Southeast Asia AI Infrastructure Race: Who Is Leading, and on What Counts

Southeast Asia AI infrastructure has a leader on every count — and no single country holds them all. Singapore leads the money and the rules: Tracxn puts about 99% of the region’s disclosed AI-infrastructure funding on the city-state, and it hosts more than 60 corporate AI centers of excellence. Malaysia leads the build: Johor and Selangor already carry roughly 8.3 gigawatts of demand from 59 projects, and more than half of all under-construction regional capacity. Indonesia leads the offshore-compute lane, Thailand the fast-follower pace, the Philippines the growth-rate sprint, Vietnam the foundations. This piece is the standing verdict that keeps updating as the map changes — the October week below is where every claim is dated, named and priced.

By 2028 the leaderboard flips if the projections hold: regional capacity rises from about 1,677 megawatts in 2023 toward 7,589, Singapore’s share falls from roughly 59% to 19%, Malaysia’s climbs to 42% and Indonesia’s to 20%. That single sentence is why “who is leading” is the wrong-shaped question. Southeast Asia AI infrastructure is six different races on one shared map — and the country that solves the three hard constraints first will be the one the 2028 map belongs to.

Key Takeaway

• 🏁 Five races, one regional scoreboard: Singapore leads money and rules (99% of disclosed $1.2B funding, 60+ AI CoEs), Malaysia leads build-out scale (8.3GW of demand in Johor-Selangor, 42% projected capacity share by 2028), Indonesia leads offshore compute, Thailand leads fast-follower approvals, the Philippines leads growth rate.
• ⚡ The constraint scoreboard decides 2028: regional capacity heads from ~1,677MW (2023) to ~7,589MW (2028); power queues, water thresholds and transformer lead times are the real grading system.
• 🏦 Financing tells the truth: a signed $3.1B senior term loan (Indonesia), state approvals (Thailand), corporate balance sheets (Singapore) and local equity (Philippines) are four different kinds of certainty — dates and instruments belong in every sentence about the race.
• 💧 Water is the newest choke point: Singapore’s PUB requires water-usage effectiveness of 1.2 or below, and Johor already rejects or delays builds that fail electricity and water-efficiency checks.
• 🇵🇭 The Philippines holds the second-highest capacity growth rate in the region (59% CAGR through 2028) — the sovereign entry is real; converting it into gigawatts is the open question.

The Southeast Asia AI Infrastructure Race: the Six-Track Standings

Judge Southeast Asia AI infrastructure as six tracks, each with a current leader and a measurable margin. The standings below are the heart of this piece: they get revised when a track’s leader changes, and every number traces to a named source.

Read the table as standings, not predictions. The 2028 capacity projections — Singapore to 19%, Malaysia to 42%, Indonesia to 20% of the region (Speeda/ASEAN data) — change who leads the build track, but they do not touch Singapore’s money-and-rules lead, and they hand the growth-rate title to the Philippines at 59% against Malaysia’s 63%. Different races, different winners, one map: that is what makes this region different from China’s single-scale model or America’s corporate-scale race we covered separately.

The capacity map: today against 2028

The projections behind the flip in the Southeast Asia AI infrastructure race come from regional market reporting (Speeda/UB-speeda data), and they line up with the DC Byte construction numbers reported across ASEAN outlets.

Two disciplines keep this table honest. First, hyperscaler attribution: Google’s Malaysia commitment is US$2 billion, Amazon’s is the US$6 billion figure often misattributed to Google, and ByteDance plans about US$2.13 billion for its Johor AI hub — the same rule we apply in every piece: verify the company behind each figure against its own announcement. Second, investment totals are announced values, not energized capacity; the standing AI data-center power pipeline piece tracks the gap between the two across the region.

On what counts: power, water and talent decide

Every megawatt converts into the same three inputs — reliable electricity, cooling water, and people who can run the plant. That is why the Southeast Asia AI infrastructure race’s real grading system is not headline dollars: it is loan tenor, grid position and water efficiency.

