SPACE, STRUCTURE, AND SCALABILITY Liquid cooling systems require dedicated space for circulation and distribution, affecting both floor loading and ceiling-mounted (hanging) capacity in AI data centers. To accommodate future growth, system design should prioritize flexibility, maintainability, and modularity. Design considerations vary across DTC, immersion, and RDHx implementations, with each introducing distinct structural demands. DTC adds weight from piping, liquid lines, pumps, and cooling distribution units (CDUs). Immersion systems impose the greatest load due to the volume of dielectric fluid and supporting infrastructure, while RDHx increases floor loading through liquid weight, piping, and heavier rear door assembly. All three systems require coordinated pipe and cable routing to maintain accessibility. Quick-coupling connectors and shutoff valves are essential to support maintenance and expansion without disrupting live operations. Leak management is also critical: DTC and RDHx systems typically use drip trays, sealed couplers, and leak sensors, while immersion systems may incorporate double-walled tanks for fortified containment. Pipe sizing should account for future capacity, and CDUs in DTC, immersion, and RDHx deployments must scale accordingly. Filtration requirements vary by system and should align with allowable particle size to ensure long-term thermal performance and reliability. KEY LIQUID COOLING MONITORING PARAMETERS Real-time monitoring helps maintain the performance, reliability, and safety of liquid cooling in AI data centers. Key operational parameters include: • Temperature and Volume: Monitor liquid temperature at system ingress and egress points— at the rack manifold for DTC, at the rack for immersion, and at the rear door for RDHx. Volume tracking varies by system. DTC systems monitor volume to support server additions or removals, immersion systems track fluid level in the tank, and RDHx systems measure total liquid volume.
REDUNDANCY AND RESILIENCE Redundancy is a foundational design requirement in liquid-cooled AI data centers, where new components and rack-level failure points must be addressed from the outset. Across DTC, immersion, and RDHx systems, redundancy should extend to new components: pumps, distribution pipes, CDUs, and heat exchangers. These systems are typically backed by dual power feeds, uninterruptible power supplies (UPS), and emergency power protocols to ensure availability. CDUs and pumps are often configured for load sharing to enable seamless failover during maintenance or component failure. DTC systems require additional safeguards due to minimal thermal inertia and high heat flux. They rely on redundant rack manifolds, server couplings, dual pumps, and heat rejection units with automatic switchover to maintain continuous coolant flow. In immersion and RDHx deployments, redundancy is typically built into CDUs and fluid distribution paths.
• Pressure and Quality: Measure liquid pressure at the pump across all three systems. Fluid quality is critical, as impurities can clog DTC cold plates, cause failures in immersion systems, and reduce heat exchange efficiency in RDHx. • Leak Detection and Environmental Monitoring: All liquid cooling systems require leak detection to prevent fluid loss and equipment damage. Air quality monitoring for two-phase vapor leaks applies to DTC and immersion systems but is not required for RDHx. Dew point monitoring— tracking both temperature and humidity— is important across all systems to prevent condensation.
Nevertheless, liquid cooling is positioned for significant growth. According to Grand View Research, cloud and hyperscale operators are expected to lead the market, which is projected to grow at a compound annual growth rate (CAGR) of nearly 24 percent over the next decade. 1 Additionally, a recent survey suggests that more than one-third of enterprise data centers plan to implement some form of liquid cooling by 2026. 2 As adoption accelerates, operators will continue to evaluate technical, operational, and regulatory considerations to guide deployment strategies. INFRASTRUCTURE CONSIDERATIONS FOR CABLING AND COOLANT ROUTING Integrating liquid cooling into AI data center rows and racks requires careful cabling infrastructure planning. Space must be allocated for liquid distribution components, including piping, manifolds, and airflow paths. In DTC systems, racks must accommodate both liquid cooling manifolds and rear airflow, while cable trays should allow space for cooling system pipes. Racks may require added depth for RDHx to support thicker rear doors to route cables and manifolds— especially when combined with DTC systems. Immersion cooling introduces additional operational requirements. All cabling must exit from the top of the rack, and any portion submerged in coolant must be chemically compatible with the fluid. Material compatibility is critical for both cabling and coolant. Cabling considerations include sheath and connector integrity, mechanical durability, electrical or optical performance, signal integrity, and compliance with flame rating and labeling standards. Fluid-related factors span potential contamination from cable materials, impact on thermal performance, and effects on system characteristics such as viscosity, filtering, and pump reliability.
SAFETY AND OPERATIONAL REQUIREMENTS
Liquid cooling deployments introduce distinct safety and operational requirements that vary by system type. Each approach requires updates to standard operating procedures, maintenance protocols, and emergency plans, along with specialized training for AI data center support personnel. Operators must monitor liquid and pipe temperatures across all systems. Coolants should be non-toxic, recyclable, and nonflammable to minimize environmental and health risks. To reduce slip hazards from leaks or drips, system designs should incorporate containment strategies and appropriate floor treatments. Weight is another critical consideration in DTC systems, which must account for the weight of individual servers. Immersion cooling, particularly with horizontal racks, requires provisions for safely lifting and removing heavy servers during maintenance. RDHx systems require structural support and handling procedures to manage the added weight of rear door assemblies. Personal protective equipment (PPE)—such as gloves, safety glasses, and protective clothing—is essential across all systems, with added emphasis on immersion cooling, where coolant handling and server access are more frequent. Operational procedures must also address new server and rack configurations, as well as the coolant distribution infrastructure.
Liquid cooling enables AI data centers to support increasing power densities driven by GPU-based workloads, while reducing energy use for thermal management, extending equipment lifecycles, and advancing sustainability and regulatory goals.
I
I
16
ICT TODAY
July/August/September 2025
17
Made with FlippingBook - Online catalogs