High-density AI clusters bring new challenges to data center cooling designs. Cooling needs to be highly targeted, energy-efficient, and integrated with the rest of the data center infrastructure. Developing a compatible design could take an HVAC OEM many hours of engineering time, with no guarantee that the end result will be completely accurate or acceptable.
The good news: following a reference design means all that engineering work has already been done for you. And the resulting cooling system is highly likely to serve the data center owner well for years to come, delivering not only the intended cooling capacity but also maximum efficiency.
Elements of a reference design
A reference design is more than a list of components. It provides a validated system blueprint that shows how various components – such as chillers, pumps, piping, manifolds, and controls – come together to form a complete system that integrates mechanical and electrical systems into a unified operational platform.
Reference designs also help HVAC OEMs make informed decisions that align with industry best practices and with client needs. They detail how to transform individual components into an intelligent, software-defined system capable of supporting digital twins, AI-driven optimization, predictive maintenance, and autonomous operations. Instead of managing isolated systems, operators gain a connected platform that provides real-time visibility, smarter insights, and room to grow.
They can also help avoid costly missteps that lead to outages, which still occur regularly. In the Uptime Institute Annual Outage Analysis 2026, half (50%) of operators reported an impactful or serious outage in the past three years, with 10% saying they had a “serious” or “severe” outage.
The good news, if you can call it that, is that “only” 8% or organizations reported an outage attributed to cooling systems. Even so, a nearly one-in-ten chance of cooling-related outages is unacceptably high for mission-critical facilities. The financial stakes are equally significant, with one in five respondents to the Uptime Institute survey saying that their most recent impactful outage cost more than $1 million.
How reference designs are developed
Reference designs for cooling systems help to reduce that risk. To develop them, Schneider Electric engineers work with counterparts at companies that make other data center components.
For its EcoStruxure Reference Design 99, for example, Schneider Electric worked with NVIDIA to develop a solution specific to AI clusters using NVIDIA GPUs. The title alone – “3818 kW, Tier III, IEC, Chilled Water, Liquid-Cooled & Air-Cooled AI Clusters” – gives a clue to how detailed the design is.
The document includes designs for deploying high-density AI clusters in two IT rooms. IT room 1 depicts three retrofit scenarios in which a new high-density AI cluster is installed alongside existing traditional IT. IT Room 2 is a purpose-built room optimized for a liquid-cooled AI cluster that uses liquid-to-liquid CDUs.
The idea is to give guidance for common scenarios that HVAC OEMs are likely to face when providing solutions for AI data centers. It provides information for four technical areas – facility power, facility cooling, IT space, and lifecycle software – and details how the various systems integrate.
Working with Schneider Electric’s reference design teams offers OEMs a faster path toreducing engineering time, increasing speed to compute, and ensuring a more integrated solution. Every reference design includes technical documentation, including engineering schematics, floor layouts, equipment lists of all components used in the design, and 3D images showing real-world illustrations. For example, the level of detail in Reference Design 99 provides OEMs with exact specifications to follow for each scenario it covers. That saves OEMs significant time in engineering their own solutions and sourcing appropriate components.
From validated designs to intelligent operations
Reference designs provide peace of mind that a design will work as intended for the use case. That greatly reduces risk, a huge benefit for critical infrastructure like cooling in AI data centers.
But Schneider Electric designs have the added benefit of detailing how to integrate with EcoStruxureTM Data Center, our open, IoT-enabled platform for monitoring and managing data centers. It has modules focused on power systems, building systems (including cooling, security, and fire), and IT infrastructure.
EcoStruxure simplifies data center management by offering complete visibility, alerting, and modeling tools. It also offers predictive analytics advice on optimizing availability and efficiency in the IT space.
EcoStruxure helps cooling OEMs ensure maximum efficiency for their systems, a key concern in AI data centers since a kW saved on cooling can be applied to generating AI tokens. So, combining reference designs with the operational benefits of Schneider Electric EcoStruxure will bring maximum efficiency and reliability to your cooling solutions.
Access AI cooling reference designs for HVAC OEMs
As AI workloads continue to push rack densities higher, proven cooling architectures become a competitive advantage. Rather than spending valuable engineering time developing the solution, HVAC OEMs can leverage validated reference designs to accelerate deployment and reduce project risk. Check out our library of data center reference designs to speed up your next project and deliver optimized performance for data centers.
Add a comment