Data Centres in Singapore: The 2026 Enterprise Buyer's Guide

Key Takeaways

• Supply is policy-rationed, not market-driven. Singapore approves new data centre capacity through competitive government calls with sustainability conditions. Pipeline exists — but it arrives slowly and much of it is pre-committed before opening.
• Vacancy below ~1.4% changes how you buy. Capacity planning must run 12–18 months ahead of need, and renewal negotiations start from a position of operator strength.
• You are buying power and cooling, not space. Contracts, pricing and evaluation all revolve around the kilowatt. Floor area is a legacy metric.
• AI density is the new dividing line. GPU racks at 40–100+ kW need liquid cooling that only newer or retrofitted facilities deliver. Verify engineering reality, not marketing claims.
• Interconnection is Singapore's real moat. Submarine cables, carrier density and cloud on-ramps — not cheap capacity — are why workloads stay here despite the price premium.
• Regulation is tightening on operators. The draft Digital Infrastructure Act would license third-party facilities of 10 MW and above and impose resilience and PUE obligations — a higher assured baseline for buyers, with compliance costs likely priced in.
• Johor and Batam are now part of the same buying decision. The practical question in 2026 is not "which Singapore facility?" but "which workloads justify Singapore at all?"
• Certifications need scope-checking. ISO 27001, SOC 2, Uptime Tiers and MTCS are table stakes among major operators — but certificates are issued per facility and per scope, so verify the building you will actually occupy.

Quick Facts

What You Are Actually Buying

A data centre is a purpose-built facility that provides conditioned power, cooling, physical security and network connectivity for IT equipment. That definition matters less to a buyer than the commercial forms it takes, because "we need data centre capacity" resolves into four quite different purchases:

• Retail colocation — you rent space by the rack or cage (typically a few kW to a few hundred kW), the operator runs the building, and you run your hardware. Sold on standard terms with service-level agreements per rack or per kW.
• Wholesale colocation — you lease a data hall or dedicated suite (hundreds of kW to tens of MW), usually on multi-year terms with ramp schedules. The economics resemble industrial leasing more than IT procurement; in Singapore, wholesale now generates the majority of colocation revenue.
• Hyperscale / build-to-suit — cloud platforms and the largest internet companies lease entire buildings or commission dedicated campuses. Relevant to most enterprises only indirectly: this is where your cloud workloads physically live.
• Cloud regions — consuming AWS, Microsoft Azure or Google Cloud in their Singapore regions is, physically, consuming these same facilities with the hardware layer abstracted away. The colocation-versus-cloud question is therefore a workload placement decision, not a facility decision — covered below.

The unit of account across all of them is the kilowatt of IT load, not the square metre. Singapore pricing, contracts and even government policy are denominated in power. A useful mental model for any Singapore data centre conversation: you are buying a guaranteed, conditioned, cooled electricity supply with a building around it — and, at carrier-neutral sites, membership of an interconnection ecosystem. For a primer on how modern facilities are engineered, see our knowledge-base explainers on AI data centres and liquid cooling.

Why Singapore — and Why Not

Singapore's position rests on infrastructure fundamentals that are hard to replicate. It is one of the world's densest submarine cable hubs, with dozens of international systems landing on the island and more committed under the national push to expand landing capacity (see our submarine cables guide). That cable density attracts carriers; carriers attract interconnection platforms; interconnection attracts cloud on-ramps, exchanges and financial trading infrastructure. The result is a network-effect ecosystem where round-trip latency to much of Southeast Asia is low and counterparties are a cross-connect away. Layer on political stability, rule of law, a deep skilled workforce and the regional headquarters concentration of banks and multinationals, and demand persistently outruns what the island can physically host.

The honest counterweights are equally structural:

• Cost. Singapore prices at or near the top of Asia-Pacific for colocation, driven by rationed supply, expensive industrial power (~US$0.21–0.24/kWh in early 2025) and scarce land.
• Availability. With vacancy under ~1.4%, the capacity you want may simply not exist on your timeline, at any price. Wholesale space is often pre-committed a year or more before a building opens.
• Energy and climate constraints. Data centres already consume about 7% of national electricity, heading toward a projected 12% by 2030 in a country with limited renewable options. That is precisely why the state rations growth — and why sustainability conditions keep tightening.
• Concentration risk. A single small territory hosting your primary and secondary sites provides limited geographic separation for disaster recovery. Many resilience architectures now pair Singapore with Johor or Batam for this reason alone.

