Compute at the landing point of the cables
Singapore should not, on paper, be a data-centre superpower. It has no domestic energy resources to speak of, imports almost all its natural gas, and measures 735 square kilometres — smaller than New York City. Yet the island carries more than 70 operational data centres and about 1.4 gigawatts (GW) of live capacity, the densest concentration of compute in Southeast Asia and one of the densest anywhere. The reason is not power or land. It is position: Singapore sits at the landing point of more than 20 subsea cables, offers political stability, an English-language legal system and a deep pool of engineers, and sells the one thing hyperscalers value above cheap electricity — latency to hundreds of millions of Southeast Asian users and certainty that the lights, and the law, stay on.
That advantage created a problem the government has spent years managing. By 2019, data centres were consuming an estimated 7% of national electricity and heading higher. For a country that imports nearly all its energy and has committed to net zero by 2050, an open-ended industrial load with double-digit annual growth was untenable. So Singapore did something no other major hub had dared: it stopped issuing approvals altogether.
Why the 2019 moratorium happened
The moratorium, imposed quietly in 2019, was arithmetic, not ideology. Data centres are among the most power-hungry buildings on earth — a single large facility can draw 100 megawatts (MW), the equivalent of a small town — and in the tropics the cooling penalty is brutal. An operator in Stockholm can pull cold air off the street for much of the year; an operator in Singapore, where the mercury sits near 30°C and humidity above 80% year-round, must chill and dehumidify continuously. That makes Singaporean facilities inherently harder to run efficiently, and every inefficient watt lands on a national grid with no hydro, no nuclear and limited room for solar.
The freeze bought time but created a scarcity premium. With supply capped and demand from cloud migration and, later, artificial intelligence surging, Singapore became one of the tightest and most expensive colocation markets in the world. The pressure valve was regional — capacity spilled across the causeway to Johor in Malaysia and to Batam in Indonesia, both of which have since attracted billions in data-centre investment precisely because Singapore would not build. The government’s calculation was that it would rather export the megawatts than the value: keep the low-latency, high-value workloads onshore and let the power-hungry bulk go next door.
The Green Data Centre Roadmap and efficiency-gated growth
The reopening, when it came, was conditional. In May 2024 the Infocomm Media Development Authority (IMDA) published the Green Data Centre Roadmap, which set out to unlock at least 300 MW of additional capacity in the near term — with more available to operators who bring their own renewable energy. Crucially, much of that 300 MW is meant to come not from new land but from squeezing more compute out of existing buildings: retrofitting cooling, raising operating temperatures and improving utilisation.
The gate is efficiency. The roadmap targets a Power Usage Effectiveness (PUE) of 1.3 or below at high IT load — an aggressive figure in a tropical climate. PUE measures total facility energy divided by IT energy; a PUE of 1.3 means 30% overhead on top of the computing itself, most of it cooling. Hitting 1.3 in Singapore’s heat is an engineering achievement, and it is now the price of entry.
That principle carried into the allocation rounds. The pilot Data Centre Call for Application in 2022 handed out just 80 MW across four winners, chosen on sustainability and economic contribution rather than willingness to pay. The successor round, DC-CFA2, was announced on 1 December 2025 with a window closing 31 March 2026. It offers at least 200 MW, with the possibility of further allocations for operators advancing genuinely novel green-energy pathways, and it sets a floor of green-sourced power — biomethane, low-carbon ammonia, hydrogen, fuel cells with carbon capture, or on-site solar. Growth is no longer rationed by land or moratorium. It is rationed by how clean and how efficient an operator can prove itself to be.
Who operates here
The tenant roll reads like a census of global compute. Amazon Web Services, Microsoft Azure, Google Cloud and Meta all run infrastructure on the island, anchoring Singapore as their Southeast Asian hub. AWS alone has committed a further US$9 billion to Singapore cloud and data-centre infrastructure — part of a broader S$12 billion package including AI programmes — a signal that the hyperscalers intend to grow onshore capacity despite the constraints, not flee them.
Beneath the hyperscalers sits a colocation layer that builds and leases the physical halls: Equinix, ST Telemedia Global Data Centres, Digital Realty, AirTrunk, Keppel Data Centres, GDS and others. Southeast Asia’s colocation capacity is dominated by Singapore, which accounts for the largest single share of the region’s live load. Scarcity has made incumbents’ existing megawatts extraordinarily valuable — a hall approved before 2019 is a scarce, appreciating asset — which is why so much current activity is retrofit and densification rather than greenfield construction.
AI’s power demand
Artificial intelligence has changed the physics. A traditional server rack in a Singapore hall might draw 5 to 10 kilowatts (kW); an AI training rack packed with GPUs can draw 40 kW, 80 kW or more, and the newest designs push past 100 kW per rack. Air cooling cannot remove that much heat from that small a footprint, which is forcing the market toward liquid cooling — direct-to-chip cold plates and, increasingly, immersion — as a practical necessity rather than a green flourish. AI has simultaneously intensified Singapore’s core dilemma: demand for compute is exploding at the exact moment the grid can least afford unconstrained load, which is why the efficiency gate has hardened rather than loosened even as the moratorium lifts.
The tropical cooling challenge and sustainability
Everything in Singapore’s data-centre economics runs through cooling. In a temperate market cooling might be 20–30% of energy use; in the tropics it can be higher, and it is the single biggest lever on PUE. The response has been a technology race: raising chilled-water temperatures, deploying liquid cooling, proving that halls can run reliably at 26–27°C rather than the traditional 22°C, and exploring floating and offshore concepts to escape land and use seawater for heat rejection. Green power is the other half: on-site solar is limited by roof area, so operators are turning to imported low-carbon electricity — Singapore plans to import up to 6 GW of clean power by 2035 — and to the alternative fuels named in DC-CFA2. In Singapore, sustainability is not corporate garnish; it is the regulatory precondition for building at all.
The industrial supply chain it feeds
For Singapore’s industrial economy, a data centre is a magnificent customer. Each facility is a dense concentration of mechanical and electrical (M&E) engineering, and the shift to AI and liquid cooling deepens that demand. Precision cooling — computer-room air handlers, chilled-water plant, coolant distribution units and rear-door heat exchangers — is a growth line in itself, and direct-to-chip liquid cooling brings a new supply chain of manifolds, quick-disconnect fittings and leak detection. Power distribution is the other pillar: switchgear, uninterruptible power supplies, busway, transformers and standby generation. Around them sit structured cabling and containment, and behind all of it a specialised construction and fit-out sector that builds shell-and-core, then commissions the M&E to exacting standards. A 100 MW facility is, for the firms Singapore Industry Index tracks, years of contracted work.
Regional context and outlook
The regional picture is a deliberate division of labour. Johor’s data-centre clusters and Indonesia’s Batam have absorbed the hyperscale bulk that Singapore chose not to house, drawing tens of billions in committed investment. Singapore has not lost that business so much as re-sorted it: latency-sensitive, high-value and AI-inference workloads stay onshore, while training and storage at scale sit across the border, often operated by the same companies. The verdict for the decade ahead is that Singapore remains the region’s control tower — dense, expensive, tightly regulated and growing only as fast as it can prove it is clean. The 200 MW of DC-CFA2, the 300 MW of the roadmap and the hyperscalers’ billions all point the same way: a market where the binding constraint is no longer whether to build, but how efficiently and how green. For Singapore’s cooling, power, cabling and construction firms, that is the most durable demand signal in the industrial economy.