The Battle for
ASEAN's AI Gravity:
why megawatts stopped being the scoreboard

Malaysia isn't full — it is the tightest data centre market in Asia-Pacific. Thailand's spare power is the wrong kind of power. And "AI investment" is not a line item in anyone's statistics.

In November 2025, Johor's authorities asked investors to postpone water-cooled expansions for roughly eighteen months. In February 2026, residents of Gelang Patah staged Malaysia's first data centre protest. Days later, Prime Minister Anwar Ibrahim told parliament that applications for data centres unrelated to AI had been stopped for almost two years.

Read together, those three events suggest an obvious story: Southeast Asia's most successful data centre market has filled up, and the capital now has to go somewhere else — Thailand, presumably, with its four decades of electronics manufacturing and its notoriously over-built power system.

That story is wrong in an instructive way. Johor is not full. It is the tightest data centre market in Asia-Pacific. And what is happening across the region is not a relay race in which the baton passes from one country to the next. It is a convergence: every government is arriving at the same gate, at different times, and installing the same kind of turnstile.

When we mapped the global stack in The 122 GW Map, the story was ASEAN's sudden promotion into the world's top tier. Fourteen months later the binding question has changed. It is no longer who can attract the capital. It is who can still say yes.

The scoreboard nobody reads

If a market were saturating, you would see it in vacancy. Johor's colocation vacancy rate is 0.7 per cent. Bangkok's is 23.3 per cent.

Johor also leads Asia-Pacific on incoming pipeline capacity at 8,542 MW, carries a total market value of US$39.11 billion — second only to Japan — and its 1,110 MW of live IT capacity ranks third in the region behind Tokyo and Singapore.

What is saturated in Johor is not demand. It is water, grid headroom and social licence — and the government's willingness to spend them on anyone who asks.

Everyone became a gatekeeper

Malaysia did not stop attracting investment. It switched from volume to selection. Johor's ad-hoc Data Centre Development Coordination Committee, formed in June 2024, has reportedly rejected around 30 per cent of applications on sustainability grounds. Fitch's BMI unit now describes Malaysia's roughly 4.6 GW of planned and under-construction capacity as almost entirely AI-driven.

Thailand reached the same decision point in July 2026, about two years behind. Roughly 70 per cent of its data centre investment is concentrated in the Eastern Economic Corridor, and the government is now preparing a "Power and Water Map" to push projects out of it, alongside screening criteria based on readiness, resource availability and national benefit. Readers of The Invisible Ceiling will recognise the pattern: the ceiling we described for chips is now binding on compute, and faster.

Thailand: the power is real, and it is the wrong kind

Thailand's headline energy position genuinely is unusual. Installed capacity is around 53 GW against a peak demand of 35.9 GW — a reserve margin of roughly 27 per cent. Around half of the large gas-fired plants sit idle, and consumers have paid an estimated THB 533.1 billion in availability payments over sixteen years to keep them there. The Thai Data Center Association's framing is that this is a sleeping national asset waiting to be converted into revenue, foreign exchange and tax.

Three caveats turn that argument inside out. The first is that the demand figure everyone quotes is not demand. Thailand's energy regulator has cited data centre power requirements approaching 30,000 MW. The industry's own association says the real number is a fraction of that, and that the 30,000 MW figure is an accumulation of speculative, duplicated and preliminary reservations.

The second caveat is that the surplus is gas, not clean firm power — and hyperscale tenants increasingly buy the latter. Thailand's Power Development Plan 2024–2037 allocates roughly 1.4 GW of new renewable capacity to the eastern region by 2030, against projected data centre demand there exceeding 4.7 GW by 2027. Direct PPA mechanisms were still being debated at a Chulalongkorn University seminar in May 2026. Fewer than 12 per cent of approved EEC projects include on-site generation.

The third caveat is price. Thailand's cheap electricity is being priced away from the sector that wants it. A dedicated data centre tariff of five to six baht per kWh is under consideration, against a general rate of 3.95 baht for the September–December 2026 period. Malaysia has already moved.

Water is the constraint that gets less airtime and moves faster. Roughly 70 per cent of Thai data centre investment sits in the EEC, where a member of parliament has publicly cited water stress and outages and argued that approved projects cannot all be sustainable once fully operational. The World Bank warned in June 2026 about excess power and water cost risk in Thai data centre investment. This is the same physical ceiling we mapped across the chip sector in The Map No One Is Drawing — arriving three years earlier than expected, on a different industry.