Who should anchor capacity in Singapore: organisations whose workloads monetise low latency to Southeast Asia, need dense interconnection (finance, trading, SaaS, content, telecoms), or carry regulatory and data-governance requirements best satisfied in Singapore's legal environment. Who should think twice: buyers whose workloads are latency-tolerant bulk compute — batch analytics, AI model training, archive storage — for which the Singapore premium buys little that Johor or Batam cannot provide.

Singapore Market Overview

Singapore operates more than 1.4 GW of data centre capacity — the fifth market in Asia-Pacific to pass the 1 GW mark, achieved on a fraction of the land area of Tokyo, Sydney, Beijing or Shanghai. Curated industry databases count roughly 66 operational colocation facilities from about 46 providers; broader tallies that include telco rooms, edge sites and announced projects run from 80 to over 130, so treat facility counts as methodology-dependent. Market researchers estimate the market at around US$4.3 billion in 2025, growing at roughly 5% annually — modest by regional standards, because growth here is capped by policy rather than demand.

Structurally, the market has tilted toward wholesale: large-block leasing accounted for an estimated 58% of colocation revenue in 2025 and is projected to keep growing as AI and cloud demand concentrates in bigger deployments. For retail buyers this matters because operators increasingly optimise new buildings for wholesale economics, keeping small-footprint supply tight.

Where the facilities are

Data centres cluster in industrial estates with substation capacity and fibre density:

• Loyang / Changi (east) — the heavyweight campus cluster: AirTrunk's SGP1, Digital Realty's SIN11 and SIN12, STT GDC Loyang and AWS sites.
• Tai Seng / Defu / Serangoon (north-east) — STT GDC's ~40 MW Defu campus, Global Switch Tai Seng, Equinix SG3, Keppel and NTT facilities.
• Jurong / Tuas (west) — Nxera's DC Tuas, Google's Jurong West campus, Equinix's SG5 and upcoming SG6 at Jalan Tukang, and the future ~700 MW Jurong Island low-carbon park.
• Woodlands (north) — Microsoft's Azure facilities and Global Switch Woodlands.
• Central / one-north / Ayer Rajah — Equinix SG1, carrier hotels and legacy interconnection-dense sites.

Cluster choice is not cosmetic: dual-site architectures should sit in different clusters on separate grid infrastructure, and network designs should confirm diverse fibre paths between them — see our structured cabling guide for the intra-facility layer.

Recent market signals

The clearest read on 2026 conditions is what happens when capacity opens. Nxera's 58 MW DC Tuas — Singapore's highest-power-density multi-tenant facility, with the country's largest direct-to-chip liquid cooling deployment and integration with a submarine cable landing station — opened in February 2026 with more than 90% of capacity already committed. Equinix's US$260 million, 20 MW SG6 will not open until early 2027 and serves a queue of AI-driven demand. New supply, in short, is absorbed before it exists.

Regulation & the Capacity Queue

No other major data centre market is as explicitly state-managed as Singapore's. Understanding the policy machinery is essential to forecasting supply — and therefore pricing and availability — over any multi-year contract term.

Three implications follow. First, supply arrives slowly and pre-sold — a 200 MW award in 2026 becomes occupiable capacity years later, much of it committed to anchor tenants during construction. Second, sustainability conditions are contractual reality: operators pass their PUE and green-energy obligations through to customers as density requirements, temperature setpoints and, increasingly, renewable energy certificate arrangements. Third, the regulatory floor under operators is rising — the Digital Infrastructure Act would give IMDA licensing, codes of practice and enforcement powers over major facilities, which is good news for buyers' assurance and one more input to cost.

