GLP's RMB4 billion Foshan data center bet starts with 40MW

The 40MW of IT capacity now under construction at GLP's hyperscale data center campus in Foshan, Guangdong Province, is the visible first slice of a RMB4 billion, 200MW plan that IREI reports would rank among the largest data center projects in the Greater Bay Area. GLP describes the region as seeing surging demand for AI computing infrastructure, and for real estate investors the budget and the megawatts are only half the story: the other half is the pace at which the remaining 160MW follows.

The distance between the 40MW first phase and the full 200MW is a capital program built around a tenant, not simply construction staging. IREI reports the campus is purpose-built for a leading internet and cloud services company that is expanding an existing partnership with GLP; the customer is unnamed, but the partnership is the underwriting anchor. The developer is adding capacity for a counterparty whose growth it can measure from the inside, rather than building a shell on the chance demand appears.

What will determine how Foshan performs in a private portfolio is how GLP structures that capital program, and the coverage is silent on whether GLP will carry the completed campus on its balance sheet, sell phases to co-investors once leased, or hold the project inside a dedicated digital-infrastructure vehicle. Each route produces a different risk and fee profile, and all of them become financeable only because the tenant relationship exists. Taken at face value, RMB4 billion across 200MW works out to RMB20 million per MW if the budget is meant to cover full build-out; without the breakdown to confirm that allocation, investors should treat the figure as scale rather than contract.

What remains unquantified is how much of the eventual capacity the expanding hyperscaler has already committed to take. Phase one, at 40MW, is in motion; the remaining 160MW is the option. IREI's article provides no timetable for the later phases, an absence that suggests GLP is preserving the freedom to wait until the customer's AI workloads pull additional construction forward. For allocators, the distinction between an option and a commitment is the difference between lending against a development pipeline and lending against stabilized data center cash flow.

Foshan also advances the sector argument this publication has pressed through the current cycle: the scarce assets in the AI build-out are power and land, not compute. GLP says it is combining its digital infrastructure and new energy platforms on the campus, an approach that acknowledges power as the controlling input; a landlord that procures its own energy supply has a better chance of holding margins when computing loads begin drawing on the grid, while one that buys power at market rates simply inherits the price move. The staged build-out gives the land and the power relationship time to appreciate rather than forcing all of the economics to be created in one construction push.

The risks cut the other way. Concentration is the visible one: a campus purpose-built around one unnamed cloud company ties the return to one expansion plan, and hyperscale customers can defer capacity when their own forecasts wobble, leaving a developer holding a large entitled site that was always meant for one tenant. The quieter risk is cost: RMB4 billion is a planned sum, not a closed contract, and the electrical and cooling systems inside a data center shell are precisely where construction inflation lands. Phasing protects against demand timing; it does not protect later phases from being built on prices that no longer fit the original underwriting.

The first phase is the easy fact to report; the second phase is the one to watch. IREI's article carries no schedule for the remaining 160MW, so investors will measure Foshan by the pace at which GLP announces the next tranche. A prompt follow-on confirms that the RMB4 billion envelope is turning into a real, staged capital program; a pause says the developer is content to hold land and power in the Greater Bay Area while it waits for the customer's expansion to catch up. Those are two different assets, and the market will be pricing the option before the next phase appears.