AI factories are no longer a future concept - they are industrial-scale production facilities already straining the limits of conventional thermal infrastructure. Yet the industry has lacked a standardized, engineering-grade roadmap for matching cooling architecture to the unprecedented heat densities those facilities generate. Johnson Controls is working to fill that gap.
On May 5, 2026, Johnson Controls (NYSE: JCI) expanded its AI Factory Reference Design Guide Series with the launch of its second guide, focused on air-cooled chillers. Building on the company's water-cooled chiller guide released in February, this latest blueprint marks the next step in what is positioned as the industry's most comprehensive set of global design guides mapping the full data center thermal chain - with additional guides covering absorption chillers and direct-to-chip liquid cooling to follow.
Why Thermal Management Has Become the Defining Constraint
At gigawatt scales, thermal management becomes a defining constraint. AI accelerators generate significantly higher heat densities than traditional enterprise IT, pushing cooling systems beyond conventional design assumptions.
As AI transforms industries, the scale and complexity of data center infrastructure is evolving rapidly. The ability to manage thermal loads efficiently at gigawatt scale is now a critical enabler for AI innovation, and the industry faces mounting pressure to deliver facilities that are not only high-performing but also sustainable and future-ready.
Operators face growing challenges: the power consumed by cooling systems, rising cooling-loop temperatures, and efficiency losses from heat islands. These pressures compound as AI inference and training clusters expand beyond hyperscale campuses into manufacturing environments, logistics hubs, and research facilities - all settings where cooling infrastructure complexity and reliability constraints are equally demanding.
What the Second Guide Covers
The new guide supports scalable data center design at all sizes up to a 1-GW AI factory utilizing air-cooled chillers. It outlines a comprehensive thermal cooling architecture integrating high-efficiency air-cooled YORK centrifugal chillers (including YDAM and YVAM models), fan coil walls (FCWs), and coolant distribution units (CDUs) to manage both air- and liquid-cooled IT loads.
The guide provides sizing references for 220 MW compute clusters, including recommended design temperatures and operating conditions across each stage of the thermal chain.
A central design decision involves chilled water supply temperature. By raising the chilled water temperature to support warm-water Technology Cooling System (TCS) loops, the architecture achieves a 30% Coefficient of Performance (COP) improvement and requires 27% fewer chillers. Designs can return up to 50 MW to the AI factory and improve annual energy consumption by 32%.
Architecture at a Glance
The guide's thermal chain spans from outdoor air intake through to rack-level heat rejection:
- Air-Cooled YORK Centrifugal Chillers (YDAM/YVAM): High-efficiency primary cooling plant; magnetic bearing models reduce mechanical losses and maintenance requirements
- Fan Coil Walls (FCWs): Manage air-cooled IT loads at row or zone level within the facility
- Coolant Distribution Units (CDUs): Interface between facility chilled-water loops and direct liquid cooling loops at rack level
- Warm-Water TCS Loops: Higher-temperature secondary loops compatible with next-generation GPU architectures, enabling COP gains at the chiller plant
The designs emphasize readiness for higher-temperature cooling loops, expected to become more common as GPU vendors push for greater chip-level efficiency. Elevated condenser water temperatures, bifurcated loops, and the use of high-lift chillers are highlighted as ways to improve annualized efficiency while supporting high-density AI deployments.
The Full Reference Design Series in Context
The air-cooled guide is the second installment in a series that will ultimately cover the complete range of thermal plant architectures available to data center designers.
| Guide | Cooling Technology | Release | Status | Key Outcome |
|---|---|---|---|---|
| Guide 1 | Water-Cooled Chillers | Feb 2026 | Published | Zero water consumption via dry coolers; up to 1 GW factory |
| Guide 2 | Air-Cooled Chillers | May 2026 | Published | 32% annual energy improvement; 50 MW heat return |
| Guide 3 | Absorption Chillers | TBA | Forthcoming | Waste heat recovery; details to follow |
| Guide 4 | Direct-to-Chip Liquid Cooling | TBA | Forthcoming | Rack-level high-density thermal management |
Each guide maps the full thermal chain, offering cooling architectures tailored to diverse compute densities, geographies, and elevations. No single approach fits every geography or workload; the series provides alternative cooling architectures that enable industry-leading power usage effectiveness (PUE) and water usage effectiveness (WUE) while maintaining the flexibility to scale across diverse climates and ambient conditions.
