Southeast Asia’s AI Data Centers Have Hit the Water Wall

Southeast Asia’s AI data center boom just met its first hard stop. Indonesia halted construction of BDx Data Centers’ 640-megawatt AI campus at Jatiluhur, West Java, in early October 2026 — a project that had broken ground barely two weeks earlier — over pending environmental and building approvals and pointed local concerns about drawing water from the Jatiluhur reservoir, a major source for Greater Jakarta. Days later, South China Morning Post’s regional read was blunt: with Thailand pausing projects and Malaysia facing grid pressure, the region has started asking Southeast Asia AI data center questions that enthusiasm once drowned out — at what cost, and who pays it?

Key Takeaway

• 🛑 The era of automatic approvals is over: Indonesia stopped BDx’s 640MW Jatiluhur AI campus over water use and permits — the largest Southeast Asia AI data center project in the country, paused by a governor’s order.
• 💧 Water is the new fault line: AI halls are thirsty before they are profitable; reservoir-front campuses now face public scrutiny that hyperscaler press releases never mentioned.
• ⚡ Power grids are the second wall: Malaysia strains its grid, Thailand pauses builds, and Singapore still rations capacity by sustainability score rather than open auction.
• 🏗️ The boom has not stopped — it has bifurcated: funded projects with cleared resources keep building ( Bitdeer’s 67MW A901, Batam’s 120MW green campus, Zankore’s $3.1B financing) while risky builds stall.
• 🌍 Why it matters beyond the region: SEA carries a third of global data center construction intent; how it balances AI demand against water and power becomes the template every emerging market will copy or reject.

The Southeast Asia AI Data Center Map After the Halt

The sequence matters, because it shows how fast sentiment turned. On September 22, 2026, Singapore-based BDx Data Centers announced construction had begun on the first of six planned buildings at its 640MW AI data center campus in Jatiluhur, West Java — its largest Indonesian campus, with 845 MVA of grid capacity secured from state utility PLN and the first 120MW building expected in service early 2027. On October 1, officials in Purwakarta regency inspected the site and found construction stopped. West Java Governor Dedi Mulyadi had ordered the halt, citing outstanding environmental and building approvals — and, publicly, the project’s planned use of water from the Jatiluhur reservoir, a major source of drinking and irrigation water for Greater Jakarta and the surrounding farmland. BDx confirmed the stop on October 3 and said it still targets first-phase service in early 2027, working with authorities to complete remaining requirements; calls for a public environmental impact assessment followed within days.

The halt did not come from an anti-AI movement. It came from the ordinary machinery of a government discovering that its fastest-growing industry consumes the two resources its citizens also drink and light their homes with. Indonesia’s data center sector is expanding rapidly — the coordinating minister for economic affairs put operational capacity near 580MW in July 2026, with investor interest worth another 1.3GW — which is precisely why the Jatiluhur campus’s reservoir became a national conversation instead of a permit footnote. One governor’s order just redrew the Southeast Asia AI data center map for every developer, financier, and regulator watching from Manila to Bangkok.

Why Water Became the Resource That Stops Projects

Data centers drink in ways the public only notices when the source has a name and a constituency. Cooling a dense AI hall at the power levels modern racks demand takes either electricity-hungry mechanical chillers or water-hungry evaporative systems — and at 640MW of critical load, even efficient designs move enormous volumes. Jatiluhur reservoir is not an anonymous aquifer: it supplies a meaningful share of Greater Jakarta’s water and irrigates West Java farmland, which made “AI versus drinking water” a story every local outlet could tell in one image. Governor Mulyadi’s framing was that local supply must not be disrupted — a bar that turns every future West Java campus proposal into a water-budget negotiation first and a technology pitch second.

The pattern is regional, not Indonesian. Environmental groups had already raised water and wildlife concerns around Google’s US$15 billion data center development in India, with government authorities rejecting claims that it was approved without sufficient environmental review. Microsoft’s Hyderabad facilities advertise air-cooled chillers that consume no water for cooling — a claim carefully scoped to cooling rather than total consumption, because the industry learned that “no water” is the new marketing front line. Singapore, the region’s most mature market, simply stopped selling open capacity: its Data Centre Call process allocates 200MW provisionally on sustainability and efficiency criteria, effectively rationing by green score. Indonesia’s own newest flagship — the Nongsa Batam 120MW campus inaugurated October 9 — leads with liquid immersion cooling at PUE below 1.15 and a “green hyperscale” identity because that is what approvals now reward.

