lennox data center solutions blog

Discover the latest in Data Center Cooling technologies, events and industry trends shaping the future of Data Center cooling. Stay informed with expert insights, awards and recognitions, real-world case studies and valuable resources from Lennox Data Center Solutions

Beyond the Megawatt: Rethinking How Data Center Cooling Scales

Updated: November 13, 2026

For years, data center cooling was specified like power: as a single megawatt figure. With AI hardware densifying unpredictably, that forecast rarely holds. Operators should start with architecture instead, asking how cooling capacity can be added, reconfigured, and redistributed over a facility’s life.  

Granularity, continuity, and convertibility define a scalable design, which matters as operators move gradually from air to liquid cooling. Five questions help test whether a cooling partner’s approach truly scales. Capacity still matters, but it should be an outcome of the architecture, not its premise. 

LinkedIn
Beyond the Megawatt: Rethinking How Data Center Cooling Scales

Why cooling architecture, not cooling capacity, should be the first decision for an AI-era facility?

For most of the past two decades, data center cooling solutions were specified much like power: as a number. Operators forecast the IT load, sized the cooling plant accordingly, added a safety margin, and built. The megawatt figure on the drawing effectively became the cooling strategy.

That approach works only when the forecast holds. Today, it rarely does.

AI training hardware is becoming denser much faster than mainstream enterprise IT. Operators are feeling the pressure. In Uptime Institute’s 2025 global survey, concern about forecasting future capacity requirements increased significantly from the previous year.

When an entire facility design depends on the project’s least certain number, operators need a different starting point.

That starting point is architecture: the way cooling capacity can be added, reconfigured, and redistributed throughout a facility’s operating life.

Why is data center cooling so difficult to plan for AI workloads?

Rack densities, the balance between air-cooled and liquid-cooled data center equipment, and hardware refresh cycles are all changing at the same time. As a result, cooling capacity fixed during the design stage may be outdated before the facility is commissioned.

Industry data illustrates how wide the range has become. Uptime Institute reports that average rack densities are still rising slowly and that few facilities exceed 30 kW. Other industry reporting, in this case from ETDatacenters, indicates that average rack loads more than doubled, increasing from approximately 12 kW in 2024 to 26 kW in 2026.

These figures describe different populations, so both can be accurate. That is precisely the planning challenge.

A cooling plant is a 15- to 20-year asset. Its owner must plan for a range of possible outcomes, not a single forecast.

Shift the planning conversation from capacity to architecture

Once uncertainty is accepted as part of the planning process, the central question changes.

Instead of asking, “How many megawatts do we need?” operators can ask:

“How will this system grow, and what will it cost if our forecast is wrong?”

A cooling architecture designed to answer that question has three essential properties:

  • Granularity: Capacity can be added in standard increments that reflect actual demand rather than projected peak demand.

  • Continuity: Each addition uses the same components, controls, and service model as the previous one, allowing the facility to grow without adding operational complexity.

  • Convertibility: The system can support air-cooled, liquid-cooled, or mixed halls without requiring replacement of the heat-rejection plant.

Capacity still matters. The difference is that capacity becomes an outcome of the architecture rather than the premise on which the entire system is built.

The air-to-liquid transition is already testing scalability

The clearest evidence for the importance of scalable architecture can be found in facilities operating today.

AFCOM’s 2026 State of the Data Center report found that 39% of respondents said their current cooling did not meet all their operational requirements. The report also found that 36% had deployed liquid cooling, while another 28% planned to do so within 12 to 24 months.

Most operators are approaching this transition cautiously. They begin with rear-door heat exchangers and a limited number of cold plates because these technologies cause less disruption to existing infrastructure.

Earlier data reflects the same pattern. Among organizations using direct liquid cooling in 2024, nearly half applied it to fewer than 10% of their racks, indicated by Uptime Institute. Uptime Institute’s 2025 research also found that operators assess a liquid cooling system largely by how easily it integrates with their existing infrastructure.

The transition from air to liquid will therefore be gradual and uneven, taking place rack by rack and hall by hall. Cooling architecture should support that transition, not only its eventual end state.

In a two-phase refrigerant architecture, the same outdoor equipment can continue serving a hall as it moves from air cooling to liquid cooling. The necessary changes occur within the white space.

Supplying refrigerant directly to the computer room air handlers, or CRAHs, also removes an intermediate heat-exchange step. This improves efficiency at every stage of the transition.

Five questions to ask a cooling partner

The following questions can help operators determine whether a cooling approach is designed to scale or designed around a single capacity figure:

  1. What is the smallest increment in which we can add capacity, and will adding it require downtime?

  2. Is redundancy built into the standard unit, or must it be engineered separately during each phase?

  3. If our air-to-liquid ratio changes, what equipment outside the building will need to change?

  4. How many heat-exchange steps sit between the chip and the atmosphere?

  5. Will the partner that commissions phase one still support the system at phase five?

Together, these questions move the conversation beyond initial capacity. They address how the cooling system will adapt, operate, and receive support throughout the life of the facility.

Plan how to grow before deciding how big to become

At Lennox, these questions shaped the design of PhaseFlow. Its waterless, modular architecture allows operators to determine how they will grow before they determine how large they will become.

The broader principle applies regardless of the technology an operator selects.

The megawatt figure will still appear on the drawing. It simply should not be the first decision.

What are modules, pods, and blocks in Lennox PhaseFlow’s data center cooling?

In PhaseFlow’s modular cooling hierarchy, the cooling module is the base unit, assembled from factory-built skids. Modules combine to form pods, and pods combine to form blocks.

A block is the standard deployment unit. It contains 16 modules, with one module held in reserve to provide the N+1 redundancy appropriate for Tier III designs.

Each level consists of identical units from the level below. This consistent hierarchy supports deployments ranging from approximately 150 kW to 300 MW and beyond.

In practice, an operator can commit to a single 18 MW block and expand it pod by pod or block by block as demand materializes. Redundancy is incorporated into the standard unit rather than engineered separately during every phase.

The fifth expansion follows the same model as the first.

How Data Center Cooling Can Scale for AI Growth