A Nasscom analysis examines how modern liquid-cooled AI data centers are handling thermal loads of up to 132 kilowatts per rack, a density level that renders conventional air cooling ineffective. The report details direct liquid cooling configurations, coolant distribution units, and rear-door heat exchangers being deployed by hyperscalers and colocation operators to manage GPU cluster heat output. Engineers cited in the analysis note that managing coolant temperature differentials and leak prevention remain the primary operational challenges at these power densities. Demand for liquid cooling infrastructure is expected to grow in parallel with continued GPU cluster buildouts through 2027.