Cooling may unlock AI capacity in data centres

By Edward Lim

Data centre operators needing more AI capacity should look first at cooling systems rather than a complete rebuild, according to Johnson Controls’ Ali Badreddine, in an exclusive interview with Entelechy Asia.

Better use of existing power, cooling and space, combined with targeted upgrades, can create room for higher-density computing without replacing an entire infrastructure.

“Cooling alone can account for 30 to 40 percent of a data centre’s total energy use, making thermal infrastructure an important opportunity to free up capacity,” said Badreddine, Vice President and General Manager of Southeast Asia Business and Asia Pacific Data Centre Solutions at Johnson Controls.

“Existing data centres can support more AI capacity without a complete rebuild. A faster path is to make better use of the power, cooling and space already available, then target upgrades where they can unlock the most capacity,” he added.

Full examination before upgrading

For operators, it’s about examining the full cooling system rather than treating individual equipment replacements as isolated projects. Upgrades spanning heat capture, heat transfer, heat rejection, and controls can be phased into live facilities while improving capacity and efficiency.

Higher capacity chillers can slash the number of units required by half. Its advanced centrifugal chiller technology can also require up to 60 percent fewer dry coolers, and occupy almost 30 percent less space than comparable alternatives, according to Badreddine.

Liquid cooling offers another route to higher-density AI infrastructure, but he stopped short of putting a universal figure on how much additional computing an existing air-cooled site could accommodate.

“The capacity gained will vary by facility, based on available power, target rack density, heat rejection and space, but the opportunity is clear: integrated liquid and hybrid cooling can unlock higher-density computing without requiring a complete rebuild,” said Badreddine.

Making those gains reliable requires operators to assess power distribution, water loops, heat rejection, controls, maintenance, and redundancy together. Across Asia Pacific, available power, climate, water supply, space and local operating requirements also influence the mix of air, liquid and hybrid cooling. Zero-water cooling options can be considered where water availability is a constraint.

Waste heat could provide another way to stretch a limited electricity supply. Rather than rejecting thermal energy, operators can use it to drive an absorption chiller to reduce the electricity needed for cooling and leave more electrical capacity for AI workloads.

In a 1GW data centre reference design, Johnson Controls modelled an absorption chiller using 57 percent of the waste heat from on-site power generation.

Badreddine said the design showed the potential to reduce cooling-related electrical demand by up to 44 percent and enable up to 97 MW of additional AI computing capacity from existing power infrastructure. Those figures describe a modelled design, rather than results from a deployed facility.

“Heat pumps offer a separate opportunity where nearby buildings or industrial processes have a ready, consistent use for recovered heat. Selecting an energy-recovery approach depends on the site’s heat source, thermal demand and operating profile,” he said.

Regional nuances

Regional conditions make validation particularly important. High temperatures and humidity place greater demands on heat-rejection systems that require cooling designs to reflect local conditions rather than a uniform approach across markets.

Johnson Controls is participating in the Sustainable Tropical Data Centre Testbed 2.0, hosted at the National University of Singapore’s College of Design and Engineering.

The programme evaluates next-generation AI and data centre technologies in a live tropical environment with research intended to inform more efficient infrastructure designs for Singapore and other tropical markets.

Yet installing new equipment is only part of the task. Maintaining gains requires visibility into cooling and building-system performance, regular reviews and coordination between facilities, operations and business teams. Smart controls and connected digital platforms can help identify performance drift and validate whether upgrades continue to deliver as workloads change.

“With optimisation built into everyday operations, sites can keep adapting as AI requirements evolve rather than treating improvement as a one-time programme,” said Badreddine.

Asked which single change an Asia Pacific operator should prioritise over the next year, Badreddine put planning ahead of any particular cooling technology.

“The most important step is to develop a clear capacity roadmap that aligns expected AI demand with infrastructure and investment priorities,” he said, adding that such a roadmap should establish how much AI capacity the organisation expects to need, what existing infrastructure can support and which investments offer the greatest long-term value.