Thailand froze 49 data centers and delayed decisions on another 117 projects
AI companies expected to scale computing capacity quickly in Southeast Asia, but infrastructure is failing to keep pace. Thailand suspended construction of 49 data centers and delayed decisions on another 117 applications. Authorities in Malaysia and Indonesia are also tightening requirements: new sites need access to water, guaranteed power, and full environmental permits. For product teams, this is no longer a background climate debate but a risk to timelines and costs.
AI-processed from Guardian; edited by Hamidun News
The race for computing capacity in Southeast Asia has, for the first time, run hard into limits not on AI demand but on power outlets, reservoirs, and permits. In September 2026, Thailand suspended construction of 49 data centers and delayed decisions on another 117 projects. In Indonesia, authorities in West Java halted construction of a facility that lacked a complete set of permits and a finalized environmental assessment.
To a developer, this may sound like distant infrastructure policy. In practice, the chain is short: a data center is not connected on time — the cloud provider does not receive new capacity — fewer accelerators are available — launching a model becomes more expensive or moves to another region. Founders must choose among a product delay, a more expensive cloud setup, and an architecture capable of surviving workload migration.
Why the region became the center of the race
Singapore was long Southeast Asia's main data center hub, but a moratorium on new facilities was in force from 2019 to 2022. The reason was simple: server campuses consume large amounts of electricity and water, while land is scarce in the city-state. After the moratorium was lifted, projects began receiving selective approval, subject to energy-efficiency and emissions requirements.
Investment flowed into neighboring countries. In the Malaysian state of Johor, operational data center capacity rose from 94 MW in 2023 to 1,110 MW in 2026 — more than tenfold. Another 3,088 MW is under construction or planned. The Jakarta metropolitan area has a declared project pipeline of 1,699 MW, while Bangkok's stands at 2,084 MW.
A familiar pattern emerged: businesses see cheap land and a growing market but evaluate a site separately from power grids, water, and residents' interests. A data center can be built faster than a new power plant, substation, or water-supply system. That is why capacity shown on a presentation slide does not yet mean the servers will actually switch on in the promised quarter.
Where the bottleneck emerged
According to Wood Mackenzie, data centers already consume about 24% of Johor's electricity. By 2035, that share could rise to 40%. This means new transmission lines, additional generation capacity, and competition among large consumers for grid connections. Thailand is considering a separate, higher tariff category for data centers because a significant share of its electricity is generated from expensive imported gas.
Water has become the second constraint. Cooling systems need it, especially in a hot and humid climate. Residents of eastern Thailand fear that industrial sites will increase pressure on already depleted reservoirs. Johor saw its first public protest against data center construction: people were concerned about dust, the facility's proximity to homes, and its potential effect on the water supply.
Operators are responding with water-reuse projects and more efficient cooling. These are useful measures, but the promises are not yet always backed by detailed, openly available data. Even Microsoft's published PUE metrics show that its Asia-Pacific sites are less energy-efficient than the company's facilities in other parts of the world.
What product teams should do
Consider a service planning to launch an AI feature for customers in Asia and relying on a single cloud region in Bangkok. The team has agreed on a budget, response latency, and the volume of available accelerators. Then commissioning of a new facility is postponed and the electricity tariff for data centers rises. The feature's economics change before release.
The launch region should therefore be evaluated on more than the price of a virtual machine. Ask the provider for information on current and future capacity, connection timelines, accelerator limits, and migration options. Check in advance whether the application can operate in a second region without violating data requirements or causing a sharp increase in latency.
The financial model should include a stress scenario: cloud prices rise, the GPU quota shrinks, and the regional launch slips by six months. Critical services need a backup provider, portable containers, and data that can be restored outside the primary site.
Southeast Asia remains a promising data center market. But the era when announcing billions in investment and starting construction was enough is coming to an end. Computing capacity now depends on three things at once: servers, regional resources, and public consent. Product architecture will have to account for all three.
- Nov 8, 2026Docx Anonymizer recognizes 50 data categories and supports nine formats
- Nov 8, 2026Microsoft scales back AI capacity rentals from Oracle, reviewing $80 billion in contracts
- Nov 8, 2026Google buys troubled Crusoe data center and will invest $17 billion in the campus
- Nov 8, 2026Sutskever, Silver and LeCun Leave the Big Labs: What the Neolabs Are Betting On
Key Questions Answered
Why the region became the center of the race?
Singapore was long Southeast Asia's main data center hub, but a moratorium on new facilities was in force from 2019 to 2022. The reason was simple: server campuses consume large amounts of electricity and water, while land is scarce in the city-state. After the moratorium was lifted, projects began receiving selective approval, subject to energy-efficiency and emissions requirements. Investment flowed into neighboring countries.
Where the bottleneck emerged?
According to Wood Mackenzie, data centers already consume about 24% of Johor's electricity. By 2035, that share could rise to 40%. This means new transmission lines, additional generation capacity, and competition among large consumers for grid connections. Thailand is considering a separate, higher tariff category for data centers because a significant share of its electricity is generated from expensive imported gas. Water has become the second constraint.
What product teams should do?
Consider a service planning to launch an AI feature for customers in Asia and relying on a single cloud region in Bangkok. The team has agreed on a budget, response latency, and the volume of available accelerators. Then commissioning of a new facility is postponed and the electricity tariff for data centers rises. The feature's economics change before release. The launch region should therefore be evaluated on more than the price of a virtual machine.
Ship AI Faster with Founders, Engineers & Weekly Architecture Reviews
Inside Vibe Club (hamidun.com/club) — weekly live masterminds with Zhemal Khamidun, Tech Radar, full LMS access, and 370+ production Claude Code skills.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.