ISO 9001 certified manufacturer · Shenzhen, China
Singapore — 1.8 MW Liquid Cooling Deployment for an AI Lab

AI & HPC · 2025

Singapore — 1.8 MW Liquid Cooling Deployment for an AI Lab

Three containerized CDU units provide N+1 heat rejection for 96 liquid-cooled GPU racks at 45 °C facility water.

1.8 MWInstalled capacity
SingaporeLocation
2025Handover
AI & HPCIndustry
1.8 MW Heat rejected
1.14 Measured PUE
45 °C Facility water
N+1 Pump redundancy

01 — The challenge

The lab had committed to a GPU cluster drawing 1.8 MW in a building whose cooling plant was sized for 900 kW of air cooling. The lease had eight years to run, so a new chiller plant was not an option, and the facility water available was 45 °C — too warm for conventional CRAC equipment.

02 — The solution

We deployed three LiquidCube CDU containers, each with two 500 kW CDUs in N+1, connected to the building's condenser water loop and to 32 direct-to-chip racks per container. Secondary loops run at 40 °C supply, which keeps chip temperatures inside vendor limits while allowing the existing plant to reject the heat.

Leak detection tape runs the length of every manifold, and the control system isolates a faulty secondary circuit automatically without interrupting the cluster.

03 — The result

The cluster runs at a measured PUE of 1.14 against the building's previous 1.62. Rack density rose from 8 kW to 45 kW per rack in the same floor area, and the existing chiller plant was retained without modification.

Singapore — 1.8 MW Liquid Cooling Deployment for an AI Lab