Southeast Asia is building nuclear reactors for AI data centres

Southeast Asian governments are actively exploring small modular reactors to power the region’s expanding AI and cloud infrastructure, a shift from decades of nuclear opposition driven by hyperscale data centre demand. Vietnam’s prime minister met Swedish SMR developer Blykalla in 2025, and Singapore is creating a blended-finance model to share the costs of first-of-a-kind projects.

The move targets a specific electricity gap: intermittent renewables cannot supply the round-the-clock power that data centres require. Malaysia’s data centre pipeline alone now exceeds 13 gigawatts, larger than Indonesia, Thailand and Singapore combined, intensifying pressure for firm low-carbon sources.

Vietnam’s prime minister visited a Swedish nuclear startup in 2025. It was the first time a Southeast Asian head of government had met a small modular reactor developer. The reason was not a broad energy transition. It was the coming surge in AI data centres, which demand round-the-clock power that renewables alone cannot deliver. The region has never operated a commercial nuclear reactor. Yet governments are now positioning SMRs not as a national electricity source but as dedicated infrastructure for the digital economy.

The region’s electricity grid is still about 70% coal and gas. Solar and wind, though expanding, remain unreliable for facilities that cannot lose power for more than a few seconds a year. Data centre operators are signing leases across Johor, Batam, and Ho Chi Minh City. They cannot wait for a grid upgrade that may take a decade. The result is a specific, and historically unusual, push: nuclear power built for the cloud.

The supply gap that forces a nuclear rethink

The scale of the demand is startling. Malaysia’s data centre pipeline alone amounts to nearly 13 gigawatts of IT load capacity, larger than the combined pipelines of Indonesia, Thailand and Singapore, JPMorgan estimates show. Average development costs hover around US$7 million per megawatt there, compared with roughly US$12 million in Singapore. According to JPMorgan’s equity research team, Singapore optimises its scarce resources while Malaysia captures the next wave of incremental capacity. The pressure to find firm, low-carbon electricity is immediate.

China’s Linglong One, a grid-scale SMR of about 125 megawatts, became the first of its kind in commercial operation in early 2026. That milestone gives Southeast Asian planners a reference for costs and performance. The World Nuclear Association projects global SMR capacity could reach about 158 gigawatts by 2040 and surpass 400 GW by 2050. Under a net-zero-aligned scenario, Wood Mackenzie sees nuclear capacity in Southeast Asia reaching 10 to 20 GW by mid-century, with Vietnam and Indonesia hosting most of it.

Data centres are now the single largest private-sector driver of planned SMR capacity globally, according to energy analyst Aidan Jad at Axis Intelligence. He notes that every major Western SMR project has already experienced at least one schedule slip. Early deployments like Canada’s BWRX-300 are therefore critical: if they overrun costs or miss timelines, the replicability that Southeast Asian governments are banking on may not materialise.

The IAEA’s milestones guidance warns that building a nuclear regulator from scratch, plus waste and emergency systems, typically takes a decade. Vietnam, Indonesia and Malaysia all base their nuclear policies on this IAEA Milestones Approach but are at different stages of readiness. None yet has SMR-specific licensing rules.

For a hyperscale operator planning a 100 MW campus in Johor, the mathematics of uptime is unforgiving. The available alternatives are diesel backup, coal-fired power, or a technology that does not yet exist commercially in the region. The economics of a power purchase agreement backed by an SMR start to resemble a future more than a fantasy.

The mismatch between the grid’s present capabilities and the performance data centres require is not hypothetical.

The grid is losing the race with data centre demand

The existing grid in Southeast Asia loses between 6% and 10% of electricity in transmission, and investment still lags where it needs to be. Hyperscale facilities cannot operate on a grid that fails. The IEA counted up to 25 GW of planned global SMR capacity in 2025, and about 30 GW of SMR-linked data centre agreements and expressions of interest. The pull from tech firms is already shaping the nuclear pipeline.

The regulatory gap is stark. The US Nuclear Regulatory Commission and EU regulators already run design certification and site licensing processes for advanced reactors. In contrast, Vietnam’s reactivated nuclear planning still relies on a legal framework that must be strengthened. The IAEA Milestones Approach is a pathway, but it is not a shortcut. A government that commits today would not see a commercial SMR before the mid-2030s at the earliest.

The hyperscalers pressing for SMRs will not be the ones licensing the reactors or storing the spent fuel. Those tasks fall to governments that are still building nuclear regulators from scratch. If Southeast Asia succeeds in pairing SMRs with data centres, Western nuclear vendors and financiers gain a major export opportunity. But the same success could draw AI compute workloads away from Europe, concentrating digital infrastructure where electricity is firm and low-carbon. The competition with Chinese and Russian reactor suppliers adds a geopolitical filter to what would otherwise be a straightforward procurement decision.

