Every data centre announced across ASEAN this year, counted

Between 1 January and 15 June 2026, companies announced 87 new data-centre projects across the ten ASEAN member states. BriefAsia counted every one — scraping press releases, planning filings, utility interconnection requests and ground-breaking notices, then cross-checking the figures with developers and grid operators. The result is a map of where the region's digital infrastructure is being poured, and a warning about what it will run on.

The headline numbers are large enough to require context to mean anything. The 87 projects represent roughly 6.4 gigawatts of planned IT capacity and at least US$48 billion in announced investment. To put 6.4 gigawatts in perspective: it is more than the entire installed generation capacity of several ASEAN members, and it is being requested by an industry that did not meaningfully exist in the region a decade ago.

But announcements are not concrete, and capacity announced is not capacity built. The most important finding of this count is the gap between the two — and the single constraint, visible in nearly every market, that will decide how much of this actually gets switched on: electricity.

Where the projects landed

The geography is concentrating, not dispersing. Three clusters absorbed the bulk of announced capacity. Johor, the Malaysian state across the strait from Singapore, led with 21 projects and roughly 2.1 gigawatts — the overflow of demand that Singapore's own moratorium-constrained market cannot site. Greater Jakarta and the Riau Islands in Indonesia followed with 18 projects. Greater Bangkok placed third with 12.

Singapore itself, once the regional hub, announced only 6 projects, all small and all explicitly tied to the city-state's stringent efficiency and low-carbon-power requirements. The moratorium it imposed years ago, and the controlled allocation that replaced it, has done exactly what it was designed to do: push the megawatt-hungry workloads across its borders while keeping the high-value, low-footprint operations at home.

Vietnam surprised on the upside with 11 projects, concentrated around Ho Chi Minh City and a newer cluster near Hanoi, several explicitly framed as AI-training facilities seeking the country's relatively cheap power and improving connectivity. The Philippines logged 9, Thailand 12, and the remaining handful scattered across Malaysia outside Johor, Indonesia outside Java, and a single ambitious announcement in Laos tied to its hydropower surplus.

Who is building

The developer mix tells its own story. The American hyperscalers — the cloud giants — account for the largest single tranche of announced capacity, roughly a third, building both for their own cloud regions and, increasingly, for AI workloads they will not specify in detail. Regional colocation specialists, several backed by infrastructure funds, account for another large slice, building the neutral capacity that hyperscalers and enterprises lease.

The newer entrant is the AI-specialist developer: companies building facilities purpose-designed for dense GPU clusters, with the exotic power and cooling that training large models demands. Eleven of the 87 projects fall into this category, and they are disproportionately power-hungry — a typical AI-training hall draws several times the power per square metre of a conventional cloud data centre. They are the projects most likely to collide with the grid.

"A normal data centre is a building that uses electricity. An AI training facility is a substation that happens to have servers in it," one Johor-based developer told BriefAsia. "The utility relationship is not a detail. It is the whole project."

The power problem, quantified

Here is where the count turns from inventory to warning. If all 87 projects were built and ran at full load, they would draw an additional 6.4 gigawatts — and the grids being asked to supply it are, in several cases, not ready. Johor's 2.1 gigawatts of announced demand exceeds the firm new capacity Malaysia's national utility has committed to that region this decade. Something has to give: either projects slip, or the power comes from somewhere unplanned.

Grid operators in two of the three lead clusters told BriefAsia that interconnection queues now stretch years, and that a meaningful share of announced projects have not secured firm power. The pattern echoes what happened in northern Virginia and parts of Ireland: announcements outran the grid, and utilities began rationing connection. The same throttle is descending on Southeast Asia, just earlier in the cycle.

The carbon dimension compounds it. Much of the new demand will, in the near term, be met by gas and, in Indonesia and Vietnam, by coal still on the grid — undercutting the green-data-centre marketing that accompanied many of the announcements. The hyperscalers have renewable-energy commitments; the grids they plug into often cannot honour them yet. The gap between the press release and the power source is, frequently, a coal plant.

The slippage forecast

How many of the 87 will actually get built? BriefAsia put the question to a dozen developers, financiers and grid officials and assembled their estimates. The rough consensus: expect 55 to 65 per cent of announced capacity to reach commissioning within five years. The rest will slip, shrink, relocate or quietly die — killed most often not by demand, which is real, but by the inability to secure firm, affordable, sufficiently clean power on schedule.

That implies the region will add perhaps 3.5 to 4 gigawatts of data-centre load from this cohort, not 6.4 — still enormous, still grid-straining, but materially below the announced total. The lesson for anyone reading the breathless investment figures: discount the headline by a third before you believe it, and watch the interconnection queues, not the press releases, to see what is real.

What the count reveals

Step back from the individual projects and the map tells a coherent story. Southeast Asia is becoming a major theatre of global AI and cloud infrastructure, drawn by power that is cheaper than in saturated Western markets, by proximity to a fast-growing user base, and by governments eager for the investment. The capital is flowing in faster than the megawatts can be generated to feed it.

The constraint that will shape the next five years is therefore not capital, not land, not chips, not demand — all of which are abundant or arriving. It is firm, clean, affordable electricity, delivered on the timeline the buildings need it. The region announced 6.4 gigawatts of digital ambition in six months. Whether it can power even two-thirds of that is the question the press releases do not answer, and the one the grids will.