Southeast Asia has established itself as one of the most consequential arenas in the global race to build artificial intelligence (AI) computing infrastructure. The region currently hosts more than 2,000 active data centres across Indonesia, Malaysia, Singapore, Thailand, Vietnam and the Philippines, with hundreds under construction and over a thousand more advancing through planning stages, according to 2026 data from Ember.
What distinguishes this build-out from comparable expansions in North America or Europe is a demanding combination of speed and constraint. Operators are deploying dense, power-intensive AI infrastructure within a tropical climate averaging between 27 and 35 degrees Celsius, across electrical grids originally designed for standard industrial and residential loads rather than the requirements of hyperscale computing.
A new technical baseline
The commercial adoption of generative AI has fundamentally altered the specifications governing data centre design. Legacy facilities were engineered around predictable workloads: web hosting, basic cloud services and enterprise file storage, with racks operating at between 8 and 12 kW. AI inference and training workloads demand sustained, intensive computation, pushing rack densities to between 25 and 40 kW per rack and beyond.
Wood Mackenzie research indicates that this shift has quadrupled the average power requirement per proposed facility in Southeast Asia, from approximately 24 MW under conventional designs to 106 MW for new AI-ready proposals.
Regional build-out
Investment across Southeast Asia is neither uniform nor uncoordinated. Singapore remains the primary Tier 1 hub, holding over 1.4 GW of capacity across more than 70 cloud, enterprise and co-location facilities. Data centres already account for roughly 7 per cent of national electricity consumption, a structural pressure that previously led to a construction moratorium between 2019 and 2022. The city-state now operates a framework of constraint by design, permitting only the most efficient and advanced facilities. Microsoft has pledged $5.5 billion between 2025 and 2029 for cloud and sovereign AI infrastructure; similarly, Amazon Web Services has committed $9 billion through 2028.
Malaysia, and specifically the southern state of Johor, has become the region’s leading scale-out market for AI processing. Abundant land, lower power costs and direct proximity to Singapore’s network infrastructure have attracted large campus-style deployments. YTL Power, in partnership with Nvidia, is developing a liquid-cooled AI campus in Johor; AirTrunk is constructing a hyperscale facility within the same corridor and Microsoft has secured land for a forthcoming cloud region. Wood Mackenzie places Malaysia’s total development pipeline at approximately 3.4 GW, representing around 60 per cent of all proposed regional capacity.
Indonesia presents a distinct dynamic, with infrastructure growth driven primarily by domestic demand from a population of approximately 280 million. Development is concentrated around Greater Jakarta, where BDx Data Centers launched the country’s first sovereign AI data centre in late 2024, built using Nvidia accelerated computing; meanwhile, established operators including DCI Indonesia continue to expand their presence. Complex permitting timelines and a national grid heavily dependent on coal-fired generation remain persistent constraints, though the volume of local enterprise demand continues to drive construction forward.
Thailand has undergone an exceptionally sharp acceleration, supported by proactive government incentives and improved connectivity. DC Byte data shows that total IT capacity grew by more than 2,000 per cent between 2019 and 2024, with AI now accounting for 28 per cent of installed capacity, up from 20 per cent per year earlier. Amazon Web Services is investing more than $5 billion over 15 years, having launched its Bangkok cloud region in January 2025; Google has committed $1 billion across Bangkok and Chonburi; and ByteDance has pledged $8.8 billion in the Eastern Economic Corridor. Chinese cloud providers including Huawei, Alibaba and Tencent are expanding alongside Western operators, making Thailand one of the few markets where American and Chinese technology stacks operate in direct proximity.
Vietnam presents a compelling cost case, with construction costs of approximately $7 million per MW, nearly 50 per cent below those in Singapore or Tokyo. Strict data sovereignty laws requiring local storage of Vietnamese data create structural demand for domestic hosting. Key projects include Viettel’s 140 MW facility in Tan Phu Trung, a $1.3 billion development by Samsung C&T and CMC, and a $2 billion consortium involving G42, Microsoft, FPT and VinaCapital. Vietnam is targeting a digital economy equivalent to 30 per cent of national GDP by 2030.
Three structural bottlenecks
Despite the scale of capital flowing into the region, three structural constraints pose genuine risks to sustained expansion.
Power is the binding variable. Southeast Asia holds the lowest renewable generation mix and the second-highest grid emissions intensity in Asia Pacific, according to Wood Mackenzie. The International Energy Agency notes that a typical AI-focused data centre consumes as much electricity as 100,000 households, with the largest facilities under construction demanding 20 times that volume.
Cooling in a tropical environment amplifies the pressure. Maintaining optimal operating temperatures under sustained heat and humidity drives up both energy consumption and water use. Across Asia Pacific, data centre water consumption is projected to rise from approximately 0.92 trillion litres in 2025 to 1.7 trillion litres by 2030. Liquid cooling systems, increasingly mandatory for high-density AI deployments, add approximately $1.80 to $2.40 per watt in capital expenditure.
The talent deficit completes the picture. Malaysia illustrates the scale of the challenge starkly: domestic universities produce approximately 5,000 engineering graduates per year against a stated national requirement of 50,000, a tenfold shortfall made worse by a persistent outflow of skilled professionals to Singapore for higher compensation.
Outlook
By 2030, Southeast Asia’s total data centre capacity is projected to reach between 5.2 and 6.5 GW, as per industry estimates, broadly triple the current installed base, driven by a roughly tenfold increase in AI-specific computing demand alongside conventional cloud growth. The division of labour already taking shape is expected to deepen: Singapore will anchor high-value financial services, research and operational command functions, while Malaysia, Thailand, Indonesia and Vietnam supply the physical land, raw compute and power infrastructure to run AI models at scale.
The long-term trajectory of this build-out depends fundamentally on whether energy and water infrastructure can expand in parallel with compute demand. If the power gap widens faster than grid upgrades and renewable procurement can close it, growth will plateau regardless of available capital. Operators must simultaneously navigate fragmented regulatory environments across 10 member states while managing the structural complexity of coexisting Western and Chinese technology stacks. That tension is increasingly not peripheral to the story of AI infrastructure in Southeast Asia.