The other race: boards, not buildings

Thailand's printed circuit board boom is the more interesting story, and the one most likely to be mis-stated. The Taiwan Printed Circuit Association projects 2026 output of US$6.09 billion for Thailand, US$4.90 billion for Vietnam and US$2.41 billion for Malaysia, with Southeast Asia now accounting for 12.3 per cent of global PCB output value. Thailand's investment board has logged more than 180 PCB-related promotion applications worth over THB 200 billion between 2022 and June 2025 — against roughly THB 15 billion a year in 2021–22.

Here is where the reporting usually breaks down. Neither Thailand's BOI nor the industry association publishes an AI-attributable share of PCB investment. What is published is a mix, and the mix matters: Thailand's new lines are optimised for standardised, high-volume production — multilayer boards, HDI, automotive electronics, server and datacom boards — which is exactly what a large OEM wants when qualifying a second source outside mainland China. That is a China-plus-one story with AI inside it, not an AI story. It is the same distinction we drew in China+1 Is Dead: the unit being relocated is a supply-chain layer, not a country bet.

Vietnam: near-parity in boards, an order of magnitude in megawatts

The instinct is to treat Vietnam as the challenger closing on Thailand. On boards, it nearly has. On compute, the distance is much larger — and it grows when you look at what is committed rather than what is built.

Vietnam is not competing for the same hyperscale anchor tenants. It is legislating a captive domestic market and building sovereign capacity to serve it — Viettel's NVIDIA partnership alone integrates roughly 800 systems and 6,000 GPUs. The binding constraints are the grid, international bandwidth, and accelerator procurement: reported lead times for high-end parts have stretched from around eight weeks to twenty-six. On the underlying fragility of that FDI-led model, see Vietnam's FDI Bet.

Four boards, four different games

The most common analytical error is to rank six countries on one axis. They are competing on four, and each has picked the one where it can win.

Indonesia is betting on scale and domestic demand rather than cost arbitrage. BDx has secured a 1.2 GW power commitment with state utility PLN — the largest held by any operator in the country — anchored by its CGK4 campus, rated up to 650 MW and Indonesia's first NVIDIA DGX-Ready site. Around 580 MW is operational nationally with roughly 1.3 GW more planned; the pipeline is put at US$15–20 billion, and Arm, NVIDIA and sovereign fund Danantara are together targeting some 15,000 trained engineers. The country also has its own Indonesian-language model family, Sahabat-AI. The weakness is visible in Figure 01: Jakarta's 20.5 per cent vacancy.

The Philippines has chosen alignment over economics. It joined the US-led Pax Silica initiative in April 2026 — the second Southeast Asian signatory after Singapore — anchored by a 4,000-acre Economic Security Zone in the Luzon Economic Corridor, billed as the initiative's first "AI-native" industrial hub. Against a DICT ambition of 18 GW over a decade, it currently has around 500 MW across 28 facilities and the region's least competitive power. Domestic critics have argued the arrangement risks locking in exactly that.

Singapore has stopped competing for megawatts and started charging for access to them. Its December 2025 DC-CFA2 round allocates 200 MW against requirements including at least 50 per cent green power and a 1.25 PUE at full load. Meanwhile it has issued conditional approvals for 8.35 GW of low-carbon electricity imports from five jurisdictions — making it the single largest committed buyer of the region's future clean generation, and therefore a shaping force in everyone else's grid planning. This is the geography argument from What Can't Be Rerouted, applied to electrons rather than hulls.

What to watch

The megawatt race is ending, and three scarcer things are replacing it: clean firm power, anchor tenants, and accelerator allocation.

Johor's water deferral expires around mid-2027. If it is extended, the spillover into Thailand and Batam that everyone has been predicting since 2024 finally starts. If it lapses on schedule, Johor's pipeline resumes converting and the regional story stays concentrated.

Bangkok's 23.3 per cent vacancy is the number to track, not Thailand's approval totals. Thirty-four approved projects from 2025–26 are still to energise. Whether that vacancy narrows or widens will tell you whether Thailand has attracted compute or merely attracted construction.

Watch whether Pax Silica membership starts correlating with accelerator access. It is currently a non-binding coordination framework. The moment it functions as an allocation mechanism, Malaysia's and Indonesia's decision to take bilateral deals instead becomes expensive.

No government or industry body in the region publishes an AI-attributable share of either PCB investment or data centre capacity. Until someone measures it, the honest formulation is that Southeast Asia is absorbing an electronics and infrastructure boom in which AI is the largest single driver — not that it is building an AI economy.

That difference will matter a great deal when hyperscaler capital expenditure guidance next turns down.

Related analysis