How Enterprise Buyers Should Evaluate Facilities

Evaluation should be structured around what actually differentiates facilities in a supply-constrained market. Six dimensions cover most of the decision:

Questions that separate serious answers from brochures

Deployment Models Compared

Most Singapore enterprises land on a hybrid: a colocation anchor for owned infrastructure, network gear and regulated systems, cross-connected to cloud on-ramps for everything elastic. The economics tip toward colocation as utilisation rises and toward cloud as variability rises; the compliance calculus tips toward colocation where auditors need physical and contractual specificity. Our AI computing guide and LLM buyer's guide cover the GPU-placement version of this decision.

Vendor Landscape

The Singapore market divides into four tiers with different buying motions. Portfolio figures below are operator-disclosed or industry-reported and should be re-verified at engagement time.

Global carrier-neutral platforms

Equinix operates five facilities (SG1–SG5) with the 20 MW, nine-storey SG6 at Jalan Tukang due in early 2027 on a US$260 million investment. Its strength is the interconnection ecosystem — the densest concentration of carriers, clouds and counterparties in Singapore, with 100% renewable energy coverage claimed since 2020. The trade-offs: premium pricing, a retail-first model that suits network and interconnection nodes better than bulk compute, and space that is chronically tight. Digital Realty runs three sites (SIN10, SIN11, SIN12 in Loyang), with SIN12 offering up to 50 MW of critical power; it is stronger than Equinix in wholesale blocks while still offering retail and interconnection. Both are publicly listed with deep financial standing. For a detailed head-to-head, see our Equinix vs Digital Realty comparison.

Singapore-headquartered regionals

ST Telemedia Global Data Centres (STT GDC), part of Temasek-linked ST Telemedia, operates one of the island's largest portfolios across its ~40 MW Defu campus, Tai Seng and Loyang — carrier-neutral, strong in both retail and wholesale, with deep local enterprise and government relationships. Keppel Data Centres (with listed Keppel DC REIT as an ownership vehicle) runs an island-wide portfolio spanning Serangoon, Tampines, Genting Lane and beyond; its REIT structure makes it unusually transparent financially. Nxera, Singtel's data centre arm, reached 120 MW of Singapore capacity with February 2026's DC Tuas — 58 MW, AI-ready, the country's highest power density and largest direct-to-chip liquid cooling deployment, integrated with a cable landing station. The caveat is availability: DC Tuas opened more than 90% committed. These operators' strengths are local depth, regulatory familiarity and (for Nxera) telco integration; their interconnection ecosystems are thinner than Equinix's.

Hyperscale wholesale developers

AirTrunk (Blackstone-owned) operates the 78+ MW SGP1 campus in Loyang, designed to a low ~1.25 PUE, and holds a DC-CFA award with ByteDance for further capacity. Princeton Digital Group operates SG1 and SG3 under a developer-operator model spanning Asia. Global Switch runs long-established Tai Seng and Woodlands facilities. These suit buyers taking hundreds of kilowatts to multiple megawatts on multi-year terms; they are not built for small-footprint retail, and ecosystem density varies by campus.

Hyperscalers and everyone else

AWS, Microsoft and Google operate self-built and leased capacity — AWS has committed a further S$12 billion to its Singapore infrastructure through 2028 (taking its total planned investment past S$23 billion), Microsoft builds in Woodlands and won DC-CFA capacity, and Google's campus sits in Jurong West. Enterprises consume these through cloud services rather than leases. Meta, Alibaba Cloud and Oracle maintain a mix of self-built and colocated capacity. A long tail matters for specific needs: NTT (Serangoon), Iron Mountain, BDx, 1-Net (Mediacorp-linked), CyrusOne and specialist operators serve niches from media to government-adjacent workloads. On the build side, an experienced contractor ecosystem — AECOM on design, ABB on power systems, and main contractors including Fortis Construction, Takenaka, Kajima, Boustead Projects, Dragages and Gammon — delivers most projects, relevant mainly to buyers commissioning build-to-suit capacity.

Operator Comparison

Pricing Overview

Singapore data centre pricing is quote-driven; published rate cards barely exist above small retail deployments. The structure, however, is consistent, and 2026 market benchmarks give usable planning ranges — treat every figure below as indicative, not quoted.