Power. Thailand alone has around 10GW of data-center projects waiting for grid connections, with approval timelines stretching toward two years and transformer deliveries on similar schedules; its utilities are spending on transmission upgrades to catch up. Singapore turned scarcity into policy years ago — the cap on greenfield data centers is the single decision that pushed the region’s GPU dollars into Johor, Batam, Pathum Thani and Central Java. Malaysia is already applying the lesson: Johor rejects or delays builds that miss electricity-efficiency requirements.

Water. The newest choke point in the Southeast Asia AI infrastructure race: Singapore’s PUB requires new data centers to hit water-usage effectiveness of 1.2 or below before approval, and water-stressed Johor faces the same discipline as its own builds scale. We maintain a dedicated water-wall brief on this track — AI servers drink first, and the countries that price water honestly will keep attracting the lenders.

Talent. Jobs follow the megawatts: construction, electrical, cooling and network-operations roles hire years before AI-researcher roles do. Vietnam’s play is the explicit counter-example — invest in regulation and talent early and accept a slower build — and the migration of skills toward whoever energizes capacity first is a standing question our own archive tracks by country.

The country deep-dives behind these standings — Singapore, Indonesia, Thailand, Vietnam, Malaysia — carry the policy detail; this piece keeps the scoreboard.

The October week that showed the pattern

The standings in the Southeast Asia AI infrastructure race were not built in a quiet week. In early October 2026, five countries moved in the same seven days — and the instruments they chose are the pattern in miniature, which is why we preserve the full detail here.

Read the table as one auction for Southeast Asia AI infrastructure. The countries are bidding for the same pool of GPU dollars, and the bidding has moved from press releases to loan documents. Zankore’s financing is the proof: a $3.1 billion senior term loan facility, with Citi as exclusive debt adviser, means banks now underwrite Southeast Asia AI infrastructure the way they underwrite ports and power plants.

Indonesia’s $3.1 billion loan: the anchor bet on offshore compute

Zankore is the anchor bet of the Southeast Asia AI infrastructure race: the AI infrastructure platform launched by Indosat Ooredoo Hutchison with Ooredoo Group, Nvidia and Nokia. In August the group announced a target of up to 1 gigawatt of Nvidia DSX AI Factory capacity. The loan moves from target to concrete. About 100 megawatts come first, serving AI companies, enterprises, startups, developers and institutions inside Indonesia and across Southeast Asia. The first 200 megawatts are slated for the first half of 2027, powered by Nvidia GB300 NVL72 systems, with construction in Batang, Central Java, targeted to begin in the first half of 2027.

The Indonesian numbers sit inside a national ambition. Jakarta has tied the project to its goal of becoming a regional AI hub, and analysts have tracked national data center capacity racing past 2 gigawatts by 2030. The constraint is no longer appetite; it is power, land and predictable permitting. Batam holds the structural hand: a 7% local growth rate, proximity to Singapore’s capped market, and room for $15-20 billion of dedicated compute on the island — the Nvidia-Firmus 360MW liquid-cooled build being the anchor. One history lesson warns against squandering it: decades ago Penang captured the assembly-and-test business that Batam’s geography had courted, which is why Indonesian industry bodies now lobby for single-window licensing and binding regulatory timelines. Our Indonesia startup map tracks the domestic demand side of the same bet.

Thailand: the state buying speed

Thailand answered the Southeast Asia AI infrastructure race with the region’s fastest-moving approval machine: administrative horsepower. The Board of Investment approved four data-center projects worth roughly US$3.1 billion: NextGen Data Centre and Cloud Services, a subsidiary of Dubai’s DAMAC Digital, an 84-megawatt facility at Navanakorn Industrial Estate in Pathum Thani valued at THB 26.7 billion (about US$826 million); local investor Zenith’s 200-megawatt hyperscale facility on the same estate at THB 54.9 billion (about US$1.7 billion); Japan’s Telehouse with a 12-megawatt Bangkok project (THB 7.55 billion, about US$234 million); and Vistas Technology — a subsidiary of China’s ZDATA — building 80 megawatts in Amata City Chonburi (THB 9.9 billion, about US$306 million).

