So a Plumbing Engineer Walks Into a Data Center Conference…
Key Takeaways
- Established plumbing and HVAC practices give data centers a foundation for managing coolant flow, balancing loads, and removing heat.
- Comparing supply and return flow may provide earlier leak warnings, while stored cooling can help bridge interruptions in chiller operation.
- Air, liquid, and hybrid cooling should match computing loads, while facility heat rejection requires weighing utility costs, climate, and water availability.
- Closed-loop cooling still needs water quality management because oxygen entering the system can sustain corrosion and produce particles that threaten cold plate microchannels.
- Shared data standards can help building and IT systems interpret equipment readings together, supporting analytics and digital twins across the cooling system.
No, I wasn’t lost. But I felt lost at first.
The data center market is the current siren song to business leaders around the water industry. It’s an explosive market, growing from $187 billion of market revenue in 2020 to a projected $517 billion by 2030, a CAGR of 10.5%. That growth comes with high stakes for reliability: Uptime Institute’s 2026 Annual Outage Analysis found that one in five operators reported their most recent impactful outage cost more than $1 million. With millions at risk when operations go down, and a race to get these facilities built as fast as technologically and physically possible, data centers are the Formula One of plumbing.
At the Open Compute Project (OCP) Global Summit held recently in San Jose, it was a showcase of emerging technology: server nodes, server racks, coolant distribution units (CDUs), row manifolds, and more, all in support of uptime. I don’t come from the data center industry, but rather from flush valve and sanitaryware design, plumbing codes & standards, certification, and now product management for data-driven water. The expo was very much focused on what’s new in compute, but I was looking at the pipes. Quick disconnect (QD) fittings, manifolds, ball valves, strainers, and pumps lined the booths, and very little of it would look out of place in a mechanical room.
During the talks, it often felt like the answers to their water-control questions had been well-known within the plumbing and HVAC community for more than 30 years. Water distribution in a data center follows the same fluid dynamics as any hydronic system, and the heat load from each server rack behaves much like the load on each floor of a building’s chilled water riser. The difference is what gets balanced. Instead of balancing based on pressure drop, the balance is based on energy (flow x ΔTemp x the fluid’s Cp across the load) since all power to chips gets converted into heat. A balancing valve controlling for power may be new to data centers; a balancing valve is not.
What’s Missing From the Data Center Resiliency Conversation?
Leaks did not get sufficient mention at the conference. Water can be destructive when escaping the plumbing system, damaging building materials and initiating mold infestations. You can imagine the deleterious effects of leaking water within a server rack or within the facility in general. The effects are rare, but the impact can be catastrophic further upstream at the rack manifold, technology cooling system (TCS) loop, facility water system (FWS) loop, between the CDU and the chillers, dry coolers, or cooling towers. The typical means of alerting for leaks is with sensing cables i.e., leak ropes. The problem is that a sensing cable tells an operator where water has already landed, which makes it the last line of defense rather than the first.
One way to more quickly identify a leak’s location is with sensing cables that detect the distance along a linear path to better follow the flow of the escaped liquid. Some compute trays themselves can have leak detection probes. Leaks can also be detected optically by a technician, as PG25 (propylene glycol, 25% by volume) contains fluorescent dye mixed in with the rust inhibitors and additives.
While leak sensing cables are necessary for racks and CDUs, the rate of nuisance alarms is concerning due to condensation, cable damage, cable connection / termination issues, human error during maintenance, and more. Another approach is to compare supply and return flow using matched, compensated flow meters in a closed loop. A difference can indicate a leak before escaped liquid reaches a sensing cable, potentially giving operators an earlier warning.
Another threat to uptime explored during the show was the speed at which devices generate heat versus the heat transfer away. Liquid cooling removes the thermal battery (room air) that once was prevalent. While a pump can run on UPS power, a chiller is too large for that. If there is a power interruption or the chiller requires a reset, it can take several minutes to get it back up and running. Minutes matter at the cold plate, and the typical thermal redundancy is a large thermal mass to account for potential downtime. Maybe stored cooling is already a known feature for thermal ride-through? I was looking for it in the resiliency talks.
Does Air Cooling Still Have a Place in Data Centers?
Not every data center is adopting liquid cooling the same way, and their choice depends on system size and compute load. At the top end, AI racks now draw well over 100 kW, far more than air can cool, so liquid cooling is the only practical answer. General-purpose compute is a different story. At the summit, a presentation by Jayesh Shah of Microsoft and Stewart Nguyen of Lenovo showed that server CPUs heading from 650 W toward 1,000 W can stay air-cooled if the rack gets wider. Moving from a 600 mm to a 900 mm rack adds about 50% more airflow area, and because pressure drop scales roughly with the square of velocity, fan-power per kilowatt-cooled falls by 20-30%. Any plumbing engineer who has upsized a pipe to cut friction loss will recognize the logic.
