Johnson Controls has released its second AI Factory Reference Design Guide, this time focused on air-cooled chillers for large-scale AI data centers. The company says the design can support facilities up to 1 GW while reducing energy use, cutting water consumption to zero, and improving thermal efficiency for growing AI workloads.
As artificial intelligence workloads continue to reshape the data center industry, cooling systems are becoming one of the biggest engineering challenges. Johnson Controls has now expanded its AI Factory Reference Design Guide Series with a new framework focused on air-cooled chiller systems for hyperscale AI facilities.
The announcement follows the company’s earlier water-cooled chiller guide released in February 2026. Johnson Controls said more reference designs covering absorption chillers and direct-to-chip liquid cooling are also in development.
AI growth is changing data center cooling
Modern AI factories consume far more power than traditional data centers. Higher rack densities, rising cooling-loop temperatures, and water shortages are pushing operators to rethink infrastructure design.
Johnson Controls said its latest guide is intended to provide a repeatable model for cooling data centers ranging from smaller deployments to facilities as large as 1 GW.
The architecture combines YORK centrifugal chillers, including YDAM and YVAM systems, with fan coil walls (FCWs) and coolant distribution units (CDUs). The design supports both air-cooled and liquid-cooled IT environments.
The company also included sizing references for 220 MW compute clusters and operational recommendations across different stages of the cooling chain.
Focus on zero-water cooling
One of the headline claims in the guide is the elimination of cooling towers through a zero-water cooling approach.
According to Johnson Controls, this could save more than 12 million gallons of water daily in large AI facilities. Water availability has become a growing concern for hyperscale operators, especially in regions already facing climate stress and utility pressure.
The guide also outlines strategies to reduce the “heat island” effect created by large air-cooled chiller plants. Johnson Controls said this approach could deliver up to 20 MW in peak power savings.
Efficiency gains and power recovery
The company highlighted several projected efficiency improvements tied to the design.
Among the key figures:
- Up to 50 MW returned to the AI factory through separate air- and liquid-cooling loops.
- A 32% improvement in annual energy consumption using redundant chillers more efficiently.
- A 30% improvement in Coefficient of Performance (COP).
- A 27% reduction in the number of chillers required by raising chilled water temperatures for warm-water cooling loops.
The growing shift toward liquid cooling in AI infrastructure has increased demand for systems that can handle heavier thermal loads without sharply increasing energy use.
Industry preparing for gigawatt-scale AI sites
Austin Domenici, president of Johnson Controls Global Data Center Solutions, said AI factories require a different approach to infrastructure planning.
“At gigawatt scale, AI factories require a fundamentally different way of thinking about infrastructure,” Domenici said. “The future requires designing integrated systems that can scale predictably, perform efficiently, and adapt as technology evolves.”
The company said the reference guides are intended to help operators create data centers that can adapt to changing workloads, regional climates, and future expansion needs.
Johnson Controls plans to continue expanding the guide series as AI-driven infrastructure demand accelerates across global markets.