Designing & Building Data Centers for the Shift to Liquid Cooling

A B S T R A C T

Rising compute density is increasing demand for high-density servers and pushing more data center projects toward liquid cooling. For owners, the challenge is not simply how to cool higher loads. It is how to bring capacity online quickly while making infrastructure decisions that will affect constructability, operations, and future expansion. In many cases, facilities may need to accommodate both air-cooled and liquid-cooled environments, raising the stakes for decisions around cooling architecture, fluid strategy, and system integration.

This white paper examines how owners can respond to that pressure without overbuilding or limiting future options. It looks at the role of fluid strategy, the delivery implications of hybrid air and liquid environments, and the importance of aligning key decisions early enough to protect schedule. Under these conditions, design-build and experienced cross-disciplinary partners, including teams with deep data center delivery, MEP, and thermal fluid expertise can help owners manage complexity, reduce delivery risk, and move capacity online faster. The result is a clearer path to speed to market, with targeted flexibility for the next phase of data center growth.

I N T R O D U C T I O N

AI and other resource-intensive workloads are changing the design requirements of data centers. Traditional CPU-based servers handle complex tasks in sequence, while AI workloads rely heavily on GPUs that can process many calculations simultaneously. The result is greater compute density at the rack level and greater demand on data center infrastructure.

Supporting these higher-density compute environments requires more than a change in server technology. Traditional CPU server racks may operate in the range of 3 to 12 kW, while GPU racks can reach 30 kW, 50 kW, or much more. That concentration of compute power significantly increases heat loads within the same footprint. At higher densities, air cooling becomes increasingly difficult to scale efficiently. If heat is not removed effectively, systems can throttle performance to protect the hardware.

“Beyond a certain computing density, the use of air as a heat transfer fluid reaches a practical limit,” says Dave Baptiste, Senior Project Manager, MEP Services, Gray. “The heat load from today’s high-powered GPUs would demand such high airflow volumes that the data hall begins to approach wind-tunnel conditions. Considerations such as sizing air handling equipment for large airflow rates, increased ductwork/plenum sizing, developed laminar flow and noise become major design concerns.”

At the same time, demand for AI-ready infrastructure is increasing. JLL projects that global data center capacity could reach 200 GW by 2030, with AI workloads potentially accounting for half of the total. Data center developers, operators, and providers are under pressure to bring new capacity online quickly.

To support higher density rack environments and manage rising thermal loads, more projects are adopting liquid cooling. “Liquid to chip cooling methodologies using specialized heat transfer fluids such as 25% propylene glycol offer greater heat capacity at lower mass flow rates due to the higher density of the liquid, compared to gaseous fluids such as atmospheric air,” says Baptiste.

“Because chip thermal design power has increased so significantly, you have to use liquid cooling,” says George Zhang, Vice President, R&D, Valvoline Global. “You need a way to remove heat to protect chip integrity and prevent thermal throttling.”

The transition from air to liquid is not a clean handoff. Infrastructure requirements, procurement constraints, operational complexity, and evolving rack requirements can all make full transitions difficult. A U.S. Department of Energy Better Buildings presentation in 2025 noted that air cooling still accounts for about 90% of data center cooling.

For many owners and operators, the near-term challenge is not replacing one cooling model with another. It is planning facilities that can accommodate a mix of air-cooled and liquid-cooled deployments while preserving schedule, flexibility, and long-term operability. Teams that can resolve challenges early and keep critical workstreams moving together are better positioned to meet startup and operational targets.

As data centers evolve to include liquid cooling for higher-density computing, owners and operators have an opportunity to bring new capacity online more effectively and support greater compute density within the same footprint. Realizing that opportunity depends on more than meeting a thermal requirement. It also requires expertise in facility delivery and thermal-management fluids.

“Data center technology, and especially computing power, is advancing faster than traditional data centers can adapt,” says Ben Burgett, Vice President of Gray’s Data Center market. “With every increase in server density, design and operational challenges grow. To overcome these, you need design and construction teams marching in lockstep.”

Gray, Gray AES, and Valvoline bring distinct capabilities to this transition. Gray and Gray AES bring extensive experience in data center construction, design-build, MEP, and facility integration, while Valvoline brings deep expertise in coolants, lubricants, and thermal-management fluids integral to liquid-cooled operations.

This white paper examines the cost, design, delivery, and operational implications of liquid cooling adoption and outlines a practical roadmap for owners and operators planning liquid-cooled data center environments.

To read more, download the full whitepaper by filling out the form above.