Data Center Cooling: How a Chilled Water Plant Is Built for Continuous Duty

A factory's chilled water plant is allowed to have a bad day. A chiller trips, production slows, a maintenance team resets it, and the shift carries on. A data hall's cooling plant is not allowed that day at all. The servers keep generating heat whether or not the plant is running, and the room temperature rises within minutes of the water stopping.

The equipment is largely the same: chillers, cooling towers, pumps, pipework, air handlers. What changes is the standard it is held to, and that standard reshapes almost every decision in the plant.

Redundancy is designed in, not added on

The first difference is that the plant is sized for the load plus the failure of something. In data-centre language that is N+1 — one more chiller, pump or tower than the load strictly needs — or, on more critical halls, two complete systems (2N) either of which can carry the load alone.

Redundancy only counts if it is real all the way through. A spare chiller that shares a single header valve, a single pump set or a single power feed with the duty chiller is not redundant; it fails with it. The piping arrangement, the valve positions and the electrical supply to each item of plant all have to be laid out so that any one component can fail, or be taken out for maintenance, without the cooling stopping.

That second condition — maintaining the plant without switching it off — is often the harder one. It is what drives isolation valves on both sides of every major item, bypasses around strainers, and pipe routes with room to work.

Ride-through: the minutes after a power failure

When the grid fails, the standby generators start, but not instantly, and chillers do not restart instantly either. For that gap the hall still needs cooling.

The usual answer is thermal storage: a buffer of chilled water large enough to keep circulating cold water to the air handlers while the chillers are down. The pumps that move it are put on uninterruptible power so that the water keeps flowing. Sizing that buffer is a calculation between the hall's heat load and the restart time of the plant, and it is one of the places where the mechanical design and the electrical design have to be done together.

Warmer water, more free cooling

IT equipment tolerates warmer air than most people expect. ASHRAE's widely used guidance for data-centre environments recommends server inlet temperatures in the range of roughly 18–27 °C. Supplying air in that range means the chilled water itself can run warmer than in a typical comfort or process system.

Warmer chilled water does two useful things. Chillers run more efficiently, because they are lifting heat through a smaller temperature difference. And for more hours of the year the cooling towers alone can do some or all of the work through a heat exchanger — free cooling — with the chillers idle. Both show up directly in the facility's power usage effectiveness (PUE), the ratio of total facility power to IT power that operators track closely.

Variable flow, matched to a changing load

A data hall is rarely full on day one. Racks are added over months and years, so the cooling load starts low and grows. Plants are therefore usually designed with variable-speed pumping and fans, so that the system delivers what the hall currently needs rather than running full-flow from the start. The control logic matters as much as the hardware: sequencing chillers, holding the right differential pressure, and keeping the plant stable at low load.

Commissioning proves failure, not just function

On a factory plant, commissioning shows that the system works. On a data hall it also has to show what happens when it stops working.

Integrated testing deliberately fails things: a chiller is tripped, a pump is isolated, the mains supply is cut and the generators are left to pick the load up. Each test is watched against what the design said would happen — which unit takes over, how quickly, and whether the hall temperature stays inside its limits. Only when the failure scenarios behave as designed is the plant ready for IT load.

That testing is coordinated across every trade at once: mechanical, electrical, controls, the UPS and generator vendors, and the client's operations team. It is the point where any boundary between contractors shows up.

Where VPSTEC fits

VPSTEC's chilled water, cooling tower, pumping, piping, electrical and commissioning capability is the same capability a data-centre cooling plant is built from. We are directing it at data-centre infrastructure, working alongside the project's consultants and specialist vendors — precision cooling OEMs, UPS and BMS suppliers — on the parts of the scope that are theirs.

What we bring is the discipline that makes the difference on a critical plant: one contract carrying the mechanical and electrical work together, so that redundancy, ride-through and commissioning are designed as one system rather than negotiated between packages. More on the data center infrastructure page.