ASEAN telecom and data centers benefit from the expansion of AI compute infrastructure

Overview

This report is an investor-oriented thematic study on ASEAN telecom and data centers, focusing on how AI infrastructure buildout is reshaping data center economics, site-selection logic, and supply-chain exposure. The report notes that NVIDIA GPUs, agentic AI frameworks, and sovereign AI demand are driving a new wave of compute infrastructure expansion; ASEAN, with its land, power, subsea cable connectivity, and localization demand, is poised to become an important hosting region for AI data center clusters.

Core views

The report’s core views include: first, MW remains the core unit of data center capacity, but revenue depends on leased load, pricing, power structure, ramp-up pace, interconnection services, and customer mix; second, AI shifts competitive focus from space to power density, cooling, and networking; third, equity investment opportunities lie not only with data center owners but also across the supply chain in power, cooling, networking, servers, semiconductors, and optical communications; fourth, Singtel/Nxera can convert regional connectivity capabilities into AI workloads through network connectivity, AI data centers, GPUaaS/AIaaS, and orchestration platforms.

Analysis framework

The report uses a “Data Center 101” framework, building an investor KPI bridge from capacity, leasing, utilization, and EBITDA to ROIC; it then explains the logic behind ASEAN data center cluster formation through connectivity, land, power, climate, demand, and policy; finally, it assesses AI infrastructure monetization potential through changes in the technology stack, supply-chain mapping, and the Singtel/Nxera case study.

Methodology notes

• Industry analysisData center KPI bridge

  Capacity → Leasing → Utilization → EBITDA → ROIC

  It links capacity, leasing, utilization, profitability, and return on capital into the data center investment logic, emphasizing that MW is only the starting point, while revenue and returns also depend on pricing, power pass-through, customer mix, and ramp-up pace.

• Infrastructure site selectionData center cluster formation framework

  Connectivity, land, power, climate, demand, policy

  Optimal site selection requires balancing fiber/subsea cable/cloud access, land availability, electricity price and reliability, climate and water resources, enterprise and sovereign AI demand, and tax and permitting policies.

• Technology stack analysisAI rack density and cooling pathway

  From traditional racks to AI GPU racks to next-generation AI platforms

  Traditional enterprise racks are about 5-15 kW, current AI GPU racks are about 40-200 kW, and next-generation platforms may reach several hundred kW or more, driving cooling to evolve from air cooling toward direct-to-chip and immersion liquid cooling.

• Value chain mappingAI infrastructure supply chain map

  Data centers, colocation, telecom, power generation, power grid, servers, networking, semiconductors, cooling, and power equipment

  The report extends beneficiaries to hardware and infrastructure chains outside the data center itself, emphasizing that AI compute buildout is a systemic investment theme spanning power, networking, servers, chips, and cooling.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• Singtel / Nxera

  Direct beneficiary and core case study

  Strengths

  It has regional telecom networks, subsea cables, edge connectivity, and the Nxera AI data center platform, and is extending toward GPUaaS/AIaaS and orchestration services through RE:AI and Paragon.

  Weaknesses

  It requires continued investment in high-density data centers, liquid cooling, power, and partner resources, with relatively high capex and execution complexity.

  Comparison

  Compared with pure colocation platforms, Singtel’s differentiation lies in the connectivity layer, enterprise customer access, and sovereign AI localization capabilities.

  Risks

  The completion of the STT GDC transaction, project financing, pre-leasing progress, customer concentration, power access, and ROIC ramp all require monitoring.

• ASEAN telecom operators and data center platforms

  Regional structural beneficiaries

  Strengths

  They benefit from sovereign AI, local data residency, regional cloud access, low-latency connectivity, and relatively available land/power resources.

  Weaknesses

  Power-grid reliability, permitting speed, land, climate, and policy differ significantly across markets.

  Comparison

  Training workloads are more oriented toward large clusters with low-cost power and high utilization, while inference workloads place more emphasis on proximity to users and data-sovereignty regions.

  Risks

  Policy changes, permitting delays, power price volatility, PPA execution, and natural-disaster risks may affect project returns.

• Power, grid interconnection, and energy equipment supply chain

  Key constraint and upstream beneficiary segment

  Strengths

  AI data centers require stable power, redundancy design, PPAs, grid equipment, energy storage, and power electronics capabilities.

  Weaknesses

  Project cycles are long and constrained by utilities, grid-connection schedules, and equipment delivery cycles.

  Comparison

  In high-density AI scenarios, the importance of power constraints may exceed that of land and building shells.

  Risks

  Grid-connection delays, equipment inflation, carbon-emissions constraints, and utility dependence may compress returns.

• Cooling and liquid-cooling supply chain

  Core enabling segment for high-density AI racks

  Strengths

  The evolution from air cooling to direct-to-chip and immersion liquid cooling improves the usability of high-density capacity.