The parallel move matters more: six new licences to expedite roughly US$9.2 billion in previously stalled projects, aimed at bottlenecks in power supply, industrial land access and visa and work-permit processing. And the scale keeps coming — later in September the BOI approved a further six-project tranche worth about US$29 billion combined, led by TikTok’s planned THB842 billion (~US$25 billion) storage-and-processing buildaround Bangkok. That is fast-follower behavior with real money: approve fast, absorb spillover, and let the grid catch up — the 10GW connection queue being the check on the whole strategy. Siam.AI Cloud’s Thailand plan reads the same way: 70 billion baht this year, 10 billion baht already invested, and an Nvidia GB200 NVL72 supercomputer module — Siam.AI says it is the first Nvidia Cloud Provider in Asia running that module — hosted at ST Telemedia’s STT Bangkok 1 facility. The country detail lives in our Thailand strategy brief.

Singapore: a different game — silicon, standards and capital

While its neighbors court GPU megawatts, Singapore answered with silicon and standards. The Singapore Economic Development Board confirmed that Vanguard International Semiconductor, a TSMC affiliate, will build a $7.8 billion semiconductor plant with NXP Semiconductors. Details remain thin — no construction timeline, capacity figures or process node disclosures came with the announcement — but the signal is not subtle: the city-state is doubling down on being the region’s stable, geopolitically neutral manufacturing anchor while it caps greenfield data center growth. Its second Data Centre Call for Application adds at least 200 megawatts of capacity, explicitly balancing AI demand against energy-efficiency and water thresholds, and January 2026 brought the S$1 billion+ public AI research commitment over five years. On the money-and-rules track of the Southeast Asia AI infrastructure race, that lead is not a vibe – it is arithmetic. Add the Tracxn finding — roughly 99% of the region’s disclosed AI-infrastructure funding concentrated there — and Singapore’s track lead on money-and-rules is not a vibe, it is arithmetic. The city-state is not trying to win the megawatt auction. It is trying to host the chips, the standards and the capital markets that the megawatt countries need — the full detail sits in our Singapore infrastructure brief.

The Philippines: sovereignty now, scale as the open question

President Ferdinand Marcos Jr. visited a newly opened AI-ready data center in Angeles City, Pampanga, operated by Converge ICT Solutions — the government’s data-sovereignty and digital-economy push made physical. The Philippine entry is smaller in dollar terms than Indonesia’s loan or Thailand’s approvals, but it carries the domestic political weight of an actual ribbon-cutting: local enterprises and government bodies getting compute on Philippine soil instead of queues in Singapore. The scoreboard’s surprise is the growth rate: the Philippines projects the region’s second-highest capacity CAGR through 2028 (59%), on a base built for catching up. The pattern to watch is whether power pricing and land availability convert the sovereignty narrative into gigawatt commitments in 2027, or whether the country remains a services and talent story feeding its neighbors’ data centers. Our Philippine data-center REIT brief tracks the capital-markets side of that question.

The supply chain is the connective tissue

None of these Southeast Asia AI infrastructure plans build themselves. Nikkei Asia reported in the same October week that TSMC and other Taiwanese companies have stepped up overseas investment – the connective tissue of the Southeast Asia AI infrastructure race — the same hyperscaler capex wave we tracked on the Nvidia capex wall — with spending concentrated in the US and Southeast Asia as supply chains rush to handle surging AI demand. The Taiwan connection cuts both ways for the region: it supplies the hardware that makes national ambitions real, and it concentrates the region’s dependencies. Whoever hosts the next-generation modules first — GB200 in Bangkok now, GB300 in Indonesia by 2027 — holds a real advantage for enterprise workloads that care about latency and data residency.

