Southeast Asia’s Data Center Expansion Faces a Grid-Capacity Constraint
DataNews Editorial Team
Data center electricity demand is rising across Southeast Asia as AI deployments and new capacity concentrate in Singapore and neighbouring markets, while grid investment and interconnection capacity lag. Governments are responding through generation additions, direct-supply arrangements, cross-border transmission and higher-density cooling infrastructure.
- Southeast Asian data center power demand is estimated at about 2.6 GW in 2025 and 10.7 GW in 2035.
- Grid modernisation needs are estimated at roughly US$18 billion per year, with current investment described as inadequate.
- Thailand has permitted direct state-utility power supply for data centers with loads above 200 MW.
- The China–Laos 500 kV interconnection entered operation in April 2026 and lifted bilateral exchange capability to 1.5 million kW.
- AI rack densities of 40–200 kW are increasing the relevance of liquid cooling in new deployments.
Southeast Asia’s data center build-out is increasingly constrained by power-system capacity. AI-driven demand and the concentration of new facilities in Singapore, Malaysia and Indonesia are increasing requirements for generation, transmission and confirmed grid connections, while network modernisation is not keeping pace.
The IEA estimate cited in the source indicates that Southeast Asian data center electricity demand will more than double by 2030 from current levels. A separate estimate in the same article places regional data center power demand at about 2.6 GW in 2025, rising to 10.7 GW by 2035. Singapore, Malaysia and Indonesia have attracted more than 70% of regional investment in computing infrastructure.
Grid limitations also affect the supply pipeline. In Vietnam, Thailand and Indonesia, around 50–60% of renewable-energy projects were cancelled or stalled between 2021 and 2025, according to the article. Southeast Asia requires about US$18 billion annually for grid modernisation, although actual investment is described as insufficient. Solar and wind capacity in the region exceeded 28 GW in 2023, following 20% growth that year.
Countries are pursuing a mix of generation, policy and interconnection measures. Malaysia plans to add 8 GW of gas-fired generation by 2030. Thailand has authorised its state power utility to supply data centers with loads above 200 MW directly. Vietnam targets renewables to provide 30.9–39.2% of electricity generation by 2030.
Cross-border transmission is also being expanded. The China–Laos 500 kV interconnection project entered operation in April 2026, increasing bilateral power-exchange capability from 50,000 kW to 1.5 million kW. China exported 2.39 billion kWh of electricity to ASEAN in the first half of 2026, up 42.9% year on year, according to the article.
At the facility level, AI rack density is increasing the need for more capable thermal infrastructure. Traditional enterprise racks are cited at 5–15 kW, compared with 40–200 kW for GPU-based AI racks. The article estimates that liquid cooling penetration exceeded 50% for AI chips and 70% for newly built AI racks in 2026. These estimates point to cooling architecture as a material part of data center deployment planning where high-density AI loads are involved.
The regional market could therefore compete increasingly for verified grid interconnection, low-carbon generation and inter-system transfer capacity, alongside sites and customers. Network infrastructure, storage, energy-efficiency equipment and liquid cooling may benefit, but overloaded networks, permitting delays and power-purchase agreement challenges could delay final investment decisions and facility delivery.
Grid access, rather than land or customer demand alone, is becoming a primary constraint on data center siting and scale-out in Southeast Asia. This could increase the value of secured interconnection capacity, low-carbon power access, cross-border transmission and efficient cooling, while raising execution risk for projects without firm energy resources.
- — / October 25, 2026 · ChineseSource →
This article was prepared by the DataNews editorial team based on the sources listed above.