For AI specifically the stakes are higher than for conventional cloud halls. AI racks concentrate tens of kilowatts each; high-density liquid cooling is becoming standard precisely because air can no longer carry the heat away economically. Direct liquid cooling still needs water treatment loops and eventually evaporative heat rejection in most designs — so the technology that makes AI halls viable is the same one the public reads as threat. Campus operators who arrive with reservoir-free designs (closed-loop dry coolers, treated seawater, or air-side economization as Batam does) are discovering that those choices, once green marketing, are now license conditions. In the Southeast Asia AI data center market of 2026, the cooling architecture is the permit strategy.

Power: the Quieter, Bigger Wall

If water makes the headlines, power sets the ceiling. BDx had secured 845 MVA from PLN — a serious grid commitment for a single campus — and Malaysia’s national grid is visibly straining under the combined load of its data center buildout, one reason SCMP framed the Southeast Asia AI data center rethink as region-wide rather than country-specific. Thailand’s pause adds a third voice to the choir: hosting the AI boom means signing up for generation, transmission, and tariff consequences that outlast any government’s term. The physics is unforgiving — a 640MW campus runs near the average electrical draw of a mid-sized city — and the political economy even less so, because every megawatt an AI hall takes is a megawatt households and factories either pay more for or wait longer to get. Grid headroom, not land, is what now constrains every Southeast Asia AI data center proposal of consequence.

That is why the bifurcation visible in 2026 matters more than any single halt. The projects still moving — Bitdeer’s A901 with its 10-year services agreement, Zankore with $3.1 billion of committed financing and NVIDIA GB300 NVL72 systems slated for its 200MW first phase, Batam’s freshly inaugurated campus — share a common design: cleared environmental footing, secured grid capacity, and financing sized for the long build. The projects stalling share the opposite profile: big ambition on paper, contested resources underneath. Southeast Asia is not rejecting AI infrastructure; it is repricing it in water and megawatts, and only plans that survive that pricing still move. Every Southeast Asia AI data center pipeline now splits the same way.

What It Means for Builders, Buyers, and the Region’s Workers

Three practical reads follow, and they reach past the boardroom. For operators and investors: resource clearance is the new schedule driver — expect water-budget studies, public impact assessments, and grid-connection queues to consume more calendar time than construction itself, and underwrite accordingly. For enterprises buying AI capacity in the region: Southeast Asia AI data center supply is bifurcating into “cleared and funded” versus “announced and uncertain,” so contract against delivery milestones (as Bitdeer’s agreement does) rather than press-release megawatts. For workers and professionals: the buildings that get built concentrate high-value operations, electrical, and cooling-engineering jobs in fewer, better-financed campuses — the talent demand follows the projects that survive scrutiny, which is one more reason the region’s AI skills gap (ADB pegs it near 48% for technical roles) concentrates in the same markets that can clear a water budget.

For Filipino readers the relevance is concrete without being forced: the Philippines competes for the same hyperscaler and neocloud capital, and the Jatiluhur lesson says the winning pitches will lead with resource security — stable grid, low-conflict water sourcing, and credible environmental processes — rather than land and tax breaks alone. The region’s next AI campus map will be drawn by water tables and grid studies as much as by fiber, and every country’s agencies are watching how Indonesia’s halt resolves for signals on where the line sits.

What to Watch as the Rethink Matures

Five signals tell you where this goes next. One: whether BDx’s Jatiluhur permits clear — and on what water conditions; a negotiated water-cap deal would become the region’s template. Two: Thailand’s decision on its paused projects, the second data point on whether pauses become cancellations. Three: Malaysia’s grid buildout pace versus its data center queue, the region’s cleanest stress test of power planning. Four: whether Indonesia formalizes the reservoir-buffer rules that halting Jatiluhur implies — codified water-protection zones would export across ASEAN. Five: pricing — if resource clearance keeps slowing new supply while AI demand compounds, SEA compute prices rise, which is exactly the opening that makes India’s subsidized export pool (covered in our India AI compute explainer) and Batam’s sub-2ms campus strategically interesting.