The first guide addressed water-cooled plants, which offer low PUE but require water availability and treatment infrastructure. The second guide extends the framework to air-cooled designs - a configuration relevant in regions where freshwater constraints or permitting restrictions limit evaporative cooling options. A headline outcome from the first guide is a zero water consumption heat rejection process using dry coolers, aimed at cutting operational costs and supporting sustainability targets in water-scarce regions. The air-cooled guide carries that water-efficiency logic forward through a fully air-side heat rejection path.
Implications for Designers, Integrators, and Operators
For electrical engineers and MEP consultants, the guide's value lies in its sizing references for 220 MW compute quadrants and defined operating conditions across each loop in the thermal chain - parameters that would otherwise require extensive site-specific modeling. Anchoring early-stage design work to a vetted, published architecture accelerates feasibility assessment and simplifies procurement specification.
Johnson Controls positions its reference designs as a way to standardize best practice while allowing operators to adapt to local constraints such as climate, water availability, and grid conditions. For systems integrators, this standardization translates into repeatable bill-of-materials structures and pre-validated commissioning sequences - both significant levers for compressing project schedules.
The guide defines temperature and operating conditions across all major facility loops, including TCS loops designed to support next-generation GPUs. By addressing both current and future thermal requirements, the designs aim to reduce the need for costly retrofits as chip architectures evolve and rack densities rise.
Johnson Controls also notes alignment with NVIDIA's DSX reference architecture, positioning the guides as compatible with emerging standards for large-scale AI factory design. For procurement officers and project managers, that alignment provides a basis for cross-vendor comparisons and simplifies capital expenditure justification.
Energy Efficiency and Regulatory Alignment
The emphasis on warm-water TCS loops and COP optimization directly addresses an escalating regulatory environment. Mandatory PUE targets - such as Germany's Energy Efficiency Act (EnEfG), which requires new data centers to reach a PUE of ≤ 1.2 within two years of coming online after July 1, 2026 - are pushing operators toward more granular efficiency accounting across each cooling plant subsystem, not just at the facility level.
The 32% improvement in annual energy consumption cited for the air-cooled architecture, achieved primarily through elevated chilled water temperatures and optimized chiller staging, translates directly into auditable energy performance metrics. For operators in jurisdictions with mandatory reporting obligations, that level of design-stage specificity is increasingly a procurement prerequisite rather than a differentiator.
Johnson Controls' AI Factory Reference Design Guide Series addresses this challenge by outlining how designers and operators can achieve industry-leading energy and water efficiency while minimizing noise impact on surrounding communities and remaining adaptable to different climates, workloads, and growth paths.
For broader context on how AI-powered systems are being deployed to optimize cooling operations in existing facilities, see the analysis of AI-driven cooling and predictive analytics in data centers.
Looking Ahead: What Comes Next in the Series
The two remaining guides - covering absorption chillers and direct-to-chip liquid cooling (DLC) - are expected to complete the series and address the highest-density end of the compute spectrum. DLC is particularly relevant for rack power densities exceeding 100 kW per rack, which are becoming standard in dense GPU cluster deployments.
Austin Domenici, president of Johnson Controls Global Data Center Solutions, noted: "At gigawatt scale, AI factories require a fundamentally different way of thinking about infrastructure. The future requires designing integrated systems that can scale predictably, perform efficiently and adapt as technology evolves. This guide reflects how Johnson Controls helps customers plan holistically for AI growth, from design to operations, anywhere in the world."
As AI workloads extend into non-traditional settings - manufacturing floors, logistics centers, and edge research facilities - the pressure to reduce per-unit energy cost of AI compute will intensify. Reference design frameworks that standardize thermal architecture while preserving design flexibility represent one of the more practical responses the industry has produced to date.
The full Reference Design Guide Series is available at johnsoncontrols.com/industries/data-centers/reference-designs1johnsoncontrols.com/industries/data-centers/reference-designs.
