Executive Summary

• Jake Roberts, CCO of Excool, explores the key to capacity planning and growth, being adaptable cooling.
• Demand is pushing developers to phased expansions and favour speed to market, which amplifies the cost of making the wrong decision. They run the risk of overbuilding and price itself out of more conventional tenants or limiting the ability to secure premium AI deployments with air-optimised infrastructure that has no room to adopt liquid cooling systems.
• Adaptable cooling enables developers and operators to reduce commitment risk while staying responsive to market demand.

Data centres now sit behind almost every mainstream digital service, from web platforms and business applications to the fast-growing wave of AI workloads. Yet, in a colocation environment, the facility team’s focus is on electrical and mechanical loads rather than the exact processes running on any particular server rack. Instead, they work to what they can specify and measure: power density, inlet temperatures, redundancy and how heat behaves under load.

That same sense of pragmatism rather than exact granular detail is increasingly shaping development decisions. A new data hall may be delivered as a blank canvas long before the final tenant mix is known, and the difference between a conventional air-cooled deployment and a liquid-cooled AI build changes everything. For developers, the challenge arises from ensuring a site can absorb changing demand without forcing expensive redesigns as the market evolves.

Data centre build programmes are long, and the technical assumptions underpinning a hall can be locked in years before the first rack even arrives. Despite this, demand pressure is pushing developers towards phased expansion and early commitments while the tenant mix is still fluid. This dynamic favours speed to market, amplifying the cost of making the wrong decision. A cooling design that is ideal for one tenant profile can become a constraint for another. The stakes are high when capacity is scarce and customer demand is volatile.

A site that is positioned for high-density liquid cooling may price itself out of more conventional tenants if the hall arrives overbuilt for most workloads. Conversely, an air-optimised hall can limit the ability to secure premium AI deployments if the required facility-water infrastructure is not available or if the means of installation are slow and disruptive.

As a result, mixed environments are becoming the norm, with air-cooled racks and liquid-cooled clusters sharing the same campus space. This improves tenant flexibility, but it can complicate plant and distribution planning while increasing the risk of expensive duplication if the site is designed too narrowly from the outset, capping growth opportunities.

Cooling flexibility has taken on greater urgency as platform roadmaps reshape facility expectations. NVIDIA’s Rubin platform has focused attention on warm water, single-phase direct liquid cooling, including a 45 degrees Celsius supply temperature target.

Where chilled water plants have often been treated as the default foundation for high-performance spaces, warm water expands the conditions under which heat rejection can be achieved using ambient air, particularly through dry coolers.

Ultimately, adaptable cooling is quickly proving itself to be the most futureproof and viable approach to data centre cooling system planning. A design that assumes a single operating regime can force costly compromise later, particularly in phased builds or mixed-tenant campuses. As such, supporting air and liquid-cooled architectures within the same site strategy, with a practical means of switching between them, is fast becoming the only viable competitive option.

One response is a shift towards packaged, perimeter-deployed hybrid cooling systems that can be installed quickly and configured as demand becomes clearer. This approach enables late-stage deployment within the build programme. Developers can defer a significant portion of cooling capex until tenant requirements are confirmed and fit-out is imminent, improving cash flow on phased builds while still being ready regardless of whether the tenant requires liquid or air-cooled facilities.

In an operating data centre, the service running on a server is often less important to facilities teams than the behaviour it creates in power and heat. A new hall is a blank canvas for a reason: it has infinite potential to accommodate anything, but cooling choices are difficult to reverse once concrete and pipework are in place.

Adaptable cooling empowers developers and operators to reduce commitment risk while staying responsive to market demand. In a sector where the next tenant requirement may not resemble the last, the ability to switch without rebuilding is integral to what makes a data centre commercially resilient.