan walls, and terminal equipment.
Clivet says its largest CDU can handle up to 3MW of thermal load at a 3K approach temperature difference or 5MW at 5K. Those figures are manufacturer specifications rather than independently benchmarked performance data, but they illustrate the scale at which liquid-cooling distribution is now being designed.
A multi-megawatt CDU is a different engineering proposition from the smaller distribution units associated with individual rows or limited liquid-cooled clusters. At several megawatts, the unit becomes a major part of the mechanical plant and has to be considered alongside pumping, heat rejection, redundancy, pipework, controls, maintenance access, and the consequences of taking equipment offline.
Clivet says the unit includes predictive temperature control intended to hold fluctuations within ±0.5°C and allows key components to be replaced while the equipment remains online. The company is pitching those features at facilities where loss of liquid distribution could affect a large amount of computing capacity simultaneously.
The portfolio also includes a magnetic CDU that combines cooling generation with secondary liquid distribution and an air-cooled magnetic-levitation chiller intended for the facility side of the thermal system.
At the compressor level, Clivet has introduced a dual-rotor, variable-speed design in which two rotors and their impellers can adjust independently. The company says that arrangement is intended to respond to changes in load, water temperatures, flow rates, and pressure ratios rather than operate around a single fixed design point.
The commercial context is the rapid shift in AI data centres from air cooling towards direct-to-chip liquid systems. Higher rack densities concentrate thermal loads into a smaller footprint, making the design of the facility water system and CDU layer increasingly important to overall availability.
Liquid cooling does not remove heat from the data centre by itself. It changes how heat is transported from the computing equipment to the facility system. Operators still need to reject that heat through chillers, dry coolers, cooling towers, or another external system, depending on the local climate and design.
That makes the boundary between IT equipment and mechanical plant more operationally important. Flow temperature, water quality, pressure control, redundancy, leak management, controls integration, and maintenance procedures all affect whether high-density hardware can operate reliably.
Clivet also demonstrated a combined power-and-cooling concept with CLOU, which supplied energy-storage and power-delivery technology. The companies are arguing that electrical and thermal infrastructure should be designed together rather than as separate workstreams.
That principle becomes harder to avoid as individual AI halls move into very high electrical densities. A change in compute load produces a corresponding thermal change, while electrical resilience arrangements and cooling resilience have to be coordinated if a facility is to avoid moving a failure from one system into another.
The equipment launch is vendor-led, and the performance figures will ultimately depend on project temperatures, redundancy requirements, control strategy, ambient conditions, and actual installation. The notable part is the scale: CDUs are moving from supporting specialist pockets of liquid-cooled hardware towards acting as multi-megawatt pieces of primary data centre infrastructure.
For European projects, that shift will affect both new-build design and retrofit feasibility. Existing facilities may have room in the IT hall for denser compute but lack the pipework, pumping capacity, heat rejection, structural routes, or electrical plant needed to support it. The CDU is only one component in that larger conversion.
