of operators run any direct liquid cooling, and among those nearly half use it on under 10% of racks. It stays surgical, reserved for AI and HPC islands. Surveys put adoption at 35–40% by 2026.
Why it stays surgicalPut to the people deciding the answer – the 90 leaders across ten markets who power, fund, permit and build it.
A continental president of data centres at a hyperscaler – whose board is weighing half a trillion dollars of cloud orders – put two questions to the desk: how widely is direct liquid cooling actually used today, and where are the high-density projects landing. The short read, on the public record. Open any highlighted panel to go deeper.
of operators run any direct liquid cooling, and among those nearly half use it on under 10% of racks. It stays surgical, reserved for AI and HPC islands. Surveys put adoption at 35–40% by 2026.
Why it stays surgicalAverage racks still sit near 8–12 kW. The head of the curve is what moves the cooling decision.
A new AI build-out across Sumaré and Vinhedo, designed from the outset for liquid cooling and densities up to 1 MW per rack – a clear marker of how far new campuses are now pushed.
Open the caseDensity is rising everywhere; liquid is still surgical. Above 50–100 kW per rack, air stops being a real choice. The hedge most operators are taking is the hybrid campus – traditional halls alongside liquid-ready AI pods – timed to the local grid rather than committed all at once.