In the largest year-over-year jump in AFCOM’s decade of research, the group’s State of the Data Centre Report 2026 found that average rack density has climbed from 16 to 27 kW in just one year. The study also found that 72% of operators expect AI workloads to increase data centre capacity requirements, while more than 60% of organisations are either already using liquid cooling or plan to adopt it within the next two years.
Scaling direct-to-chip liquid cooling demands a complete rethink of the cooling stack, with the industry challenged to design integrated, scalable architectures that support next-generation high-density compute.
Six paths to heat rejection
To select the most appropriate liquid cooling architecture, data centre operators must weigh four factors: compatibility with existing cooling infrastructure, deployment size, speed of deployment, and energy efficiency. Three heat rejection methods combined with one of two coolant distribution unit (CDU) form factors create the six most common liquid cooling architectures.
The first step is to identify the heat rejection method which determines how heat is transferred outdoors, and dictates the type of heat exchange used by the system.
Reject to Air: Liquid to air, existing system. The CDU expels heat into data centre air, working with existing air-cooled infrastructure. This is best for small deployments from a single server to a few racks where speed of deployment is the priority.
Reject to Facility Water: Liquid to liquid, existing system. The CDU connects to chilled or condenser water loops. This is suited to mid- to large-scale sites with chiller plants, prioritising efficiency over rapid rollout.
Reject to Independent Water: Lliquid to liquid, dedicated system. This builds a new heat rejection system for liquid cooling. It is ideal for major deployments focused on efficiency and heat reuse, where longer timelines and new piping are feasible.
The CDU form factor determines how cooling fluid is distributed among IT equipment. A rack-mounted CDU is installed in the rack, creating an isolated loop. Best for one to ten racks needing rapid installation, it limits failures to one rack, but becomes less efficient and costlier per kW as scale grows.
A floor-mounted CDU is placed on the facility floor, serving multiple racks or rows via a shared loop. Suited for more than ten racks where workloads can tolerate a common-cause failure, it offers better efficiency, saves rack space and lowers cost per kW at scale, though a single failure affects all connected racks.
Retrofitting existing facilities
Integrating high-density AI clusters into existing brownfield facilities presents specific engineering constraints. Most legacy halls were designed for power densities and cooling methodologies that are fundamentally challenged by modern AI accelerators. Successful retrofits depend on localised heat rejection strategies that minimise disruption to the central plant.
Rear-door heat exchangers stabilise rack temperatures, even under peak AI workloads, making them a practical server cooling option for GPUs. Where water access is limited, heat dissipation units can remove heat from AI servers and reinject it directly into the room air, eliminating the need to route facility water to the rack. Direct-to-chip liquid cooling at the rack level also allows for staged upgrades, preventing major central plant overhauls and making modernisation of smaller, constrained sites more feasible.
The logic of efficiency: the 7+3 framework
Sustainability in the AI era is governed by the trade-offs between energy and water consumption. Optimising power usage effectiveness requires a sophisticated understanding of approach temperatures, the temperature difference between the primary and secondary cooling loops. The sustainability and efficiency of a direct-to-chip liquid cooling system are driven by a holistic view of the architecture, from the chip to the chiller.
Seven core design factors determine energy and water consumption:
• Outdoor heat rejection: Air-cooled chillers with economisers save water compared with cooling towers, which consume heavily year-round.
• IT inlet fluid temperatures: Raising supply temperatures, for example, from 25 to 45°C, cuts compressor energy by 30 to 40% and water use by 40 to 60%.
• Rack power density: Higher AI-driven densities boost liquid heat capture, slightly improving plant efficiency.
• CDU type: Liquid-to-liquid CDUs are more efficient than liquid-to-air units, avoiding extra heat exchange cycles.
• Component selection: Efficient pumps and heat exchangers with low approach temperatures enable higher facility water temperatures, more economiser hours and less compressor energy.
• Coolant type: Propylene glycol at 25% concentration is standard for reliability. While deionised water offers better thermal capacity, it requires more complex chemical management.
• Heat reuse: Higher liquid cooling temperatures enable waste heat recovery for district heating or industrial use, improving the energy reuse factor.
Three operational practices complete the framework:
• Control scheme: Advanced systems use digital twins and dynamic pump control to match real-time loads, avoiding wasted energy.
• Maintenance: Regular coolant monitoring and filter cleaning prevent damage and reduce pump strain.
• Adaptive deployment: Modular, phased builds align infrastructure with evolving IT demand, avoiding stranded capacity and excess carbon.
Designing for AI requires mastering fluid dynamics and material science at scale. A physics-based architectural strategy, rather than an anecdotal rule of thumb, allows facility engineers to deploy direct-to-chip cooling with confidence.
| Tel: | +27 11 254 6400 |
| Email: | [email protected] |
| www: | www.se.com/za/en/ |
| Articles: | More information and articles about Schneider Electric South Africa |
© Technews Publishing (Pty) Ltd | All Rights Reserved
printer friendly version