Making every drop count in efficient liquid cooling systems

AI may be digital, but its growth depends on physical infrastructure: chips, power, water, and heat-removal ability.

As rack densities climb past 100 kW and, in some cases, 130 kW, the industry’s central question has shifted from how to power more compute towards how to cool it without placing unsustainable demands on local resources. This has become an ongoing puzzle.

At the forefront is sustainable solutions provider Ecolab, which supports international data centre organisations in their cooling endeavours. Matthew McLaughlin shared his expertise on optimising sustainable water and liquid cooling solutions with Capacity.

“Data centre operators increasingly recognise that water and energy infrastructure can determine a facility’s license to operate,” McLaughlin said. “When managed in ways that support surrounding communities, these resources can enable responsible growth.”

Traditional linear cooling approaches (which take water and energy in, and send heat and waste out) were not designed for rapidly growing AI density.

A more resilient model uses water circularly through reuse, recycling, and liquid cooling technologies that keep fluid in productive service, McLaughlin acknowledged. To scale responsibly, operators must optimise water, energy, and performance as connected outcomes. The goal is using less water while maximising the value of every drop.

For McLaughlin, the solution starts by focusing on the physics: “Liquid – specifically water – moves heat far more effectively than air. Water can carry roughly 3,500 times more heat than the same volume of air and transfer heat about 23 times faster.”

“Think of a hot engine: a fan can cool its surface, but circulating coolant through the engine removes heat directly from where it is generated. The same principle applies to high-density compute. At today’s densities, that difference is decisive. Air cooling depends on fans and chillers pushing enormous volumes of air across hot components, and that movement takes significant power.”

Bringing liquid closer to the heat, whether through water in a facility loop or coolant flowing through a technical loop closer to the chip, captures heat more directly at higher temperatures – with less energy spent moving it.

The energy story is also a water story. Much of the world’s electricity depends on water somewhere along the way, from cooling thermal power plants to running hydroelectric dams. Every kilowatt-hour a cooling system avoids also avoids the water used to generate it upstream.

McLaughlin noted that this part of the equation often goes uncounted because it happens outside the data centre’s fence: “But it’s real, and it means efficient cooling improves the water math in two places at once. On site, water is used purposefully and kept in circulation. Off site, lower power demand reduces the water embedded in the grid that feeds the facility.”

“It is tempting to frame cooling as a choice between competing technologies,” McLaughlin said.

“In practice, high-density sites operate through a cooling chain: heat moves from the chip to the rack, from the rack to the facility, and from the facility to the atmosphere. Because each stage has different requirements, most sites use a combination of cooling methods rather than relying on a single technology.”

Direct-to-chip liquid cooling takes on the hottest components, while air or evaporative cooling handles the remaining load in the rack and the room. At the facility level, heat is rejected through evaporative cooling towers, dry coolers, adiabatic systems, or hybrids, depending on three main factors: climate, power, and water availability.

Evaporative cooling remains one of the most energy-efficient ways to reject heat, and reusing water or using non-potable sources helps make it water-efficient.

Regardless of cooling method or combination, cooling must consider the local environment and performance requirements, looking at impacts both inside and outside the facility fence. The right mix is site-specific and, for McLaughlin, the operators leading the way are designing for local conditions rather than copying a template.

“Where operators can go wrong is optimising one metric in isolation. Water Usage Effectiveness (WUE) and Power Usage Effectiveness (PUE) are often reported as separate scorecards, but they’re linked. An approach that uses less water at the point of cooling may need more energy, and generating that energy can consume more water upstream at the power plant.”

The better approach treats heat, water, and energy as one connected balance, measured inside and outside a facility based on climate, power grid, water availability, and performance requirements. That means asking where a site’s power comes from, how water-stressed the local watershed is, and how cooling choices shift load between the two.

With those questions answered, operators can then make trade-offs and determine a cooling approach that uses resources as effectively as possible.

Regardless of cooling method, efficiency only holds if the loop stays healthy. In any kind of system cooled by water or liquid, the fluid itself is critical infrastructure. Corrosion, microbiological growth, particulates, water variability, and coolant drift all erode heat removal effectiveness over time.

McLaughlin explained: “Even in direct-to-chip liquid cooling applications, water and coolant fluids are not ‘set-and-forget’ utilities. Real-time monitoring of coolant chemistry, temperature, and flow helps operators identify out-of-spec conditions and manage risks proactively, so they can protect servers and mitigate risks.”

The proof is how it works in practice. To demonstrate this principle, McLaughlin described several case studies. At a hyperscale site, continuous monitoring caught corrosive particles that routine maintenance had missed, enabling an early fix. Meanwhile, at a colocation provider, it flagged diluted coolant during commissioning, and an immediate correction kept the facility on schedule.

Both instances mitigated the risk of a potential outage down the line, preserving uptime and operator peace of mind.

Water reuse, recycling, and direct liquid cooling all require careful monitoring to maintain fluid performance. Without it, circularity at scale introduces too many variables to support reliable performance.

“For teams planning their next high-density build or retrofit, a few principles go a long way,” said McLaughlin. First, he advised that operators start with the site. Climate, power source, water availability, and workload profile should shape the cooling design – “not the other way around.”

Operators should also count water and energy together. To do this, McLaughlin advised they model WUE and PUE side by side, including the water embedded in the power generated upstream in the process. Operators should also match the method to the heat: “Use liquid where density demands it, and pair it with the facility-level heat rejection that fits local conditions.”

Finally, operators should treat the fluid as infrastructure. McLaughlin stressed again that the fluid itself is critical infrastructure, adding: “Put monitoring in place from commissioning onward so efficiency holds over the life of the system.”

AI growth will continue to increase demand for compute, power, and cooling. However, resource waste does not have to increase in kind. It’s imperative that it doesn’t; while additional power capacity can be built, water availability remains constrained by local conditions. Too much interference can have wider ecological and human impacts.

“Operators that use water and liquid cooling effectively – and manage water and energy as one connected system – will be better positioned to deliver more compute with fewer resource demands,” McLaughlin concluded.