Summary
- GLP has started a Foshan campus planned to exceed 200MW, with 40MW of IT load in the first phase.
- The facility is designed around a mix of air and liquid cooling, including liquid-cooled rack densities of up to 140kW.
- GLP is using prefabricated and modular construction to shorten delivery times for high-density AI infrastructure.
GLP has started development of a data centre in Foshan, Guangdong, planned to exceed 200MW and designed to support liquid-cooled rack densities of up to 140kW.
The first phase will deliver 40MW of IT load for an unnamed internet and cloud-services customer. GLP puts total investment in the development at RMB4bn and expects the campus to become one of the larger high-density computing projects in China’s Greater Bay Area.
The 140kW rack specification is the most technically notable element. That density is well above the range for which conventional enterprise data halls and many existing colocation facilities were designed.
At those loads, thermal management is no longer predominantly an airflow problem. Heat has to be captured much closer to processors and transferred through liquid circuits capable of carrying a large and continuous thermal load.
GLP says the Foshan design will combine air and liquid cooling rather than using a single cooling method throughout the facility. That allows different halls or workloads to operate at different densities while the highest-powered clusters use liquid systems.
Moving towards 140kW per rack changes more than the cooling equipment mounted around the servers. Pipework routes, coolant distribution units, pumps, heat exchangers, leak detection, controls, maintenance access, and heat-rejection plant all become part of the data hall design.
The electrical system has to change alongside it. GLP describes an A/B dual-circuit and dual-busbar architecture designed to separate different modules while retaining common infrastructure.
The company is also relying heavily on prefabrication. Core components are manufactured away from the site while civil and assembly work progresses in parallel.
GLP says the approach can reduce construction time substantially and support delivery of modules in as little as four months. That target should not be confused with construction of the full 200MW-plus campus, which will still require staged civil work, utility infrastructure, commissioning, customer hardware, and energisation.
Prefabricated mechanical and electrical systems are nevertheless becoming more valuable as AI facilities shift towards repeatable high-density blocks. Factory assembly can reduce onsite labour requirements and allow more equipment to be tested before installation.
The first 40MW phase is being built for an existing GLP customer. The company has not identified the tenant beyond describing it as a major internet and cloud-services business.
GLP says its wider Chinese data centre platform consists of around 20 facilities with approximately 1.4GW of IT capacity across operating and development sites.
The Foshan design provides a useful technical comparator for European operators because the underlying rack-density problem is not geographically specific. Accelerator hardware being deployed in China, Europe, and North America is driving similar requirements for power delivery and heat removal.
New-build facilities can design liquid loops, floor layouts, pipe routes, structural loads, and heat-rejection systems around those densities from the outset. Existing air-cooled data centres face a more difficult retrofit problem, particularly where electrical and mechanical plant was sized for racks consuming a fraction of 140kW.
Headline campus megawatts therefore provide only part of the capacity picture. How those megawatts are distributed across individual racks determines the mechanical system, electrical topology, maintenance regime, and usable density of the building.
GLP’s first 40MW phase will provide the initial operating test for the Foshan design before the campus grows towards its planned 200MW-plus scale.
