The Global "Reliable Power" Shockwave. Article 2 of 2: ASEAN Power Situation to 2028 and Beyond

Subtitle: Singapore-Johor-Batam-Jakarta-EEC: New 24/7 capacity or growth bottleneck?

James Soh. First published on 16th of January, 2026.

Article 1 shared the race by U.S. hyperscalers to restart retired reactors—such as Microsoft’s Three Mile Island deal (835MW 2027), Meta's 6.6GW nuclear hunt, and Amazon/Google reactor contracts—because "speed-to-power" trumps everything else in the AI race.

Southeast Asia faces identical pressure: regional data centers are projected to explode from 2.6GW today to 10.7GW by 2035, claiming 3–4% of total regional peak demand. Individual racks are now pulling an average of 106MW—a massive jump from the 20–30kW seen just three years ago—with annual electricity costs reaching $10 billion USD.

The core question isn't just "growth," but survival through evening peaks when solar disappears while AI inference runs 24/7. Will each hub deliver enough NEW reliable 24/7 capacity by 2028, or will power constraints throttle their expansion plans? 

AI Workloads: Different Beasts, Same 24/7 Crisis

To solve the power crunch, we must distinguish between the two types of AI workloads:

• Training AI (Building Models): These are absolute "power monsters," drawing 30–150kW per rack. They are relatively latency-tolerant and belong in massive-scale campuses like Johor and Batam where land and grid connections are more abundant.

• Inference AI (Running Models): This is the "volume beast" used for real-time queries. It requires constant 24/7 power and ultra-low latency (<100ms), anchoring these workloads close to end-users in hubs like Singapore’s financial district and Bangkok.

Singapore: The High-Value Pivot and Jurong Island Bet

Singapore’s approach is the most nuanced in the region. The city-state is not racing for raw scale; instead, it targets high-value economic add, such as the AI+ workforce, corporate HQs, and associated financial activities. Because large-scale AI and hyperscale data centers do not directly create a high volume of high-value jobs and the nation is land-scarce, the DC-CFA2 and overall power capacity will not be increased indefinitely to satisfy every data center demand.

• Jurong Island as a Pilot: The move to Jurong Island is a considered pilot to transition the area to green power. Data centers are positioned to augment or even replace the oil and chemical industry on the island, which is predicted to decline over the next one or two decades.

• The Strategy: By allocating 20 hectares for a low-carbon data center park (~700MW potential), Singapore is testing waste heat recovery and hydrogen/ammonia fuel blends.

• CFA2 Constraints: The Dec 2025 call for 200MW new capacity enforces strict rules: 50% green energy sourcing and a PUE ≤1.25. At the world’s highest tariff of $154/MWh, Singapore is betting that premium reliability and high-value integration will retain critical inference workloads.

Malaysia/Johor: Explosive Scale and TNB Momentum

Johor has transformed from 10MW of capacity in 2021 to 1.5GW IT load by late 2024—a 150x growth in three years. It is now Southeast Asia’s premier AI training hub.

• Momentum: TNB secured a 280MW dedicated capacity agreement for TM Nxera in Dec 2025.

• Pipeline: A 1GW pipeline is expected to be operational by end-2026, including Microsoft SEA3 and Google Elmina.

• Risk: While the $133/MWh tariff is competitive, the 58% coal dominance remains an ESG hurdle for green AI financing.

Thailand: EEC Core and the 2GW Direct PPA

Thailand is moving aggressively with a split strategy: EEC (Chonburi/Rayong) for industrial scale and Pathum Thani/Bangkok for urban edge.

• New Capacity: BOI approved $3.1B across 376MW in late 2025, including Zenith’s 200MW hyperscale in Pathum Thani and Digital Edge’s 100MW AI-ready campus in the EEC.

• Policy Innovation: The 2GW Direct PPA pilot (Jan 2026) mandates 100% renewables and battery storage (BESS) for 50MW+ loads. This is critical as evening reserves are projected to drop below 15% by 2028.

Indonesia: Jakarta Urban and Batam Pressure Test

Jakarta leads Indonesia’s growth with a 709MW pipeline, while Batam serves as the capacity test case for the SJB (Singapore-Johor-Batam) triangle.

• Batam’s Race: PLN aims to deliver 369MVA by 2026, and a Maxpower 150MW gas plant is expected by 2027.

• Shortfall Risk: While Batam is safe through 2028, a shortfall crisis looms in 2029 if cumulative demand hits 1,000MVA. Without local hydro options, it remains reliant on gas and potentially Sumatra hydro imports via subsea cable.

The Tropical Pitfalls: Solar+BESS and Direct PPAs

While green energy is the goal, the tropical reality of Southeast Asia introduces significant pitfalls for operators:

• Monsoon Reality: The rainy season (Oct–Apr) can slash solar output by 40–70%. Passing clouds cause instant drops in power, and BESS typically only handles hours of storage, not the weeks of reduced capacity common during monsoons.

• Direct PPA Traps: Operators in Direct PPA schemes face curtailment exposure—the grid takes priority during peaks—and price volatility, where solar costs can jump 20–40% during rainy months.

• Degradation: High heat and humidity can reduce BESS operational life by 20–30%.

The Nuclear Frontier: SMR "Buffer Zone" Risks

Small Modular Reactors (SMRs) are a long-term interest for the region, but they highlight a major geographic risk for Singapore.

• Land vs. Safety: Thailand and Malaysia have the vast industrial zones required for the several-kilometer safety perimeters needed for nuclear emergency planning.

• Singapore’s Constraint: For a densely populated city-state, the "buffer zone" is an existential bottleneck. Finding the radius required for nuclear safety without overlapping residential areas remains a primary puzzle. Unless "zero-buffer" designs emerge, Singapore may remain dependent on power imports while its neighbors build sovereign nuclear capacity.

What to Monitor: Government Power Directives

Since power is government-controlled, these are the signals that determine which hubs stay investable:

Series Summary: Hub may be slow to bring new, 24/7, reliable capacity on-stream by 2028, which will affect its ability to meet data center expansion needs. Speed-to-power is the new currency; those who solve the reliable power capacity issue first will capture the AI billions.

Note: The values (all currency values are in USD) provided in this article reflect average on-grid electricity costs specifically for data centers rather than peak-only or general industrial rates. While Malaysia has historically been known for its competitive energy pricing, the most recent data for 2025–2026 shows a shift in the regional hierarchy:

• Thailand ($108/MWh): Currently offers the region's lowest tariff for data center operators. This is driven by high reliance on natural gas (two-thirds of the mix) and a new 2GW Direct Power Purchase Agreement (DPPA) pilot launching in January 2026.

• Malaysia ($133/MWh): While still competitive compared to Singapore, this rate reflects recent adjustments, including a hike in the base tariff for "Regulatory Period 4" (2025–2027) that took effect in July 2025.