The regional story of 2026 is now set: Southeast Asia’s AI data center boom continues, but on rations. Water, power, and public consent moved from footnotes to gates — and every Southeast Asia AI data center campus that clears those gates defines the region’s AI capacity for the decade.

Frequently Asked Questions About Southeast Asia AI Data Centers

Why was Southeast Asia AI data center construction halted at Jatiluhur?

Construction of BDx’s 640MW AI campus at Jatiluhur, West Java was stopped on the West Java governor’s order in early October 2026, over outstanding environmental and building approvals and concerns about the campus’s planned water use from the Jatiluhur reservoir, which supplies Greater Jakarta and regional irrigation. BDx says it still targets first-phase operation in early 2027 pending approvals.

How much water does an AI data center use?

It depends on cooling design: evaporative systems consume substantial water per megawatt of capacity, while closed-loop liquid cooling and dry coolers cut direct consumption sharply. This is why operators now differentiate on cooling design — Microsoft’s Hyderabad facilities advertise no-water cooling, and Batam’s new campus leads with immersion systems — while reservoir-front proposals trigger public scrutiny first.

Are Southeast Asia AI data centers still being built?

Yes — the boom bifurcated rather than stopped. Bitdeer signed a 10-year agreement for a 67MW Malaysian campus (A901) in October 2026, Zankore secured $3.1 billion in financing for its NVIDIA-powered buildout toward 1GW, and Indonesia inaugurated a 120MW green AI campus in Batam. The projects advancing share cleared environmental footing and secured grid capacity; contested-resource builds stall.

Which countries are rethinking Southeast Asia AI data center projects?

Indonesia (the Jatiluhur halt plus calls for public environmental assessment), Thailand (projects paused for review), and Malaysia (grid pressure flagged) — with Singapore as the long-running case study of a market that allocates capacity by sustainability score rather than open demand.

Does the power grid constrain AI data centers in Southeast Asia?

Increasingly, yes. BDx alone required 845 MVA of grid capacity from PLN, and Malaysia’s grid strain is documented. A 640MW campus draws on the scale of a mid-sized city, so the binding constraint on new campuses is often transmission and generation headroom long before land or fiber run out.

What is Singapore’s Data Centre Call?

A capacity-allocation process where Singapore provisionally grants data center build rights — 200MW in the second call — to projects meeting sustainability and efficiency criteria. It functions as regional rationing: capacity is awarded on green credentials rather than first-come demand.

What should companies buying SEA AI capacity do differently in 2026?

Contract against delivery milestones and verified grid capacity, not announced megawatts; prefer campuses with cleared environmental footing; and watch resource-politics risk as a line item in any multi-year capacity agreement, because halts like Jatiluhur now sit on every credible risk register.

What happens next for the Jatiluhur project?

BDx is working with Indonesian authorities to complete environmental and building approvals and still targets first-phase service in early 2027. Watch for a public environmental impact assessment and any negotiated water-use conditions — the terms of that resolution will shape how every future SEA AI campus proposal gets judged.

Financial Disclaimer

This article is general information about AI infrastructure trends, not financial or investment advice. Project details, financing terms, and regulatory status reflect reporting cited below as of October 2026 and change often. Verify independently before decisions. WorldNgayon is not a financial adviser.

Sources

Primary reporting: The Straits Times on the Jatiluhur halt and BDx’s 2027 target (October 4, 2026); SCMP’s regional rethink feature; eWeek on Zankore’s $3.1 billion financing and the 1GW plan (October 6, 2026); Bitdeer AI’s A901 announcement (October 8, 2026); Komdigi’s Nongsa Batam 120MW campus inauguration (October 9, 2026, Indonesian-language primary); DCD on the Naver Cloud–FPT AI buildout MoU. Regional strategy context: our Southeast Asia AI infrastructure overview and the AI data center power problem explained; regional skills evidence via the ADB–Global Fintech Institute ASEAN skills framework report. For the supply-side counterpoint, see why India is building AI compute to export.