Three structural points. First, power drives everything: your monthly bill scales with committed kW and the pass-through of some of Asia's most expensive industrial electricity, so hardware efficiency directly cuts facility cost. Second, wholesale contracts are leases in substance — expect multi-year terms (commonly 5–10 years), ramp schedules, take-or-pay commitments on reserved capacity and annual escalators; the negotiation levers are term length, ramp pacing, expansion rights and renewal caps. Third, the scarcity premium is real and durable: with vacancy under ~1.4% and supply arriving pre-committed, operators hold pricing power, and renewal uplift caps negotiated at signing are worth more than headline rate discounts. Budget additionally for migration (typically S$50k–500k+ depending on scale), structured cabling fit-out, and connectivity — and note that no government grant meaningfully subsidises enterprise colocation, though adjacent digitalisation costs may qualify under schemes covered in our IT grants guide.

Compliance & Security Considerations

Facility assurance in Singapore runs on a stack of standards and regulations that do different jobs. The buyer's task is matching the right evidence to the right risk — and verifying scope.

Two practical notes. Certificates are issued per legal entity, per scope, per site — a portfolio ISO 27001 certificate can exclude the building you are contracting for, so ask for the certificate and read the scope statement (our cybersecurity standards guide covers how to do this). And physical placement does not settle data governance: the PDPA permits cross-border transfers with appropriate safeguards, while some sectoral rules and customer contracts impose stricter residency — resolve the legal layer before the facility layer, or you may pay Singapore premiums for workloads that never needed to be here (or vice versa).

Implementation Considerations

A colocation programme is a facilities-plus-networking project with an IT migration attached. Realistic phasing for an enterprise deployment:

• Requirements and market engagement (2–3 months). Power/density/growth model, compliance requirements, site shortlist, RFP. In this market, start earlier than feels necessary — target sites may have no availability.
• Diligence and contracting (2–4 months). Site visits, certificate verification, reference calls, legal negotiation of SLAs, ramp, renewal and exit terms. Financial institutions add MAS-aligned outsourcing diligence here.
• Fit-out (1–4 months). Cage/rack build, power whips, structured cabling, cross-connects, out-of-band management. Cross-connect and carrier delivery times are a common critical-path surprise — order early.
• Migration (1–6+ months). Wave-planned moves with rollback criteria, parallel-run where the risk warrants it, and hypercare. Physical moves of production hardware need specialist logistics and insurance.

Stakeholders beyond IT: facilities/real estate (lease mechanics), finance (capex/opex treatment of long leases), risk and compliance (regulatory mapping), and network engineering early — because carrier diversity, on-ramps and latency budgets drive site selection more than floor plans do. For AI deployments, add hardware logistics (GPU lead times), liquid-cooling commissioning, and power-ramp coordination with the operator; density retrofits mid-term are far harder than provisioning correctly at fit-out. Plan exit from day one: document what leaving would take, negotiate exit assistance into the contract, and avoid single-site architectures for anything critical.

Common Mistakes

• Buying space instead of power. Contracting on square metres or rack count without committed kW and density guarantees — the classic error, and expensive to unwind in an AI refresh.
• Assuming capacity will be there when needed. In a <1.4% vacancy market, "we'll expand next year" is a hope, not a plan. Secure expansion rights contractually or accept the risk explicitly.
• Ignoring renewal mechanics. Scarcity gives operators pricing power at renewal. Uncapped renewal uplift is the most costly clause buyers routinely leave unnegotiated.
• Treating design PUE as delivered PUE. Marketing figures describe the building at full load in ideal conditions. Ask for measured annualised PUE, and remember partially loaded buildings run less efficiently.
• Underestimating cross-connect and bandwidth spend. Interconnection-rich architectures accumulate hundreds of dollars per connection per month. Model the network bill, not just the rack bill.
• Taking certification claims at face value. "Tier III equivalent", portfolio certificates that exclude your building, expired SOC 2 reports — verify documents, scopes and dates.
• Framing colocation vs cloud as ideology. The costly failure modes are symmetric: lifting steady-state workloads into cloud at a premium, or anchoring elastic workloads in a fixed-capacity contract.
• Ignoring the border. Paying Singapore rates for latency-tolerant bulk compute that Johor or Batam would host at materially lower cost is now a board-level oversight, not a technical nuance.
• No exit strategy. Migrations out of a facility are as hard as migrations in. If exit terms, data destruction and assistance obligations are not in the contract, you are negotiating them later without leverage.

