Liquid cooling removes heat from servers using a circulating fluid rather than air. Coolant is pumped to each rack, absorbs heat from high-density components such as GPUs and CPUs, then returns to a Coolant Distribution Unit (CDU) where that heat is transferred to the facility's broader cooling infrastructure. The IT-side loop is closed and isolated from the building's main water supply.

Modern AI processors generate far more heat than air cooling can remove. Air cooling reaches its practical limit at around 30kW per rack. High-performance AI deployments routinely operate above 100kW per rack, with next-generation platforms expected to exceed 200kW. Liquid is approximately 60 times more efficient at transferring heat than air, making it the only viable option for dense AI infrastructure at scale.

Direct-to-chip cooling routes coolant through a cold plate mounted directly on the surface of high-heat components, typically GPUs and CPUs. Heat is removed at the source rather than from the air surrounding the rack. This allows processors to run within their optimal temperature range, supports higher rack densities, and significantly reduces the load on room-level air conditioning systems.

Not necessarily. The coolant loop from the rack to the CDU is a closed system that does not consume water. Whether the facility uses water depends on how heat is rejected at site level, through cooling towers, dry coolers or chillers. Some facilities running direct-to-chip liquid cooling operate with no evaporative water consumption at all. The right question is not whether liquid cooling uses water, but how heat is ultimately rejected at your specific site.

Air cooling uses fans and chilled air to remove heat from IT equipment. It is well understood and widely deployed, but becomes increasingly inefficient above approximately 30kW per rack. Liquid cooling uses fluid, which transfers heat approximately 60 times more efficiently than air at equivalent mass flow rates. For AI, HPC and other high-density compute environments, liquid cooling has become the standard approach because the heat loads involved simply cannot be managed any other way.

Air-cooled infrastructure reaches its practical limit at around 30kW per rack. NEXTDC is designing for rack densities up to 2MW as next-generation GPU hardware scales beyond what air can handle. Achievable density for a specific deployment depends on the hardware platform, power availability and site conditions. Engage early, design decisions at this stage have the most impact.

NEXTDC operates liquid cooling infrastructure across six data centres in Australia, including M3 Melbourne, S3 Sydney, M2 Melbourne, S4 Sydney, S6 Sydney and A1 Adelaide. NEXTDC is expanding into APAC with KL1 Kuala Lumpur. The S3 Sydney facility hosts a live demonstration suite where customers can observe liquid cooling operating in a production environment. Additional deployments are underway across APAC, with future sites designed for liquid cooling from day one.

NEXTDC operates a guided demonstration suite at S3 in Sydney where sessions are tailored to each visitor's hardware platform, target rack density and deployment context. The greatest decisions in any AI infrastructure project are made during the design phase, before hardware is ordered and rack layouts are finalised. Engaging early gives teams the most flexibility. Contact NEXTDC's AI and HPC Solutions Director to arrange a session.

Direct-to-chip liquid cooling routes coolant through a cold plate mounted on each processor and works with standard server designs from leading GPU manufacturers. Immersion cooling submerges entire servers in dielectric fluid and requires purpose-built hardware that most organisations are not currently deploying. Direct-to-chip is the dominant approach for enterprise and hyperscale AI deployments today. NEXTDC's liquid cooling infrastructure is built around direct-to-chip technology, enabling customers to deploy standard AI hardware at high-density across liquid cooled colocation facilities in Sydney, Melbourne and Adelaide and expanding into APAC.