Transforming data centres: The rise of sustainable liquid cooling for AI

TODAY, single AI query uses roughly ten times the electricity of a typical Internet search, and demand is climbing at lightning speed.

- STEVEN SANTINI

In fact, projections suggest data centres could account for approximately 3% of total global electricity consumption by 2030, nearly doubling their current share and, significantly, growing four times faster than the growth of total electricity consumption from all other sectors.

The result, a massive amount of heat generation which leads to a thermal challenges legacy data centres simply werent designed to handle. Traditional air cooling, the longtime workhorse of data centres, is being pushed to its practical limits by high-performance, high-density racks.

To unlock Al's full potential, data centres must move beyond the status quo and embrace advanced, sustainable liquid cooling. The move to high-density, liquid-cooled infrastructure

As mentioned, AI workloads are breaking the cooling mould and pushing rack power densities to new norms. Current rack densities can range from 40 kW to well over 100 kW, which is impractical to manage with air cooling.

Also, demands continue to climb rapidly with each new generation of GPU-accelerated servers.

For example, a fully populated NVIDIA-based GPU racks draws around 132 kW, and that number is set to rise.

The next generation, expected within a year, is projected to reach 240 kW per rack, and the industry is already preparing for future power densities of 1 MW per rack.

Unlike standard Central Processing Units (CPUs), the Graphics Processing Units (GPUs) and other AI accelerators that power these AI models generate intense, concentrated heat loads that require targeted, highly efficient cooling to maintain optimal performance.