  Weaknesses

  There is still uncertainty around technology pathways, retrofit difficulty, maintenance capability, and customer certification.

  Comparison

  Operators and suppliers with liquid-cooling experience are more differentiated than traditional data hall builders.

  Risks

  Cooling-system failures, retrofit cost overruns, supplier delivery delays, and changes in technical standards.

• Networking, optical communications, servers, and semiconductor chain

  Spillover beneficiaries in the AI data center value chain

  Strengths

  High-performance GPUs, HBM, high-speed switching, optical modules, server ODM/EMS, and network interconnection form the foundation of AI cluster expansion.

  Weaknesses

  The supply chain is fragmented and cyclical, and some companies are not covered by the report.

  Comparison

  Equity exposure is not limited to data center landlords, but also includes the NVIDIA ecosystem, networking equipment, optical communications, and server components.

  Risks

  GPU supply, customer capex cycles, technological substitution, inventory volatility, and geopolitical restrictions.

Key data

• Report date2026-06-29The cover shows June 29, 2026 12:00 AM GMT.
• AI infrastructure assessmentHistoric inflection pointThe report says AI infrastructure buildout is driven by NVIDIA GPUs and agentic AI frameworks, with ASEAN at the intersection of land, power, connectivity, and sovereign AI demand.
• Data center revenue driversContracted capacity, billing model, power pass-through, ramp-up pace, interconnection services, customer mixThe report emphasizes that MW is not the only revenue metric.
• Traditional rack power5-15 kWTraditional enterprise rack power density is significantly lower than that of AI GPU racks.
• Current AI GPU rack power40-200 kWAI training and high-density inference are driving higher current delivery, networking, and cooling requirements.
• Next-generation AI platform powerSeveral hundred kW+Higher-density platforms will further increase the importance of liquid cooling and power systems.
• Singtel/Nxera four-layer stackConnectivity、AI DC platform、GPUaaS & AIaaS、OrchestrationCorresponding to network connectivity, Nxera, RE:AI, and Paragon, with the goal of converting regional connectivity and capacity into AI workloads.
• STT GDC platform scale50 data centers, about 673MW of operating capacity, 12 marketsThe report says the platform brings operating scale and geographic footprint to the consortium.
• STT GDC medium-term pipeline>400MW medium-term pipeline；combined platform about 2.8GW design capacity / 12 marketsThe report also mentions an approximately 1.7GW cross-market pipeline.
• STT GDC transaction metricsEV S$13.8bn；S$6.6bn cash acquisition of 100% of STT GDC；Singtel to invest S$740mn for a 25% stakeExpected to be accounted for under the equity method, with no impact on dividends; management expects completion in 2H26.
• Singtel incremental capexAbout S$400-500mn over the next three yearsRelated to the STT GDC transaction and platform expansion.
• Singtel valuation metricAbout 4% EV uplift based on MSeA value indication provided by the report, not equivalent to a formal target price.

Impact & implications

For investors, the data center theme should not be assessed only by building area or nominal MW; the focus should instead be on power access, liquid-cooling capability, network density, customer pre-leasing, financing structure, and ROIC. If ASEAN telecom operators can integrate subsea cables, edge nodes, sovereign AI demand, and local partners into AI-ready capacity, they may achieve higher strategic value than traditional colocation.

Risks

• Power access, PPAs, grid-connection dates, and redundancy design may fall short of expectations.
• High-density AI racks may drive electrical, cooling, and networking equipment capex above expectations.
• Insufficient pre-leasing, high tenant concentration, or declining customer credit quality could weigh on utilization and EBITDA ramp-up.
• Unclear billing models, power pass-through, and contract duration may affect revenue visibility.
• Changes in data sovereignty, permitting, tax incentives, and regulatory policy may alter project economics.
• Financing structures, JV arrangements, project debt, and asset-rotation progress affect returns on capital.
• Orbital data centers are a long-term option rather than a near-term substitute, but technological breakthroughs could become a distant disruptive factor.

What to watch

• Whether the Singtel/STT GDC transaction is completed in 2H26 as management expects.
• Construction, pre-leasing, and customer mix at Nxera’s AI-ready campuses in Singapore, Thailand, Indonesia, and Malaysia.
• Whether RE:AI, Paragon, and GPUaaS/AIaaS truly deliver monetization above wholesale colocation.
• Grid-connection dates, PPAs, backup power, and utility dependence for each project.
• Rack power density, liquid-cooling retrofit capability, and the pace of transition from air cooling to direct-to-chip/immersion liquid cooling.
• Billing methods in leasing contracts for each kW, each kWh, shell-and-power, or bundled services, and whether power can be passed through.
• capex/MW, steady-state EBITDA, ROIC, and downside scenarios under leasing delays.
• Launch costs, radiation shielding, thermal design, orbital debris, safety, and replacement cycles related to orbital data centers.