Bangkok vs Johor vs Singapore: Where to Place Data Center Capacity in 2026

Southeast Asia’s data center decision in 2026 comes down to a triangle: Singapore for network density at $330-475/kW/month, Johor for gigawatt-scale capacity at the region’s lowest cost (now AI-only for new approvals), and Bangkok for a fast-growing, incentive-rich market serving Thailand and Indochina. The standard answer for most operators is a split architecture — latency edge in Singapore, bulk compute in Johor, and Bangkok when the workload or the customer base is Thai. The details, and the February 2026 policy shock in Malaysia, change the math case by case.

The three markets at a glance

Metric	Bangkok (Thailand)	Johor (Malaysia)	Singapore
Live capacity	~216 MW nationally (Q2 2026); Bangkok 145.8 MW + Chonburi 70 MW	Multi-hundred MW live; ~5.8 GW aggregate incl. committed (mid-2025)	~1 GW-class operational hub, minimal new supply
Pipeline	~944 MW under construction/planned; ~2.9 GW total incl. announced	Toward 7 GW (largest in SEA)	Small, sustainability-gated tranches
Colo pricing	Low; well below Singapore (current index)	Region’s lowest for wholesale scale	$330-475/kW/month, avg ~$403
Vacancy	Ample new supply	Tightening for AI-ready blocks	~2% (lowest in APAC)
Policy stance	Actively courting investment (BOI)	Non-AI approvals frozen ~2 years (Feb 2026)	Post-moratorium, efficiency-gated growth
Anchor tenants	AWS ($15B), ByteDance ($4B), Google ($1B)	Microsoft, ByteDance, Oracle, GDS/DayOne et al.	Every major cloud and carrier

Sources: TNGlobal/BMI, DC Byte, IndexBox, Aggasys.

Singapore: the premium hub that stopped growing

Singapore built its position the old-fashioned way: it is the region’s subsea cable landing hub, the densest carrier and cloud interconnection point in Southeast Asia, and the default home of every regional network edge. Then it ran out of room. The 2019-2022 moratorium on new data center approvals was lifted only in favor of a controlled-allocation regime that releases small tranches of capacity gated on energy efficiency and renewable commitments.

The result is textbook supply constraint: vacancy around 2% — the lowest in APAC — and asking rents of roughly $330-475/kW/month, averaging near $403/kW/month, the highest range of any APAC market, per Aggasys’s 2026 Singapore guide. Power is also the region’s most expensive, and sustainability requirements (PUE, renewables) add compliance overhead.

Nobody places bulk compute in Singapore anymore at these prices. What stays is what must stay: network edges, financial-trading infrastructure, interconnection nodes, and regulated workloads with Singapore data-residency requirements. Everything else has been migrating 20 km north across the causeway for five years — which is precisely how Johor happened.

Johor: gigawatt scale meets a policy wall

Johor is Southeast Asia’s capacity story. Aggregate capacity roughly doubled in a year to about 5.8 GW by mid-2025 (live plus committed), with a development pipeline running toward 7 GW — more than every other SEA market combined. The pitch: land and industrial power far cheaper than Singapore, 1-2 ms latency to the Singapore ecosystem, and a state government that moved fast on approvals. Microsoft, ByteDance, Oracle, and the major pan-Asian operators all built or committed at scale, pushing the market past $11 billion.

Then the wall. On February 24, 2026, Prime Minister Anwar Ibrahim confirmed that Malaysia has frozen approvals for new non-AI data centers for roughly the next two years, citing electricity and water strain and the risk of pushing tariffs higher for ordinary consumers (Malay Mail). Anwar noted approvals had informally been limited “for almost two years” already (DCD). Cloud storage, enterprise IT, and CDN-type projects are effectively unapprovable; AI and hyperscale-AI builds still clear (IndexBox). Water is the quieter constraint — Johor’s draw-down has become a live political issue (The Diplomat).

Read this carefully as a buyer, because the freeze cuts two ways. Capacity already approved or built keeps operating and leasing — there is no shortage of AI-ready wholesale space today, and committed supply keeps delivering through 2027-2028. But the two-year approval gap means non-AI capacity tightens from here, and operators holding approved-but-unbuilt non-AI entitlements suddenly own scarce assets. Expect Johor pricing for general-purpose colocation to firm through 2027 while AI campuses keep breaking ground. Browse live Johor facilities in our data center catalog.

Bangkok: the fast riser with incentives to spend

Thailand is earlier in the curve but moving fastest relative to base. BMI counts about 216 MW live nationally as of Q2 2026 with 944 MW under construction or planned — a quadrupling — and the total pipeline including announced projects approaches 2.9 GW (TNGlobal). Bangkok holds 145.8 MW live plus a 902 MW pipeline; Chonburi in the Eastern Economic Corridor adds 70 MW live and a 300 MW pipeline (DC Byte). The anchor tenants are serious: AWS has committed roughly $15 billion, ByteDance $4 billion, Google $1 billion.

