Artificial intelligence is driving a fundamental shift in data center operations, with higher rack densities pushing traditional air cooling to its limits. As organizations deploy more AI and high-performance computing workloads, facility teams are increasingly evaluating liquid cooling technologies to improve efficiency and future capacity.
In this manufacturer roundtable, Facility Maintenance Decisions spoke with data center cooling system manufacturer about how cooling strategies are evolving and the infrastructure and organizational hurdles facilities must overcome to successfully adopt liquid cooling.
FMD: As data center densities continue to rise, how are facility teams rethinking the fundamentals of cooling strategy, and what operational challenges are emerging as a result?
“As AI workloads continue to increase rack densities, facility teams are rethinking how cooling systems are designed and managed. This reflects a broader shift, where managing facility equipment safely and precisely is now as critical as managing the IT racks themselves.
Traditional air-based systems are becoming harder to scale efficiently in these environments. This is driving greater interest in hybrid approaches in which air is supplemented by hydronics and direct-to-chip (DTC) cooling systems. Because water transfers heat more efficiently than air, these systems can support higher-performance computing while helping improve efficiency and reliability. Closed-loop systems further enable water and heat reuse by converting and repurposing waste heat into supporting adjacent applications.
However, these systems present new operational challenges. Teams must manage more complex cooling loops, tighter control sequences and greater interdependence between IT and facilities. Liquid cooling also adds new requirements for redundancy planning, water quality management, leak detection, flow assurance and integrated monitoring across mechanical and IT systems.”
— Matt Johnson, business development manager, building services, Xylem
“As data center rack densities continue to rise driven largely by AI and high-performance computing workloads, facility teams are increasingly treating cooling as a core design constraint rather than a supporting utility. With rack power levels now regularly exceeding 120 kilowatts (kW) and continuing to increase in some deployments, traditional air-based cooling is approaching physical and efficiency limits in heat removal and airflow management. Liquid transfers heat roughly a thousand times more effectively than air at the chip level, which is one reason direct-to-chip and other liquid approaches are moving from optional to the default. This is driving greater adoption of liquid and hybrid cooling approaches, such as direct-to-chip and rear-door heat exchange, which are being designed from the outset rather than added later.
This shift introduces operational complexity, including tighter coupling between IT and thermal systems, increased coordination between IT and facilities teams and new requirements around fluid handling, monitoring and specialized maintenance skills, particularly when integrating with legacy air-cooled environments. AI workloads also behave differently than traditional IT loads. They are inherently spiky, with multi-megawatt swings that are difficult to predict, which puts a premium on cooling and power systems that can respond quickly and on tighter day-to-day coordination between IT and facilities teams.
From a facility level and data center operations (DCOPS) personnel perspective, traditional operational skill sets require up-training in the areas of DTC. Ideally, the fundamentals of heat transfer through a liquid medium are already familiar to DCOPS teams. What is less familiar is technology cooling system (TCS) loop maintenance. Depending on the frequency of equipment swap-outs, the TCS loop can be contaminated through improper server swap procedures. A properly trained process would eliminate the potential for loop contamination and prevent impact on other expensive server equipment. Operating temperatures for the facility water system and TCS are also much higher than traditional cooling loops.”
— Mike Donahue, senior solution architect, Schneider Electric
FMD: Liquid cooling is gaining traction, but adoption varies widely. What factors are influencing whether facilities make the shift, and what organizational or infrastructure barriers tend to slow that transition?
“A key driver of liquid cooling adoption is the rise of AI and high-density computing. These environments generate significantly higher thermal loads. In many cases, liquid cooling is becoming a practical path to support higher rack densities, improve power usage effectiveness and future-proof capacity to maintain reliable performance as computing demands continue to grow.
Adoption tends to be slower in retrofit environments or facilities not originally designed with hydronic infrastructure in mind and that offer physical constraints, such as limited space for the required equipment like coolant distribution units and piping.
Organizational alignment is another challenge. When IT, engineering and facilities teams are not aligned early in the planning process, adoption can stall. As a result, many organizations are taking a phased or hybrid approach as they transition to newer cooling strategies.”
— Matt Johnson, business development manager, building services, Xylem
“Adoption of liquid cooling is primarily driven by increasing rack densities from AI and high-performance computing workloads, which are pushing thermal loads beyond the practical limits of air cooling in some environments. It is also influenced by energy efficiency goals, space constraints and sustainability targets, particularly in new builds or AI-focused retrofits. However, adoption is often slowed by infrastructure complexity, including the need to redesign mechanical systems for fluid distribution, heat rejection and redundancy.
Organizational barriers also play a role, including limited in-house expertise, uncertainty around standards and interoperability and the challenge of integrating new cooling architectures into existing air-cooled facilities and capital planning cycles, which often leads to phased or hybrid adoption strategies.
Validated reference architectures and integrated portfolios developed jointly with chip and server vendors are emerging as one way to reduce integration risk and shorten deployment timelines, particularly for teams adopting DTC cooling for the first time. Addressing these barriers will require earlier collaboration between IT, facilities and engineering teams, along with greater investment in workforce training, standardized design approaches and long-term infrastructure planning that accounts for future AI-driven density growth.
Closer coordination between construction and DCOPS teams is required when deploying DTC systems. A key constraint is project sequencing and commissioning timelines, which must accommodate startup and commissioning processes that differ from traditional data center builds, including load-bank-testing approaches specific to liquid cooling systems. In many cases, DTC systems are brought online after core facility systems have been started, allowing DCOPS teams and equipment owners to establish expectations and develop structured testing scripts for validating DTC equipment performance.”
— Mike Donahue, senior solution architect, Schneider Electric
Jeff Wardon, Jr., is the assistant editor for the facilities market. With more than three years of experience, he covers topics including technology, wellness, sustainability and emerging industry trends.
How AI Is Reshaping Data Center Cooling
Balancing Water, Energy and Sustainability: Data Center Cooling Strategies Are Evolving
Preparing Facility Teams for Next-Generation Data Center Cooling Systems
