Johnson Controls launches series of thermal management reference design guides for gigawatt-scale AI factories

Supporting the new AI-based economy

As AI continues to evolve – and gigawatt-scale AI factories become more prevalent – the industry faces mounting pressure to deliver facilities that are not only high-performing, but also sustainable and future-ready. The designs for these infrastructures must be repeatable and globally scalable to empower the deployment of AI accelerators (purpose-built hardware designed to handle AI workloads) and keep up with the growing demand of an AI-based economy.

To help data center operators and facility managers respond to this rising challenge, Johnson Controls has launched the first of a series of global reference design guides for gigawatt-scale AI data centers.

Focused on the emerging class of AI factories, these reference design guides outline how to build high-performance, scalable facilities from a thermal management perspective. Each guide in the series maps the full thermal chain, from capturing heat at chip level to managing the final atmospheric rejection.

Water-cooled chiller plant for a 1 gigawatt AI factory

The series kicks off with a blueprint for a water-cooled chiller plant capable of serving a gigawatt-scale AI factory. Future guides will turn the spotlight on air-cooled and absorption chiller solutions.

The water-cooled reference design guide outlines a complete thermal architecture that incorporates the major components of thermal management such as:

• Computer room air handlers (CRAHs)
• Fan coil walls
• Coolant distribution units (CDUs)
• Heat exchangers
• Thermal controls
• High-efficiency YORK centrifugal chillers

The guide also provides detailed sizing guidance for 220MW compute quadrants and a 105MW central core. It defines temperature and operating conditions across all major facility loops, including technology cooling system (TCS) loops supporting next-generation GPUs.

The series has been designed with a global audience in mind, where each data center location comes with its own unique set of challenges to overcome. As 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, ambient temperatures and elevations.

Critical outcomes for data center operators

Reference Guide 402 from Johnson Controls allows data center operators to plan and design a water-cooled chiller plant capable of serving a full 1 gigawatt AI factory. The approach outlined in the reference design guide delivers the following critical outcomes:

• Zero water consumption: A fully water-free heat rejection process using dry coolers, reducing operational costs and advancing sustainability objectives
• Future-ready thermal flexibility: High-temperature TCS loop readiness aims to ensure compatibility with forthcoming GPU architectures
• Optimized high-density AI performance: Alignment with NVIDIA DSX reference architecture enables scalable deployment of 1-GW-class AI factories
• Energy-efficient operation: Elevated condenser water temperatures, bifurcated loops and YORK high-lift chillers aim to deliver good PUE and improved annualized efficiency

Speaking at the launch of the series, Austin Domenici, Vice President & General Manager at Johnson Controls Global Data Center Solutions, put it this way:

“AI factories are production facilities – the places where intelligence is manufactured at an industrial scale. By supporting the NVIDIA DSX reference architecture and improving water and energy efficiency in the cooling process, while maintaining high temperature loop compatibility, our Reference Design Guide equips customers to deploy gigawatt-scale AI infrastructure that is scalable, repeatable, resilient and sustainable.”