For several years, the race to build AI infrastructure has centred on one question – where will the power come from? Today, across Europe and beyond, the next major constraint on AI growth is the resources required to remove the vast quantities of heat generated by high-density computing.
As rack densities move beyond 100kW and AI deployments scale towards gigawatt-sized campuses, water availability is becoming a critical factor in determining where new capacity can be built, how quickly it can be deployed and whether it can expand over time.
In this environment, cooling is a strategic infrastructure decision.
AI’s hidden resource challenge: water
For many operators, water availability is increasingly important when evaluating future capacity. A site may have access to land and power, but if cooling depends on large volumes of evaporative water use, long-term growth plans can quickly run into practical limits.
While rack density is rising and global datacentre electricity demand is projected to more than double by 2030, drought, extreme heat and water uncertainty are becoming real constraints on where and how facilities are built and add pressure to already-strained infrastructure.
Operators also now need to increasingly consider the availability of water. The question has shifted from how to cool a single rack to how the entire thermal system should be designed so cooling never becomes the constraint on growth.
A facility may have access to sufficient power and land, for example, but limited water availability could affect how much computing capacity can ultimately be supported. Similarly, a cooling architecture designed around today’s workloads may become a limitation as rack densities increase.
Thermal management, therefore, needs to be considered alongside capacity planning, rather than after decisions around compute and site design have already been made.
Zero-water and higher-temperature cooling are expanding what’s possible
The industry’s challenge is not to adopt a single new cooling technology. It is to integrate the entire thermal ecosystem. From direct-to-chip liquid cooling and coolant distribution units through to chillers and heat rejection systems, operators increasingly need architectures that optimise water, energy efficiency and computing performance simultaneously. They need to consider an approach which focuses on treating these elements as a connected thermal strategy rather than individual infrastructure components.
Zero-water cooling is particularly significant because it changes the resource equation. Designs that eliminate evaporative heat rejection can remove dependence on a water-intensive cooling process and give operators greater flexibility when evaluating potential sites and planning future expansion. At the gigawatt scale, zero-water designs have modelled the potential to eliminate more than 45 million litres of cooling water consumption per day.
That flexibility has implications beyond sustainability. If water availability no longer determines whether a site can support a particular cooling architecture, more locations can potentially become viable for datacentre development.
Similarly, higher condenser temperatures, up to 74°C, are expanding the role dry coolers can play in datacentre design. By allowing more heat to be rejected directly to ambient air, these approaches can reduce reliance on water-intensive cooling towers while also cutting noise levels by as much as 20dBA. Depending on the application, this can equate to a perceived noise reduction of up to 75%, resulting in sound levels comparable to a normal conversation or a quiet household appliance.
They can also reduce the amount of infrastructure dedicated to heat rejection, and give operators greater flexibility in site design and future expansion planning. Less water, less noise and a smaller cooling footprint open up sites near communities or in land- and water-constrained areas and can help facilities deploy and expand faster.
The business case goes beyond saving water
The value of zero-water cooling is therefore not simply the amount of water it can save. It is the capacity and flexibility that can be unlocked when water is no longer a fundamental constraint on cooling. Datacentres use significant energy beyond compute, with cooling accounting for an estimated 30% to 40% of total datacentre energy use.
Every resource used by cooling is a resource that cannot be used for compute, and every constraint that limits cooling can ultimately limit the amount of compute a facility can support.
Optimised thermal architectures can effectively return resources to the compute side of the facility. Modeled 1GW air-cooled architectures have demonstrated the potential to support up to 50 MW of additional AI capacity through thermal optimisation alone, and absorption-based designs that use on-site waste heat have shown potential for up to 97 MW of additional AI capacity.
There is another factor that is beginning to influence datacentre development – social licence to operate.
Across Europe, communities and regulators are paying closer attention to how datacentres consume local resources. Water use is increasingly part of that discussion. Reducing dependence on evaporative cooling does more than improve sustainability metrics. It can strengthen stakeholder confidence, simplify planning conversations and improve the long-term viability of future projects
Designing for growth from day one
As AI infrastructure continues to scale, thermal management will be a deciding factor in where and how new datacentre capacity gets built. The question is no longer simply whether a site has enough power to support AI. It is whether the site has the resources, infrastructure and cooling architecture needed to support that growth over time.
That means building thermal management into early decisions around site selection, capacity planning and infrastructure design, rather than treating cooling as a fixed requirement to solve after the workload is already set. Zero-water and higher-temperature cooling approaches can give operators more flexibility as rack densities increase, climate conditions change and resource constraints grow.
As the industry enters a period of unprecedented expansion, the next decade will require a broader question – can infrastructure be designed to grow without becoming constrained by the resources needed to cool it?
Organisations that answer that question early will gain greater flexibility in where they build, how quickly they scale and how effectively they navigate future resource constraints. In the AI era, cooling is no longer a support function, it is a strategic enabler of growth.
John Keane is VP and general manager, EMEA global data centre solutions, at Johnson Controls
