Singapore allocates 200MW to four data centre operators on Jurong Island
A capacity call with low-carbon strings attached shows how power, not land, now decides where data centres get built.
Singapore's Economic Development Board and Infocomm Media Development Authority have provisionally allocated 200MW of new data centre capacity to four operators, with Digital Realty, Equinix, Keppel Data Centres and ST Telemedia Global Data Centres each receiving 50MW, Data Center Dynamics reported on 21 August. The awards were made under the country's second Data Centre Call for Application, known as DC-CFA2, which opened in December 2025 and closed to applications in April. The agencies said more than 20 proposals were submitted by local and international players.
All four sites are to be developed at a low-carbon data centre park on Jurong Island, a reclaimed island south-west of the mainland that hosts Singapore's petroleum, chemical and energy industries. The park is being developed by JTC, the government agency responsible for industrial infrastructure, and the facilities are expected to draw on the island's low-carbon energy options. Under the terms of the allocation, at least half of each facility's capacity must be supplied by low-carbon sources such as biomethane, low-carbon ammonia or hydrogen. The operators have also committed to exceed the scheme's minimum sustainability requirements and to deploy liquid cooling.
Keppel said its allocation will support Keppel Data Centre Singapore 11, a S$1 billion project to be funded by Keppel Data Centre Fund III and co-investors. Digital Realty said its capacity will go into a new facility at the JTC park and put its total potential investment in Singapore at nearly S$7 billion. Development timelines were not disclosed.
Singapore has restricted new data centre construction since a 2019 moratorium. A first call in 2023 allocated 80MW across Equinix, GDS, Microsoft and an AirTrunk-ByteDance consortium. The EDB and IMDA said they will review the need for a further call in 18 to 24 months.
Why it matters
In land- and power-constrained markets, governments are becoming the gatekeepers of data centre growth, rationing grid capacity and attaching efficiency and fuel conditions to it. The Jurong Island model, where a low-carbon energy park is planned alongside the facilities, points toward siting decisions driven by energy supply chains rather than proximity to customers. Liquid cooling commitments in a tropical climate signal that high-density AI racks are the expected tenant. Other dense hubs facing similar constraints are likely to study how this allocation performs.
In land- and power-constrained markets, governments are becoming the gatekeepers of data centre growth, rationing grid capacity and attaching efficiency and fuel conditions to it.
Call opens (December 2025)
EDB and IMDA launch DC-CFA2 under the Green Data Centre Roadmap
Applications close (April 2026)
More than 20 proposals received from local and global operators
Provisional allocation (August 2026)
Digital Realty, Equinix, Keppel and STT GDC each receive 50MW, 200MW in total
Conditions attached
At least 50% low-carbon supply, liquid cooling, sustainability targets above the minimum
Next review
Agencies to reassess the need for another call in 18 to 24 months
What you can learn from this
- Power capacity, not floor space, is the scarce resource. A data centre's size is quoted in megawatts because every watt of IT load must be delivered by the grid, converted, distributed and then removed as heat. Utilities plan generation and transmission years ahead, so a sudden cluster of 50MW requests can exceed what a substation or feeder can carry. That is why regulators in dense markets allocate power in tranches and why the announcement counts capacity in MW rather than square metres.
- Power usage effectiveness explains why efficiency conditions matter. PUE is the ratio of total facility energy to the energy used by IT equipment; a value of 1.0 would mean no overhead at all. Cooling is usually the largest overhead, and in a hot, humid climate mechanical chillers run harder for more of the year. Requiring operators to exceed minimum efficiency standards is effectively a way to squeeze more useful compute out of the same scarce grid allocation.
- Liquid cooling exists because air has physical limits. Air can only carry so much heat per unit volume, so once a rack draws tens of kilowatts, fans and cold aisles struggle to keep silicon within its thermal envelope. Direct-to-chip cold plates and immersion systems move heat with a fluid that has far higher heat capacity, allowing much denser racks and warmer coolant loops. This in turn lets a facility use less chiller energy, which links the cooling commitment back to the efficiency requirement.
- Low-carbon fuel mandates shape the electrical design. Biomethane, hydrogen and low-carbon ammonia are not plugged in like a grid feed; they typically run through on-site or nearby generation, which changes how redundancy, fuel storage and switching are engineered. A facility designed for 50 percent low-carbon supply needs to handle two sources with different availability profiles and failure modes. This is why co-locating data centres with an energy park is attractive: the supply chain for the fuel is already next door.
- Staged calls for applications are a capacity-planning tool. By opening a call, collecting proposals and allocating a fixed total, an authority can compare projects on efficiency and sustainability rather than serving requests first-come, first-served. It also gives grid operators a known load to plan against, and gives applicants a clear set of criteria to design toward. The stated plan to review another call in 18 to 24 months shows the allocation is intended to be periodic rather than a one-off.
We teach this
How to use this in practice
- Calculate the electrical story of one rack. Take a typical AI server specification you can find online, note its rated power, multiply by servers per rack and then by an assumed PUE of 1.3 to see facility draw. Work out how many such racks 50MW supports. Done means a small spreadsheet where changing PUE or rack density visibly changes the number of racks a fixed allocation can host.
- Draw the power path from grid to chip. Sketch the chain: utility feed, transformer, switchgear, UPS, power distribution unit, rack PDU, server power supply, and add a second branch for an on-site low-carbon generator. Mark where redundancy exists, where a single failure would drop load, and roughly how much energy each conversion stage loses as heat. Done means a one-page diagram with every conversion stage labelled and at least one note on where the low-carbon source joins the path and how the switch between sources would happen.
- Measure PUE-style overhead on your own hardware. If you have a smart plug or a machine that reports power draw, log the idle and full-load consumption of a computer over an hour while also noting any fan or air-conditioning load in the room. Compute the ratio of total room energy to computer energy as a rough PUE analogue. Done means a short table with the two figures and a sentence on which overhead dominated.
- Compare air and liquid cooling on paper. Look up the specific heat capacity of air and water and estimate how much of each is needed to remove 1kW of heat with a 10 degree temperature rise. Then write three sentences on why the difference matters at 100kW per rack, and one more on what a leak or pump failure would mean for the servers in that rack. Done means a worked calculation you could explain to a colleague without notes, plus a short list of the new failure modes liquid cooling introduces and how each might be detected.
Sources
- Singapore gov't allocates 200MW of power for four data center developers on Jurong Island — Data Center Dynamics
Our reporting is an original summary; full coverage is at the links above.
Don't just read about it — build it.
Square 1 teaches the skills behind the headlines, with every line of your work graded by AI. Find your starting point in 3 minutes.
Get your free skill report