Summary
- Johnson Controls has released a reference design guide for absorption chillers in AI data centres.
- The design targets sites with high-grade waste heat from on-site power systems such as gas engines or fuel cells.
- The approach ties cooling design, power availability, waste heat, and AI density into one facility-level energy calculation.
Johnson Controls has released an absorption chiller reference design guide for AI data centres, targeting facilities where on-site power generation creates recoverable heat that can be used to reduce electrical cooling load.
The design uses absorption chillers to convert high-grade waste heat into cooling. Johnson Controls says the approach can allow more electrical capacity to be used by IT equipment rather than mechanical cooling, particularly at large-scale AI sites considering gas engines, fuel cells, or other on-site power systems.
The guide forms part of the company’s wider reference design series for gigawatt-scale AI factories, alongside water-cooled, air-cooled, and direct-to-chip liquid cooling approaches. Its focus is the interaction between power generation, heat recovery, cooling architecture, and usable compute capacity.
Mechanical design is joining the energy plan
Absorption chillers are established industrial systems, but they have not been a default choice for most data centres. They use heat rather than electricity as the main driving energy source, making them suitable where steam, hot water, or other high-grade thermal energy is available.
AI data centres change the design conditions. Higher rack densities place heavier loads on cooling systems, while grid-constrained sites may lack enough electrical headroom for both IT load and traditional chiller plant. Where developers are already considering on-site generation to secure power or resilience, the heat from that generation can become part of the cooling system rather than a by-product to reject.
That creates a different energy balance. Cooling is no longer treated only as an electrical overhead measured after the IT load is known. It becomes part of the site’s power strategy, affecting how much compute can be supported within a given grid connection or on-site generation envelope.
Project-specific performance will depend on the heat source, operating profile, ambient conditions, cooling temperatures, redundancy design, water strategy, maintenance requirements, and the carbon intensity of the energy used. A gas-backed design may reduce electrical cooling demand while still raising emissions, fuel supply, permitting, and air-quality questions. A fuel cell or lower-carbon thermal source changes the calculation, but not the need for transparent evidence.
AI sites need full-system cooling decisions
Absorption cooling will not suit every facility. The technology needs a dependable source of usable heat, enough plant space, compatible cooling temperatures, and operations teams able to maintain another layer of mechanical complexity. It will be more relevant to large campuses with on-site generation, industrial heat sources, microgrid designs, or power-constrained locations where every megawatt of electrical capacity is valuable.
The wider trend is clear. AI data centre design is becoming less tolerant of standard templates. Electrical capacity, cooling topology, heat rejection, water use, backup systems, and controls now have to be modelled together. A design that looks efficient in one metric may be weak if it increases water exposure, adds maintenance burden, or relies on a thermal source that is not available at the right time.
Consulting engineers will need to test absorption cooling against direct-to-chip liquid cooling, conventional chilled-water systems, air-cooled plant, heat reuse, and hybrid configurations. Each carries different consequences for capital cost, redundancy, energy use, operational skill, and future retrofit.
The Johnson Controls guide does not make absorption cooling a default path for AI facilities. It does show how far cooling has moved from the mechanical room into strategic power planning. As grid connections become harder to secure, designs that convert waste heat into useful cooling will receive more attention.
The strongest test will be in real projects. Modelled savings and reference designs can guide early-stage engineering, but data centre operators will judge the approach by uptime, maintainability, energy cost, emissions, and whether it allows more IT capacity to be delivered within the same power constraint.
