Southeast Asia data centers Report Interpretation
Goldman Sachs’ expert-call takeaways portray Southeast Asia as an AI-driven data-center growth market with 26–28% CAGR through 2030, high utilization and pricing premiums for AI workloads. Singapore remains a constrained premium hub, while Johor is expected to become the region’s largest hub within 8–12 months.
Summary
Goldman Sachs’ expert-call takeaways portray Southeast Asia as an AI-driven data-center growth market with 26–28% CAGR through 2030, high utilization and pricing premiums for AI workloads. Singapore remains a constrained premium hub, while Johor is expected to become the region’s largest hub within 8–12 months.
- The expert expects regional data-center demand to grow at a 26–28% CAGR through 2030, led by AI.
- Utilization is 80–90% across most markets and reaches 99% in Singapore.
- Johor is expected to surpass Singapore as the largest regional hub within 8–12 months.
- AI workloads may command US$200–250/kW monthly pricing, versus US$90–130/kW for hyperscale capacity.
- Powered land, long-lead equipment and skilled construction labor are the principal capacity bottlenecks.
Report Interpretation
Overview
This expert-call note examines AI-led demand, supply constraints, hub positioning, contract economics and operator advantages in Southeast Asian data centers. The report’s central view is constructive: demand is strong and diversified, utilization is high, and constrained supply supports favorable market conditions despite execution and regulatory risks.
Core views
The expert identifies AI as the largest structural demand driver for Southeast Asian data centers and expects the regional market to grow at a 26–28% CAGR through 2030. Deployments have become materially larger than historical orders. US and Chinese hyperscalers are attracted by customer proximity—given that roughly two-thirds of the global population lives in Asia—and by quicker deployment enabled by comparatively available power, land and construction capacity versus constrained US and European markets. AI-cluster rental demand and some spillover linked to Middle East tensions are additional sources of demand. The report is constructive on regional demand-supply conditions. Utilization is 80–90% across most regions and 99% in Singapore; new projects are commonly already 50–60% committed and then fill rapidly. The expert argues that infrastructure owners benefit from multiple demand sources: if one AI-demand category shifts, other AI and non-AI customers can absorb capacity. This diversification underpins the view that data-center assets face no major demand risk relative to neocloud players. Hub dynamics are shifting. Singapore is currently Southeast Asia’s largest hub, with close to 2GW of capacity and Goldman Sachs estimates of 1.5GW of live capacity by 2026. Land scarcity makes it a premium, seller’s market, prompting operators to migrate regional workloads elsewhere while reserving capacity for critical local demand. A DC-CFA2 announcement released another 200MW after a four-year moratorium. Johor is projected to become the region’s largest hub within 8–12 months, supported by Singapore spillover and inbound AI demand, including significant Chinese-hyperscaler demand. It has about 1GW of live capacity, Goldman Sachs estimates 1.4GW by 2026, a 7–8GW early-stage pipeline and 2GW under construction. The expert does not view Johor as oversupplied, while highlighting Malaysia’s relatively greater exposure to GPU-related demand and related regulatory risk. Other regional hubs are at different stages. Indonesia has about 1GW of capacity and is growing steadily. Thailand is seeing significant development and expansion, partly as customers diversify Malaysia exposure, aided by incentives such as those in the Eastern Economic Corridor. The Philippines could accelerate after data-locality laws were announced, whereas Vietnam remains early stage. Supply growth is constrained less by land in most markets than by obtaining powered land on a workable timetable. Converting pipeline into live capacity takes two to five years depending on location, partly because power generation is often distant from data-center sites. Power providers nevertheless favor data-center loads for their stability. Land is a severe constraint in Singapore, Hong Kong and Tokyo, while long-lead equipment—including cooling systems, generators, power equipment and AI chips and servers—also affects build schedules. The expert views water concerns as overstated because efficiency can improve and AI workloads use closed-loop systems that evaporate less than traditional facilities, but skilled labor and construction materials remain physical execution constraints. Contract structures and pricing reinforce the attractiveness of the sector, though customer bargaining power limits a uniformly seller’s-market outcome. Typical leases exceed 10 years; examples include 15-year commitments with unilateral five-year renewals and terms up to 25 years, while the shortest contracts run five years. The expert also sees a healthy secondary market for chips after five to seven years of use. Average monthly service pricing is estimated at about US$350/kW for retail enterprise colocation and US$90–130/kW for hyperscale facilities. AI workloads can achieve US$200–250/kW, a 150–200% premium excluding power costs. Wholesale hyperscale contracts pass through electricity costs, unlike retail colocation, and pricing is generally linked to CPI or a flat escalator such as 3% year on year. Hyperscalers retain meaningful bargaining power because of their scale and experience building their own facilities, and not all hubs compete directly. The operator landscape favors early entrants and scaled platforms. Early local players have first-mover access to critical power resources, while larger operators can procure long-lead equipment earlier and deploy faster. The expert especially highlights Chinese-origin leading platforms’ engineering, program-management and deployment capabilities, including prefabricated cabinets. Chinese hyperscalers are described as more price-sensitive than Western peers and may accept lower redundancy and resilience standards to reduce costs.
Analysis framework
The report synthesizes views from a Southeast Asia data-center expert call, moving from AI-driven demand and utilization to hub-by-hub capacity, supply constraints, contract pricing and operator competition. It uses capacity estimates, pipeline data, utilization, lease terms and price comparisons to explain the regional demand-supply balance.
Methodology notes
Regional data-center demand and supply analysis
The report assesses AI and hyperscaler demand against utilization, committed capacity, powered-land availability, construction timelines and equipment constraints to judge market tightness.
Infrastructure bottlenecks and operator deployment advantages
It links power access, land, cooling and power equipment, chips, construction labor and project execution to the pace at which announced capacity can become live data-center supply.
Key data
- Regional market growth26–28% CAGR through 2030Expert estimate for the Southeast Asian data-center market, driven primarily by AI.
- Regional utilization80–90%Utilization across most regions.
- Singapore utilization99%Indicates a particularly tight premium market.
- Singapore capacityClose to 2GW; GSe 1.5GW live capacity by 2026ESingapore is currently the largest Southeast Asian hub.
- Johor capacity and pipelinec.1GW live capacity; GSe 1.4GW live capacity by 2026E; 7–8GW early-stage pipeline; 2GW under constructionJohor is projected to become the largest regional hub within 8–12 months.
- Pipeline conversion timeline2–5 yearsTime to convert pipeline into live capacity, depending on location.
- AI workload pricingUS$200–250/kW per monthImplied 150–200% premium over US$90–130/kW hyperscale pricing, excluding power costs.
- Typical lease durationMore than 10 years; up to 25 yearsExamples include 15-year commitments with unilateral five-year renewals.
Impact & implications
The report argues that high utilization, long leases, AI pricing premiums and supply constraints support favorable conditions for regional data-center infrastructure owners. It also suggests that location, power access, procurement scale and execution capability are key differentiators, while hyperscaler bargaining power and uneven hub conditions temper the strength of pricing across the region.
Risks
- Malaysia is relatively more exposed to GPU-related demand and may therefore face higher regulatory risk.
- Powered-land availability, long-lead equipment, skilled labor and construction materials can delay capacity delivery.
- Large hyperscale customers retain bargaining power because of their scale and ability to build their own data centers.