Between those two approaches sits the hybrid model. Cold plates cool the processors and memory while fans handle everything else, and the presenters estimated that going fully fan-less adds about 50% more liquid cooling infrastructure cost on top of that. Existing air-cooled halls can host a handful of liquid racks this way, often through liquid-to-air CDUs that reject heat into the room without new facility piping. The presenters’ conclusion was that liquid makes sense once a rack exceeds the air envelope, not before, which leaves the MEP engineer with a familiar job: match the system to the load.
The other area that air has an effect is in the cooling tower. The industry touts closed-loop liquid cooling. However, the economic decision between evaporative (open-loop) and dry (closed-loop) heat rejection is based on regional water cost vs electricity cost of a potential DC site. A presentation by Dr. Sammy Houssainy and Dr. Karim Shalash of Daikin Applied introduced the concept of the Water-Energy Selection Index (WESI) to help make this decision based on electric and water utility rates. For example, the WESI shows that it is more cost advantageous to perform dry cooling in the Pacific NW, and wet-cooling in Arizona. Note that this is based on utility costs and weather and there was no consideration for water availability or future water stress.
How Does Water Quality Affect Closed Loop Cooling?
Water quality is the third area where plumbing practice applies, and it was the subject of a Cooling Environments Lightning Talk by my Watts colleague Stirling Boston from EasyWater and Kelley Mullick of Avayla Consulting Group. Their argument is that a closed loop is only nominally closed. Oxygen comes in dissolved in the makeup water, or through vented tanks and faulty air vents, or during shutdowns and maintenance. An untreated closed loop typically carries 3 to 7 ppm of dissolved oxygen, enough to sustain continuous corrosion. The corrosion products become the fouling and abrasive particulate that threaten cold plate microchannels. Chemical treatment slows that process without removing its cause: passivation protects the metal but leaves the oxygen in place, and dosing enough biocide to hold bacteria in check can itself accelerate corrosion.
Their alternative is continuous side-stream treatment: remove dissolved oxygen through physical conditioning and clear iron oxide and sediment with sub-micron filtration, all without chemical treatment. In building hydronic loops, they reported dissolved oxygen falling from 3-7 ppm to <1 ppm within 2-3 weeks, and validation on a live TCS loop is under way. Hydronic designers will recognize the remedy, since air elimination has been part of good closed-loop practice for as long as there have been expansion tanks and air separators. It also ties back to the leak discussion, as every gallon of makeup water that replaces a leak brings fresh oxygen with it, so a slow leak and a corrosion problem are closer cousins than they first appear.
How Can Building and IT Systems Share Cooling System Data?
Digital twins are another crossover area for engineers. We have a problem in the plumbing industry in that BMS controls typically do not include anything domestic; the controls for a water heater, pump, or digital mixing valve live inside the equipment at the edge. As such, we are behind with our journey into digital twinning and analytics that can suggest potential next steps. However, the Brick schema is trending in the direction of making space for plumbing so that applications can be built on top of tagged data.
ASHRAE 223P and Brick describe building systems (mostly HVAC), however there is very little there yet for data center applications. Part of this may be because the means to communicate with and identify the equipment in the white space (i.e. the tech side) is increasingly Redfish. Redfish is a REST API using HTTPS to communicate using a JSON structure, and defines its own schema-based data model rather than tags. Further, a BMS manages the OT (operations tech) system while a DCIM (data center infrastructure management) system tracks the IT infrastructure, including rack power, capacity, and assets, while each server’s individual baseboard management controllers (BMC’s) manage each server. Below is a diagram that relates the two domains. Notice how there is a distinct split in communication and control except for shared equipment like the CDU, which bridge the two.
What does this mean for the plumbing engineering community that typically does not interact with Division 25 controls requirements in the construction spec? Many plumbing engineers operate in MEP firms that do chilled water designs. Lend your expertise, comment on ASHRAE 223P, and give some suggestions to Brick so that both can expand and prepare for plumbing. Your expertise is needed in the Open Compute Project in order to improve those standard designs operating at the bleeding edge.
Conclusion
What this means for the plumbing industry is that we can both learn current practices from the data center world as it pertains to using information, as well as teach them what we have learned over time, while also improving our own understanding of hydraulics given their high performance-driven needs. Consider participating in OCP workstreams and let’s learn more while sharing our knowledge of what goes on inside a pipe.
About the Author
Conrad Jahrling
Product Manager, Data Centers – Watts®
For nearly two decades, Conrad worked to make water systems create better outcomes: co-inventing high-efficiency flush valves and fixtures at Sloan Valve that cut water use by 87%, and leading ASSE International water safety standards development and product certification. He’s glad to talk water modernization in the built environment, product management, or ontologies