The Regional Dimension: Johor & Batam

Singapore's constraints created two satellite markets, and any serious 2026 capacity strategy weighs all three. Johor has become one of the fastest-growing data centre markets globally: roughly 850 MW completed, about 1,800 MW under construction and a multi-gigawatt pipeline, with JLL projecting Malaysia's national capacity to more than double to over 2 GW by end-2026, concentrated in Johor. Land and power are abundant and materially cheaper; hyperscalers and AI platforms have committed at scale. Batam is smaller but rising — Nongsa Digital Park hosts a growing cluster including BW Digital's planned 144 MW NDP-1, positioned for Singapore spillover and Indonesian data-residency workloads. Operators increasingly market the triangle as one platform: Nxera, for instance, brings additional AI-ready capacity in both Johor and Batam onstream in the second half of 2026.

The honest trade-offs: cross-border latency is low but real, and network design across the Causeway or the strait needs diverse paths; power-grid maturity, water and skilled operations talent are still catching up with the pace of construction; and legal regimes differ — Malaysian and Indonesian data, employment and tax law apply, which cuts both ways depending on your residency requirements. The pattern that works: Singapore for interconnection, latency-critical and regulated workloads; Johor/Batam for AI training, bulk compute, storage and disaster recovery — with the split revisited as the satellite markets mature.

Future Trends (2026–2030)

• AI density becomes the default spec. Liquid cooling — direct-to-chip first — moves from differentiator to requirement in new builds, as GPU racks push past 100 kW. Facilities that cannot support it will bifurcate into a lower-value tier serving legacy workloads.
• Growth stays rationed and green-conditioned. Expect further CFA rounds on the DC-CFA2 template, each tying capacity to green energy sourcing — biomethane, hydrogen-ready generation, ammonia, fuel cells — with the ~700 MW Jurong Island park as the decade's flagship supply event.
• The Digital Infrastructure Act reshapes operator obligations. Once enacted, licensing, resilience codes and PUE mandates formalise what leading operators already do — and consolidate the market around those who can comply.
• The SG–Johor–Batam triangle operates as one market. Cross-border fibre, shared operator platforms and workload arbitrage deepen; buyers will contract regionally, not per-city.
• Pricing stays firm. Structural scarcity, rising compliance costs and AI demand point to sustained premiums; the negotiable variables remain term structure, ramp and renewal protection rather than headline rates.
• Sustainability reporting reaches tenants. Operators' green obligations flow down as customer-level energy and emissions reporting, renewable energy certificate arrangements and efficiency clauses — build this into vendor management now.

Frequently Asked Questions

How many data centres are there in Singapore?

Counts vary with methodology. Curated databases list roughly 66 operational colocation facilities from about 46 providers; broader counts including telco rooms, edge sites and announced projects run 80–130+. The more useful measure is capacity: over 1.4 GW operational, the fifth Asia-Pacific market past 1 GW.

Why is colocation in Singapore so expensive?

Deliberately rationed supply (vacancy below ~1.4%), some of Asia's most expensive industrial electricity (~US$0.21–0.24/kWh in early 2025) and scarce land. Market benchmarks put Singapore colocation at roughly US$330–475 per kW/month — at or near the top of Asia-Pacific.

Can we still get capacity in Singapore in 2026?

Retail colocation in the tens of kilowatts remains obtainable, with limited site choice. Wholesale capacity is largely pre-committed before buildings open — DC Tuas launched more than 90% committed. Engage operators 12–18 months ahead of need and expect ramp schedules.

What is the Green Data Centre Roadmap?

IMDA's May 2024 policy framework for sustainable capacity growth: at least 300 MW of additional capacity near-term, around 200 MW more for operators deploying green energy, and a PUE target of 1.3 or better at full IT load across the industry within a decade.

