The real work is keeping liquid cooling running
As direct liquid cooling, DLC, becomes a baseline for high-density AI servers, designers tend to focus on selecting the cooling method. In practice, the bigger issue is the set of operational requirements needed to keep that cooling system stable. Operating pressure, coolant temperature, leak testing, monitoring sensors: if these are missed, even high-density racks can stop on thermal limits. This article organizes the key operating numbers and standards for liquid cooling from primary information.
Operating pressure: how high to design
Typical operating pressure for a liquid-cooling TCS, Technology Cooling System, loop is 140-450kPa, or 20-65psi. This pressure range is the starting point for pressure ratings on piping, fittings, quick disconnects, and other components. Higher pressure makes it easier to secure flow rate, but raises leakage risk and component cost. Design sets the operating point within this trade-off.
Leak testing: the safety margin standards require
The electrical-equipment safety standard IEC 62368-1 requires leak testing at three times normal operating pressure, and two times pressure under single-fault conditions. If the operating pressure is designed at 450kPa, testing must therefore be planned up to roughly 1,350kPa. In liquid cooling, where liquid and electricity coexist, this leak-test safety margin directly constrains component selection and enclosure design. Setting operating pressure also sets test pressure.
Coolant temperature classes: ASHRAE W1-W5
The allowable coolant temperature also shapes design. ASHRAE defines facility-side coolant temperature classes in steps: W1 at 17 C, W2 at 27 C, W3 at 32 C, W4 at 45 C, and W5 above 45 C. Higher allowable-temperature classes make it easier to use outside air or dry coolers and reduce chiller load and power consumption, but the equipment-side thermal design becomes more demanding. Which class to run depends on site climate and PUE targets.
Managing the coolant itself: concentration and biological contamination
Coolant is not a set-and-forget item; it becomes an aging management target. Glycol-based coolants are considered to prevent bacterial growth at concentrations of 25% or higher. Concentration has to be set with freeze protection, viscosity, and biological-contamination control in mind at the same time. If concentration drops, biofilm risk inside flow paths rises, reducing heat-exchange efficiency and reliability. Coolant selection and concentration management need to be built into maintenance planning from the design stage.
How much heat to take through liquid: heat-capture ratio design
Even within "liquid cooling," the configuration changes depending on how much IT-equipment heat is captured by liquid. In a Hybrid Basic configuration, CPU/GPU cold plates can recover 70-75% of IT-equipment heat. Air cooling handles the rest. In contrast, Full Liquid classifications approach nearly 100% of components cooled directly by cold plates. A higher liquid-captured share reduces air-conditioning load, but routing liquid cooling to peripheral components such as memory and VRM increases design difficulty.
Monitoring: sensor requirements for staying online
Operational stability is secured through monitoring. Minimum liquid-cooling sensor requirements correspond roughly to ASHRAE DCIM Tier 2. Continuous monitoring of temperature, pressure, flow rate, and leakage is what allows abnormalities to be detected before systems stop. Sensor requirements may look like an ancillary specification, but in practice they are a core requirement for keeping high-density racks online.
Standardization status: OCP loop requirements
These operating requirements are also being developed as industry standards. OCP, the Open Compute Project, published the first version of its coolant document as Revision 1.0 on October 9, 2019, and later released Cold Plate Cooling Loop Requirements Rev 2 in November 2022. Rev 2 covers coolant, cold plates, internal chassis tubing, rack manifolds, QD, and CDU, showing a direction where the full cooling system is treated as a connected set of requirements. Aligning operating requirements with standards, rather than holding them as private specifications, preserves room for multi-source procurement and interoperability. Related articles cover the broader standardization picture.
The base layer: mass-production track record of components
Designers also need to check whether components that meet these operating requirements can be mass-produced. Boyd, a major cold-plate supplier, has delivered a cumulative total of five million liquid-cooling cold plates to hyperscalers. The emergence of suppliers that can not only meet numerical requirements but also supply reliably is what is making liquid-cooling operations realistic.
Pressure and leak testing
TCS loop operating pressure is 140-450kPa. IEC 62368-1 requires leak testing at 3x operating pressure, or 2x under single-fault conditions. The operating point determines test pressure.
Coolant temperature classes
ASHRAE W1, 17 C, through W5, above 45 C. Higher classes reduce chiller load but make equipment-side thermal design tougher. Select by location and PUE.
Coolant management
Glycol-based coolants suppress bacterial growth at concentrations of 25% or higher. Lower concentration can cause biofilm and efficiency loss. This is a maintenance item.
Heat capture and monitoring
Hybrid Basic captures 70-75% of heat through CPU/GPU cold plates, while Full Liquid approaches nearly 100%. Monitoring should be around ASHRAE DCIM Tier 2 level.
Business implications and checkpoints
When adopting liquid cooling, the key thing to verify is not the name of the method but the operating requirements. Check whether operating pressure and the leak-test safety margin align with component specifications, whether coolant temperature class fits the site and PUE target, whether concentration management and monitoring sensors are built into maintenance planning, and whether these requirements align with OCP and other standards so that multi-source procurement remains possible. Liquid cooling is not a one-time selection at installation. Treating it as requirements design for continuous operation is what brings a data center closer to staying online.