The money scoreboard: how Southeast Asia AI infrastructure gets financed

Look at how each project is moneyed. Indonesia: a senior term loan with a global bank advising — institutional debt. Thailand: Board of Investment approvals that unlock tax breaks, land privileges and expedited permits — state-backed attraction of mostly private capital. Singapore: a fab commitment by two established chipmakers inside a stable manufacturing base — corporate balance sheets. The Philippines: a ribbon-cutting on domestic capital from a listed telecom-infrastructure company — local equity. Four different capital structures, one shared race, and each betrays what its backer believes the risk actually is.

That framing matters for anyone tracking Southeast Asia AI infrastructure numbers: a $3.1 billion loan facility, a $9.2 billion unblocking program and a $7.8 billion fab are three different kinds of certainty. The loan is signed intent with a schedule. The unblocking program is administrative hope with a target. The fab is a commitment whose timeline remains undisclosed. Treating them as interchangeable dollars is how regional tech coverage inflates — and why dates and instruments belong in every sentence about the race.

WorldNgayon Insight: what the race is really about

What changed: Southeast Asia AI infrastructure moved from app-layer ambition to megawatt math, with banks setting the tempo. Why now: Singapore’s greenfield cap created the spillover, the hyperscaler price war made compute the binding constraint, and IDC now projects global AI-infrastructure spending near $497 billion for 2026 — Southeast Asia is the fastest-growing beneficiary lane. Why it lasts beyond the news cycle: capacity, power queues, water thresholds and talent pipelines are structural — they decide where workloads sit for a decade.

Who is affected: enterprises choosing data residency, developers pricing latency, professionals choosing where data-center skills pay. What to do next: track the three checkpoints below instead of headline dollars, and use the six-track standings — not country pride — to place a bet. Risks, opportunities, costs, skills: grid queue and water stress are the risks; the arbitrage is in the spillover lanes; the cost of being late is renting your own regional capacity; the skill demand is shifting toward data-center trades and energy engineering. Connection to work and money: the compute map of 2028 is also the hiring map — and the cooling and water jobs lane is already visible.

The verdict, stated once: Southeast Asia AI infrastructure has six leaders because it has six races — and the single country that solves power, water and talent first will absorb the crown the others are sharing. Watch the constraints, not the ceremonies.

What it means for Filipino readers

Thailand’s BOI explicitly tied its approvals to employment and broader economic development. Indonesia’s ministry describes Zankore as a step toward the regional hub ambition, projected to reach up to 1 gigawatt within three years. For Filipino engineers, technicians and data professionals, the signal is directional: the compute jobs of 2027-2030 are being located now, and the region that wins capacity will import talent as readily as power. The OFW read is the same one that applies to every infrastructure boom: these projects hire construction workers, electrical and cooling engineers, security staff and network operators long before they hire AI researchers, and the window to re-skill into data-center-adjacent trades is open while the concrete is still wet.

For the Philippines, the strategic question stays blunt: Southeast Asia AI infrastructure momentum is being financed in signed documents, and the countries writing them will set latency prices in 2027. The Converge facility in Pampanga gives the country its first real entry ticket into that conversation, and the 59% growth-rate projection makes it the fastest-rising challenger by percentage. Converting sovereignty into gigawatt-scale commitments — the way Indonesia converted an August announcement into a $3.1 billion October loan — is the difference between hosting national workloads and renting regional capacity forever.