The pivot to nuclear is not a choice born of ideology. It is the consequence of a simple, arithmetic gap between what data centres demand and what the current grid can supply. The region has no commercial reactors, no SMR-specific regulations, and no track record of managing nuclear waste. The hyperscale leases signed this year, however, will not wait for the decade it takes to build a regulator.

Beyond the headline

The bigger picture

The SMR push for data centres crystallises a broader shift in how emerging economies think about industrial power: instead of treating nuclear as a general grid resource, governments are now weighing it as a sector-specific enabler for digital infrastructure. That reframes nuclear from a legacy baseload debate into an instrument of competitiveness in the global AI race, where reliable low-carbon electricity becomes part of the location calculus for cloud platforms and chip-heavy computing clusters.

The money trail

Hyperscalers’ need for firm, green power is quietly restructuring who finances nuclear risk. Rather than utilities alone carrying multi-decade balance-sheet exposure, blended models in Singapore and beyond envisage government-backed funds, multilateral banks and tech firms sharing first-of-a-kind SMR costs. That alignment could turn data centre power purchase agreements into de facto anchors for new nuclear asset classes, with returns shaped as much by computing demand growth as by traditional electricity tariffs.

The reach

One non-obvious ripple effect runs through European and US climate policy: if Southeast Asia succeeds in pairing SMRs with hyperscale data centres, it reduces the region’s reliance on coal-heavy grids and weakens arguments for strict carbon border adjustments on digital services. That, in turn, may influence how Western regulators design emissions accounting for cross-border data processing, potentially easing compliance costs for tech firms sourcing compute from low-carbon Southeast Asian hubs instead of domestic facilities.

Three decisions for Western firms watching the SMR race

With the first Southeast Asian SMR deployments unlikely before the mid-2030s, Western companies face choices that will shape their exposure to this market for years.

• Western SMR technology developer or investor: You need to map regulatory readiness in Vietnam, Indonesia, and Malaysia, paying particular attention to licensing gaps and upcoming PDP8 revisions. Early engagement with government-linked entities and potential offtakers will be critical, as will a clear-eyed assessment of how Chinese and Russian vendor competition may affect deal terms.
• Hyperscale data center operator with APAC presence: Begin talks with regional SMR developers and Singapore’s green finance platforms now. Power purchase agreement structures that incorporate SMR risk-sharing are still being designed; early involvement will let you shape the templates rather than inherit them. Site selection in Johor, Batam, or Ho Chi Minh City should factor in the realistic timeline for firm low-carbon power.
• Western policy professional focused on Indo-Pacific energy security: The success or failure of SMR adoption will affect regional power balances and supply chain resilience. Watch whether multilateral development banks join Singapore’s blended-finance vehicle, and track Vietnam’s PDP8 revision for signals of nuclear’s real weight. Both will indicate how quickly this market could solidify—and whether Western or Chinese vendors are positioned to lead it.

Explainer

Small modular reactors (SMR)

Nuclear reactors producing up to 300 MW per unit, designed for factory fabrication and modular deployment. SMRs offer a higher capacity factor than intermittent renewables, making them suitable for baseload applications like data centres. Unlike large reactors, they are meant to be scalable and have lower upfront capital costs.

Capacity factor

The ratio of actual output over time to the maximum possible output. A 90% capacity factor means a reactor runs at full power 90% of the year. Solar and wind typically have capacity factors of 10–30% depending on location, which is one reason hyperscalers look to nuclear for round-the-clock power.

IAEA Milestones Approach

A framework developed by the International Atomic Energy Agency that guides countries through three phases of building a nuclear power programme. Phase 1 involves establishing national position and infrastructure; Phase 2 covers preparatory work and contracting; Phase 3 leads to the first concrete pour. Each phase typically takes three to five years, meaning a country starting from scratch may need at least a decade before commercial operation.

Power Development Plan 8 (PDP8)

Vietnam’s national electricity generation plan for 2021–2030 with a vision to 2050. PDP8 sets targets for coal, gas, renewables, and now includes nuclear as a potential post-2035 option. The plan originally faced delays due to high coal reliance and funding gaps, and the next revision is expected to clarify nuclear’s precise role.

Blykalla

A Swedish nuclear startup developing a lead-cooled small modular reactor design. Blykalla’s concept emphasises safety and simplicity, using natural circulation of coolant. Founded in 2013 as a spin-off from KTH Royal Institute of Technology, it has engaged with Vietnam after the prime minister’s 2025 visit, though no formal agreement has been announced.