Until just a few months ago, Southeast Asia's data center investments looked unstoppable. Now Thailand and Indonesia are suspending projects. The problem isn't chip supply; it's electricity, water and environmental permits. And that shows the real bottleneck in the AI boom is physical resources.

What happened?

In early 2026, Thailand's board of investment approved data center projects from players such as TikTok, UAE-based Skyline and Singapore's Bridge. The total planned investment stood at around $27 billion. But in September 2026, the country halted construction on 49 projects and postponed approval decisions on 117 more. The rationale was that there wasn't enough detail for regulation and oversight. In neighboring Indonesia, the governor of West Java—the most populous province—also halted construction of a data center over missing permits and an incomplete environmental assessment.

Why does it matter?

The region drew in tech giants with promises of cheap energy, affordable land and flexible regulation. The logic was simple: saturated markets like Singapore had already capped new data centers, so demand spilled over into surrounding countries. Thailand and Indonesia wanted to turn that spillover into an industrial push. Foreign capital, jobs and the prestige of becoming a "digital hub" were all appealing.

The trouble is that a data center has a different load profile than a factory. It demands uninterrupted, intensive electricity; it consumes serious amounts of water for cooling. GPU-dense facilities built for AI training draw many times more energy than traditional cloud data centers. A single facility's power demand can approach the consumption of a mid-sized town. Adding that load to the existing grid means competing with households and industry.

There's a timing mismatch here. A data center can be built in 18 to 24 months; a new power plant or transmission line takes years. Capital flows fast while infrastructure can't keep up. The agencies that granted approvals didn't factor this in from the start, and the belated recognition of that gap is exactly why the brakes went on.

Not a chip shortage, a resource bottleneck

For a long time the AI debate revolved around GPU scarcity. Yet chips aren't the only thing you need to build a facility. The same picture has emerged recently in the US and Europe: utilities are queuing up connection requests, and some regions are placing moratoriums on new data center applications. Ireland imposed that limit years ago. The pause in Southeast Asia is the same lesson learned later but harder.

PromisedWhat actually turned up
Cheap, abundant energyLimited grid capacity, competition with households
Fast build-outGridlock in permitting and environmental review
Flexible regulationAn oversight gap, tightened after the fact
Easy access to waterCooling water and climate pressure

This picture holds for any country that wants to become "the new data center hub." Turkey, too, is occasionally mentioned as a regional data center candidate thanks to its geographic location, young population and energy projects. The example here shows that what matters isn't the incentive package but the grid's real carrying capacity and water budget. Approving the investment is easy; supplying the electricity and water sustainably is hard.

What's next?

There are a few things to watch. First, how many of the halted projects end up canceled versus how many continue with conditions attached. Second, companies' tendency to generate their own power: solar, wind and long-term energy agreements are now an inseparable part of the facility plan. Third, whether cooling methods that consume fewer resources than water will become more widespread.

It would be wrong to read this pause as a collapse. Demand is still there; what's changing is that the rules are tightening and the costs are being priced realistically. Investment will likely shift in the coming period toward regions that offer energy security and run transparent permitting processes. The real lesson is clear: the future of AI infrastructure will be written wherever the most reliable kilowatt-hour and water rights are on offer—not the cheapest land.