What is DC-CFA2 and who won the first round?

The second Data Centre Call for Application (announced 1 December 2025, closed 31 March 2026) offers at least 200 MW, requiring at least 50% green power. Awards were unannounced as of July 2026. The 2023 pilot awarded ~80 MW to Equinix, GDS, Microsoft and an AirTrunk–ByteDance consortium from over 20 proposals.

What is the Digital Infrastructure Act?

A draft law (consultation closed 22 July 2026) creating a licensing regime for major digital infrastructure: third-party data centres of 10 MW+ critical IT load and cloud providers earning S$100 million+ from Singapore customers would face resilience, security and PUE obligations, with penalties up to S$1 million. For buyers it raises the assured baseline — and likely flows into pricing.

Should we choose colocation or cloud?

Decide per workload. High-utilisation steady workloads, owned GPU clusters and audit-heavy regulated systems favour colocation economics and compliance; elastic and fast-changing workloads favour cloud. Most enterprises run hybrid — and the clouds' Singapore regions physically live in the facilities this guide covers.

What certifications should a facility hold?

Uptime Tier III+ (or TIA-942 equivalent) for resilience, ISO 27001 for security management, SOC 2 for operational assurance, plus PCI DSS, MTCS SS 584 or Green Mark where relevant to your workloads. Always verify the certificate's scope covers the specific building, and its expiry date.

What does colocation actually cost here?

Indicatively in 2026: ~US$330–475 per kW/month at market benchmark; a retail 42U rack ~S$2,200–4,000/month depending on power; cross-connects ~S$200–800 each per month; electricity typically passed through. Wholesale is individually negotiated. Obtain multiple quotes — variance between operators and terms is wide.

Should we consider Johor or Batam instead?

For bulk compute, AI training and DR — increasingly yes: Johor has ~850 MW built and ~1,800 MW under construction at materially lower cost, and Batam is emerging. Keep interconnection-dependent, latency-critical and Singapore-regulated workloads on the island; split the rest deliberately.

Do MAS-regulated firms face extra requirements?

Yes — MAS Technology Risk Management and Outsourcing requirements apply to data centre arrangements: due diligence, audit and inspection access, incident notification and exit planning must be contractually secured. Major operators handle these requests routinely, but the regulatory burden stays with the institution.

How much rack density do AI workloads need?

Conventional enterprise racks run ~4–10 kW; current GPU training racks draw 40–100+ kW and need liquid cooling. Newer facilities (DC Tuas, SG6, SGP1) are engineered for this; many older buildings cannot retrofit economically beyond limited zones. Verify delivered density, cooling method and floor loading before you buy hardware.

Final Recommendations

Singapore remains the strategic anchor of Southeast Asian digital infrastructure — for reasons of connectivity, stability and ecosystem that a decade of supply constraint has not eroded. But it is a market that punishes casual buying. The winning posture in 2026 is deliberate scarcity management: plan capacity 12–18 months out, contract in kilowatts with density and expansion rights in writing, cap your renewal exposure, verify every certificate's scope, and route each workload to the cheapest geography that satisfies its latency and compliance requirements — which increasingly means a Singapore core with Johor or Batam scale-out.

Anchor in Singapore if your workloads monetise low regional latency, dense interconnection or Singapore's legal and regulatory environment — finance, trading, SaaS platforms, telecoms, regional headquarters infrastructure. Look across the border first if your requirement is latency-tolerant bulk compute, AI training or archive capacity, where the Singapore premium buys little. Run hybrid in either case: the colocation-plus-cloud-on-ramp pattern remains the most defensible enterprise architecture in this market.

Above all, treat the policy environment as a market force. The Green Data Centre Roadmap, DC-CFA2 and the Digital Infrastructure Act are not background compliance noise — they determine how much capacity exists, what it costs, and what assurance you can demand from the operators competing for your workload.

Browse Data Centre Vendors in Singapore

Evaluating providers? TechDirectory lists verified data centre operators, colocation providers and connectivity partners across Singapore — with company profiles, certifications and community reviews.