What changes when the cooling design is driven by monsoon humidity rather than by a temperate reference site.

Why the reference site does not transfer

Most cooling performance curves are published against a temperate design condition. In tropical markets the constraint is not dry-bulb temperature alone but the number of hours per year above a given wet-bulb threshold. Those hours determine whether an economiser mode is a design assumption or a rare event.

The practical consequence is that a hall sized on a temperate curve will spend most of the year running mechanical cooling, and the PUE that appears in the feasibility study will not survive the first wet season. The design basis has to be local weather data over at least ten years, not a single design day.

What liquid cooling actually buys you

Cold-plate loops move heat at a higher temperature difference than air, so the chiller does less work per kilowatt of IT load. The gain is real but it is not the whole story: the loop also concentrates heat into a smaller footprint, which changes how the hall is laid out, how maintenance is scheduled and how a leak is detected.

Immersion has a different profile. It carries higher densities and removes fans from the server, but it complicates serviceability and fluid logistics. In our experience the choice is driven less by peak density than by who will operate the hall for the next decade.

Metering as a design input, not a report

If the PUE target is contractual, it needs instrumentation at the cooling loop, the power distribution and the hall boundary from day one. Retrofitting sub-metering after commissioning usually costs more than the savings it was meant to prove.

We specify the metering points together with the cooling topology, so that the operating team can attribute consumption to a row, a loop or a tenant without a separate data project.

A phased sequence that matches grid reality

In most of the region, the grid connection arrives in steps. Designing the hall as a single integrated block forces either idle capacity or a delay to first revenue. Modular cooling and power trains let the first phase open on a partial connection, with the second phase commissioned when capacity is released.

The trade-off is more interfaces and more commissioning events. That cost is predictable; the cost of a hall waiting for a transformer is not.