Summary
- FRA6 provides 6,600 sq m of colocation space and supports direct-to-chip liquid cooling.
- Liquid-cooled deployments introduce CDUs, pipework, controls, monitoring, and separate facility and IT liquid circuits.
- Operators increasingly need to accommodate liquid-cooled and conventional equipment in the same facility portfolio.
Equinix is using its FRA6 data centre in Frankfurt to support direct-to-chip liquid cooling alongside more conventional thermal systems, illustrating the changes required below the rack as facilities accommodate higher-density compute.
The purpose-built four-storey facility provides around 6,600 sq m of colocation space. Equinix lists N+2 UPS redundancy, N+1 cooling redundancy, and direct-to-chip liquid cooling among the site’s specifications.
The engineering difference becomes significant once heat is moved into liquid rather than being handled entirely through server airflow.
Direct-to-chip systems place cold plates on high-heat components such as CPUs and GPUs. Coolant carries heat away from those components into a liquid circuit, reducing the portion that must be removed through conventional air cooling.
That does not make the facility-side cooling system disappear. The liquid needs to be circulated, monitored, conditioned, and ultimately connected to the data centre’s wider heat-rejection infrastructure.
A coolant distribution unit typically provides the interface between the facility water system and the secondary circuit feeding IT equipment. Keeping the circuits separated allows operators to manage flow, temperature, pressure, and water chemistry without passing the facility loop directly through customer hardware.
That creates new infrastructure requirements inside and around the white space: distribution pipework, CDUs, pumps, valves, leak detection, sensors, controls, maintenance access, and suitable connections back to the mechanical plant.
Liquid cooling therefore cannot be treated simply as a different server specification. Where a customer needs coolant delivered to the rack, the facility has to be designed or modified to support it.
Air cooling also remains part of many direct-to-chip deployments. CPUs and GPUs may account for a large share of the heat load, but memory, storage, power supplies, network equipment, and other components can still release heat into the server chassis.
The result is often a hybrid thermal architecture rather than a complete change from air to liquid.
That is particularly relevant to colocation operators, whose data halls may have to support several generations of equipment at once. Conventional enterprise racks, high-density air-cooled systems, rear-door heat exchangers, and direct-to-chip installations can coexist inside the same metro estate.
The retrofit question is correspondingly important. Rear-door heat exchangers can often be added closer to the rack with less intervention in the facility water system, while direct-to-chip installations require a defined liquid-distribution path into the IT environment.
New data halls can reserve space and routes for that infrastructure from the start. Existing rooms may need pipework changes, additional CDUs, control integration, floor or ceiling routes, and alterations to the mechanical system before comparable loads can be supported.
The transition creates a capacity-planning problem. Building liquid infrastructure too slowly risks leaving a facility unable to accommodate customers deploying high-density AI and HPC systems. Building too much before demand materialises can leave expensive mechanical capacity underused.
FRA6 shows how operators are managing the transition by maintaining several thermal options rather than assuming one cooling method will immediately replace another.
For future high-density deployments, the competitive difference will increasingly depend on more than whether a facility can advertise liquid cooling. The practical questions are how coolant reaches the rack, how the loops are isolated and monitored, how failures are contained, and how the additional heat is rejected efficiently once it leaves the IT equipment.