What changes the standings: the checkpoints and triggers

This is a standing piece — the standings update when reality moves, and the triggers are public:

• The three October checkpoints: whether Zankore’s first 100 megawatts energize on schedule; whether Thailand’s expedite licences move the $9.2 billion from stalled to steel; whether the Vanguard-NXP fab discloses a construction timeline that anchors Singapore’s silicon play in reality.
• The 2028 share projections (Singapore 19% / Malaysia 42% / Indonesia 20%) — any quarter’s live-capacity print that diverges from the projection curve re-ranks the build track.
• Grid, water and policy events: Thai grid-connection approvals, PUB’s WUE threshold revisions, Johor efficiency rulings, Malaysia’s next investment round, ASEAN AI-governance moves — each one can move a track’s leader. Our ASEAN AI summit brief follows the governance lane.

Frequently Asked Questions

Who is leading the Southeast Asia AI infrastructure race in 2026?

There is no single leader – the region runs six tracks. Singapore leads money and rules (about 99% of disclosed AI-infrastructure funding and 60+ corporate AI centers), Malaysia leads build-out scale (more than 50% of regional under-construction capacity), Indonesia leads offshore compute (Batam), Thailand leads fast-follower approvals, the Philippines holds the second-highest growth rate, and Vietnam lays foundations.

How much data-center capacity does Southeast Asia have today?

Regional computing capacity has already surpassed about 4.4 gigawatts, with market reporting projecting the 2023 base of roughly 1,677MW rising toward 7,589MW by 2028 – a regional build-out measured in gigawatts, not megawatts.

Why is Malaysia winning the build-out?

Johor and Selangor combine cheap land and power, streamlined permitting and proximity to Singapore’s capped market – 59 projects with roughly 8.3GW of power demand are already in the pipeline, and DC Byte counts more than half of five-country under-construction capacity in Malaysia.

Can Singapore keep leading without new megawatts?

On the money and rules tracks, yes – its funding dominance, 60+ AI centers of excellence, the S$1B+ research program and standards like the 1.2 WUE threshold are independent of raw capacity. On the build track, its projected share falls to 19% by 2028.

Where does the Philippines stand?

The country holds the region’s second-highest capacity growth projection (59% CAGR through 2028) and a real sovereign entry point in Converge’s Pampanga facility. The open question is converting that rate into signed gigawatt-scale commitments the way Indonesia converted its loan.

What is the single biggest constraint on the race?

Power first (Thailand’s ~10GW grid queue, two-year approval timelines), water second (PUB’s WUE 1.2 requirement, Johor’s efficiency rulings), talent third – the winner of the race is whoever clears all three, not whoever announces the most.

Which investment figures should builders verify themselves?

Every named company figure: Google’s Malaysia commitment is US$2 billion (the US$6 billion figure belongs to Amazon), ByteDance plans about US$2.13 billion in Johor, Microsoft US$2.2 billion for its Malaysia AI center and US$1.7 billion for Indonesia, AWS US$5 billion committed to Thailand plus a planned $12 billion expansion – always to the company’s own announcement, not secondary reporting.

What should I watch next?

Three October checkpoints – Zankore’s first 100MW energizing, Thailand’s $9.2 billion moving from stalled to steel, and the Vanguard-NXP fab timeline – plus any divergence of the quarterly live-capacity prints from the 2028 projection curve.

Financial Disclaimer

AI World This Week-style regional coverage of Southeast Asia AI infrastructure involves investment figures, loan values and market projections — this article is editorial analysis, not investment or financial advice. Announced values are not guarantees of energized capacity; verify terms, timelines and company commitments against primary sources before any capital or career decision.

Primary reporting followed for the October week: eWeek on Zankore, Thai Times on the BOI approvals, Die Signal on the Singapore fab, Manila Times on Pampanga, and Thai Times on Siam.AI — each linked figure dated in the text above. Standing-coverage sources for the scoreboard: Tracxn data reported by Southeast Asia Infrastructure; capacity projections and country roles reported by Speeda ASEAN; DC Byte construction data reported by Malay Mail and AFP; hyperscaler investment mapping in the EDB’s State of AI in Southeast Asia; and the power-water-talent race framework as reported by Asia Econ News. Figures are as announced or reported; no capacity estimates were added where primary sources did not disclose them.