Southeast Asia is no longer a sideshow in the map of artificial intelligence. It has become a third geography of computing: a chain of ports and substations from Jurong Island to Batam where American hyperscalers, Chinese cloud groups and local governments are planting steel and silicon.
Tracked AI-oriented facilities in the region already add up to about 7 gigawatts, much of it still under construction. The contest is not only power usage. It is who trains models on whose chips, whose language those models speak, and which capital sets the terms.
R. Ballater Aegis
Makati, M. M.
Sep. 24th, 2026
Scarcity in Singapore, space in Johor
Singapore remains the premium node with its miles of cables, talent, and law, but it is rationing power and land. A second capacity auction in 2026 awarded only 200 MW, with green-power conditions. Microsoft still added US$5.5 billion; AWS (Amazon Web Services) has pledged about US$9 billion through 2028. The city-state’s bet is to stay the control tower while the warehouses move next door.
And they have. Johor is now a market of its own. Malaysia has drawn more than US$34 billion from Microsoft, Google, AWS, Oracle, Alibaba and Tencent. Capacity was projected to more than double to about 2,055 MW by of this year. The flagship American-linked site is the YTL–NVIDIA AI campus at Kulai. Google’s US$2 billion Malaysian region and Microsoft’s US$2.2 billion package sit nearby. Earlier this month, OpenAI reserved space at two Firmus factories in Malaysia.
Kuala Lumpur wants more than rent. The 2026 budget set aside about US$490 million for a sovereign AI cloud and paused non-AI data-center applications. The risk, researchers say, is that Malaysia will get caught hosting other people’s machines faster than it builds its own models.
Bangkok / Jakarta / Hanoi / Manila
Thailand seems currently to be the fastest riser, attracting higher levels of FDI in the AI realm than most of its neighbors. Google put US$1 billion into Chonburi and Bangkok; AWS opened a Thailand regional office on a US$5 billion-plus commitment; Microsoft pledged more than US$1 billion with Gulf and AIS. In September Bangkok froze scores of approvals until power, water and siting rules are set.
Indonesia sells scale plus sovereignty but insists designated data must stay onshore. Microsoft’s US$1.7 billion sits beside NVIDIA-linked ventures and a US$5 billion Batam Campus – a Chinese-backed hall on an island that is a low-latency annex of Singapore. Batam is now a presidential priority digital zone for Prabowow.
Vietnam has broken ground on wholly foreign-owned data centers and initiated a standalone AI law in March, which reportedly pulled more than US$7 billion in commitments.
The Philippines is the late entrant. A US$34.4 billion masterplan aims to lift AI capacity from about 50 MW to 1.5 GW by 2033, tentatively anchored on Pampanga – Bataan. Two unnamed U.S. hyperscalers are each studying about 200 MW. Power prices, capacity, and indigenous resistance remain handicaps.
U.S. labs, Chinese halls
Microsoft’s 2024 tour set the pace; Google writes large site cheques. AWS often leads on committed dollars; NVIDIA supplies the factories others brand.
OpenAI for Singapore is a multi-layered effort that will set out more than S$300 million with its first Applied AI Lab outside the United States, and 200-plus technical jobs. It is rolling ChatGPT ads across six ASEAN markets and named a Manila reseller. Anthropic opens a Singapore office in October. Temasek holds equity in both.
Never to be outdone in their own backyard, Chinese groups are not spectators. Alibaba’s largest Southeast Asian footprint is now in Malaysia. Reporting in late 2025 said Alibaba and ByteDance used regional halls to train on NVIDIA chips scarce at home. ASEAN is courting both bids and hoping rivalry raises the price rather than forcing a choice.
Models that speak the neighborhood
SEA-LION, from AI Singapore, is the regional open-source family for Indonesian, Thai, Vietnamese, Filipino, Malay and more. Around it are Indonesia’s Sahabat-AI, Thailand’s Typhoon; Vietnam’s PhoGPT/VinAI, a Philippine LLM targeted for 2027 and Singapore’s MERaLiON for speech. They may not beat frontier English models but will certainly make up for it in enthusiasm.
The emerging division of labor depends on how Singapore rations high-trust work. Malaysia and Thailand take the halls, Indonesia sells market size, and the Philippines writes plans with hopes the grid keeps up. Countries that treat data centers as an end will remain tenants. Those that use the bidding war to fund sovereign clouds, local models and cleaner power will have a say in the next map of intelligence – not only a place on it.
The bill comes due in megawatts.
The geography is already splitting. Singapore rations high-spec, high-trust work. Malaysia and Thailand take the training and inference halls. Indonesia sells market size and onshore rules. Vietnam sells legal clarity and cheaper land. The Philippines is writing a master plan and hoping the cables and the grid catch up.
The constraints are physical. AI halls drink power and water. Johor has felt the strain; Thailand has frozen permits; several ASEAN states are looking again at nuclear. A standard campus can use as much electricity as a small city. If the grids stay on gas and coal, the region will host the models and export the carbon.
The strategic risk is the one The Diplomat flagged: ASEAN is buying capacity faster than capability. Hyperscaler money builds rooms. It does not automatically build chips, model weights or the talent to audit them. Treat data centers as an end and you remain a tenant. Use the American and Chinese bidding wars to fund sovereign clouds, local models and cleaner power, and you get a say in the next map of intelligence — not only a place on it.