The draw is a combination Johor can no longer offer new entrants: an open-door policy plus incentives. Thailand’s BOI grants corporate income tax exemptions capped at 100% of investment value — 8 years for high-efficiency facilities (gated on PUE, water-use efficiency, and local-benefit plans), 5 years standard — alongside import duty exemptions and 100% foreign ownership (Tilleke & Gibbins).

The caveats are power and connectivity. The general industrial tariff sits near 3.95 baht/kWh ($0.12), but regulators have approved the principle of a dedicated data center tariff, with 5-6 baht/kWh ($0.15-0.18) under consideration so grid investment is not socialized onto households (Energy Connects); a 2 GW Direct PPA pilot launched in January 2026 offers a renewable workaround for large buyers. Grid delivery, not generation, is the binding constraint — the transmission “megawatt gap” is real, and Krungsri Research flags both it and Thailand’s thinner subsea cable connectivity versus Singapore as the market’s structural gaps. Bangkok is roughly 30 ms from Singapore — fine for domestic and Indochina workloads, wrong for anything that needs the Singapore ecosystem at single-digit milliseconds.

Cost and policy comparison

Factor	Bangkok	Johor	Singapore
Industrial power	~$0.12/kWh (DC tariff of $0.15-0.18 proposed); 2 GW DPPA pilot	Cheapest of the three; tariff reform pending	Most expensive in the region
Tax incentives	BOI: 5-8 yr CIT exemption, duty-free machinery	Standard incentives; AI projects prioritized	None comparable
New-build policy	Open, actively courting	Non-AI frozen ~2 years from Feb 2026	Tightly rationed, efficiency-gated
Water risk	Moderate (EEC planning)	High — politically live issue	Managed, but a gating criterion
Latency to Singapore	~30 ms	1-2 ms	—
Subsea connectivity	Improving, still thin	Via Singapore + new landings	Regional hub, densest in SEA

Connectivity and latency: the triangle’s geometry

The three markets form a deliberate geometry. Singapore is the hub: the densest subsea cable landing point in Southeast Asia and the region’s richest carrier, cloud, and internet exchange ecosystem. Johor’s entire value proposition depends on sitting 1-2 ms from that hub — close enough that a Johor data hall and a Singapore network edge behave like one campus for most applications. Cross-border dark fiber and diverse causeway routes are now standard offerings from the major Johor operators, and buyers should still contract at least two physically diverse paths, since the causeway crossings are the obvious shared-risk point.

Bangkok plays a different game. At roughly 30 ms from Singapore, it is not an extension of the Singapore ecosystem — it is an alternative center of gravity for Thailand and mainland Southeast Asia (Vietnam, Cambodia, Laos, Myanmar), a combined consumer market of over 250 million people mostly served today from Singapore at 20-40 ms. New subsea landings and terrestrial routes through the Malay Peninsula are steadily improving Thailand’s international connectivity, but Krungsri Research is right that the cable gap versus Singapore remains the market’s structural weakness. For workloads whose users are in Bangkok, Hanoi, or Ho Chi Minh City, 30 ms saved is the point; for workloads serving Singapore, it is disqualifying.

There is also a sovereignty dimension pushing capacity north. Thailand’s draft data-residency and cloud policies, like Indonesia’s before them, favor in-country hosting for government and financial workloads — demand that cannot be served from Johor or Singapore at any latency.

Which to choose if…

• You run latency-critical, interconnection-heavy, or Singapore-regulated workloads → Singapore. Pay the $330-475/kW and keep the footprint minimal: network edge, exchange connectivity, data-residency workloads only.
• You are deploying AI training or large-scale inference → Johor. AI builds still get approved, the wholesale cost per kW is the region’s lowest, and you keep 1-2 ms adjacency to Singapore. Move fast on grid-secured, water-secured campuses; those are the scarce ones. For GPU economics versus renting capacity, check our GPU price tracker.
• You need general-purpose (non-AI) capacity in the region → Bangkok, or existing Johor space. New non-AI approvals in Johor are frozen; Thailand is the open market with 8-year tax holidays and nearly 3 GW of pipeline. Existing approved Johor space works but expect firming prices.
• Your customers are in Thailand or Indochina → Bangkok, full stop. Local capacity, BOI incentives, and hyperscaler-grade supply arriving through 2028.
• You are hedging policy risk → split Johor + Bangkok. Malaysia’s freeze showed how fast SEA policy can turn; Thailand’s proposed DC tariff shows cost certainty is not guaranteed either. Two-market deployments with a Singapore edge are becoming the default enterprise architecture.

Benchmark rates for all three markets are tracked in our colocation price index, and facility-level detail for 202 facilities is in the catalog. If you want current quotes from operators in any of the three markets, request a quote and we will run them against index data.