Why Air Cooling Runs Out at AI Densities
Conventional air cooling with hot/cold aisle containment becomes impractical above roughly 15-20kW per rack — the airflow volume required starts demanding unreasonably large CRAH capacity and floor space. AI training and inference racks in UAE data halls increasingly run 30-100kW+, pushing facilities toward direct-to-chip or full immersion liquid cooling.
Direct-to-Chip vs Immersion Cooling
Direct-to-chip cooling circulates coolant through cold plates mounted on CPUs/GPUs, removing heat at the source while the rest of the rack still uses air — a moderate design change from conventional layouts. Full immersion cooling submerges entire servers in dielectric fluid, removing airflow requirements almost entirely but requiring a fundamentally different rack, containment and fire-suppression design.
The UAE Climate Factor
UAE ambient temperatures affect the design of the cooling loop's heat rejection side — dry coolers and cooling towers sized for temperate climates need significant derating for UAE summer conditions, and design should account for the actual local wet-bulb and dry-bulb extremes rather than generic manufacturer datasheet figures calculated for other climates.
Impact on ELV and Fire Detection Design
Liquid cooling changes fire detection design — traditional aspirating smoke detection assumptions about airflow patterns don't hold in a liquid-cooled hall with dramatically reduced air movement, and leak-detection systems for the coolant loop become a new ELV scope item that didn't exist in air-cooled designs. ASDV designs fire and leak detection for liquid-cooled halls as a distinct scope from conventional data centre fire alarm design, not a simple extension of it.