Southeast Asia is in the middle of an infrastructure build-out that will reshape its electricity systems within a single decade — and most of the region’s climate commitments were written before anyone modelled it properly.
Data centre power consumption across ASEAN sat at roughly 9 terawatt-hours in 2024. Current projections put it at 68 TWh by 2030. That is close to a sevenfold increase in six years, arriving in economies that are simultaneously trying to phase down coal, expand electricity access, and industrialise.
Capacity is forecast to grow at roughly 19–20% annually through the second half of the decade, with regional data centre investment projected to reach USD 30 billion by 2030. In some ASEAN member states, data centres are expected to account for anywhere between 2% and 30% of total national power demand.
Thirty percent. For one category of building that did not meaningfully exist in the region twenty years ago.
The Regional Picture, Country by Country
The build-out is not distributed evenly. It is concentrating in a handful of markets for reasons that have more to do with land, water, and permitting than with where the demand originates.
Malaysia: The Steepest Curve
Malaysia is absorbing the largest share and facing the sharpest increase. Power consumption from data centres is projected to rise from 8.5 TWh in 2024 to 68 TWh by 2030 — a sevenfold jump concentrated in a country whose generation mix remains heavily dependent on coal and gas.
Malaysia now holds the largest project pipeline in Southeast Asia, with approximately 3.4 GW of proposed capacity, or around 60% of all announced projects across the region. Johor in particular has become the destination of choice, largely because it offers what Singapore across the strait no longer can: land, water, and grid headroom.
Whether the generation to serve that pipeline arrives clean is the defining question for Malaysia’s emissions trajectory this decade. Building 3.4 GW of continuous, high-utilisation load onto a fossil-heavy grid does not produce a modest increase in emissions. It produces a step change.
Singapore: Constrained by Design
Singapore remains Southeast Asia’s leading hub and ranks among the largest data centre markets globally, with roughly 1 GW of operational capacity. It got there first and it got there efficiently.
But Singapore imposed a moratorium on new data centre development in 2019 precisely because the trajectory was unsustainable within a city-state’s land and power constraints. The moratorium has since been replaced with a stricter, criteria-based approval process that heavily weights energy efficiency.
The result is instructive and slightly uncomfortable: Singapore’s constraint did not reduce regional demand. It relocated it. Much of the capacity that would have been built in Singapore went to Johor instead, roughly thirty kilometres away, onto a grid with a higher carbon intensity. Serving, in many cases, the same customers.
This is carbon leakage in its clearest form. A jurisdiction tightened its rules, the demand did not disappear, and the emissions ended up higher than if the facility had been built under the stricter regime. It is a pattern climate policy keeps reproducing, and it argues strongly for coordination at the regional level rather than at the national one.
Thailand: Accelerating Fast
Thailand has moved from peripheral to serious in a very short window. In March 2025 alone, Thai authorities approved data centre investment applications worth THB 90.9 billion — roughly USD 2.7 billion.
Thailand’s advantages are geographic and political: reasonable land availability, improving connectivity, active government incentives through investment promotion schemes, and a position that serves the broader Mekong region. Its challenge mirrors Malaysia’s — a generation mix still dominated by natural gas, and renewable capacity that is expanding but not fast enough to cover the incoming load on its own.
Indonesia and Vietnam
Both represent enormous latent demand — Indonesia through sheer population, Vietnam through manufacturing and a fast-growing digital economy — and both face grid reliability constraints that have slowed hyperscale development relative to Malaysia and Thailand. Indonesia’s coal-dominated generation mix in particular means that data centre growth there carries an unusually high carbon cost per megawatt-hour.
What Is Actually Driving the Load
The convenient explanation is artificial intelligence, and AI training and inference genuinely account for a large and growing share of new capacity. But attributing the whole build-out to AI obscures a more mundane and more persistent driver.
The bulk of the load is ordinary consumer digital activity, running continuously, at scale, across a region of nearly 700 million people who came online fast and use their connections heavily.
Video streaming is the largest single component. Social platforms with algorithmic feeds that transcode and serve video at enormous volume come next. Then e-commerce, digital payments, ride-hailing, food delivery, messaging, and cloud storage — services that are now basic infrastructure across urban Southeast Asia.
Interactive entertainment is a substantial share of what remains. Asia-Pacific hosts more than 1.3 billion mobile players and accounts for over half the global gaming market by revenue. Regional entertainment platforms of every description — game publishers, streaming operators, tournament services, and interactive portals such as UFA GAME — all depend on the same underlying resource: continuously available compute sitting close enough to users to keep latency low. That proximity requirement is precisely why the servers are being built in Johor and Bangkok rather than in Oregon or Ireland. Latency-sensitive consumer services cannot be hosted an ocean away, which means the electricity demand they generate lands squarely on ASEAN grids rather than someone else’s.
Understanding this composition matters for policy. If the load were purely AI training, it would be relatively location-flexible and could be sited wherever clean power is abundant. Because a large share is latency-bound consumer services, it has to be built near the users — and that removes the easiest decarbonisation lever available.
The Grid Problem
This is where the numbers become uncomfortable.
Coal and gas still dominate generation across most of ASEAN. Adding tens of terawatt-hours of new, continuous, non-deferrable demand to those grids means, in the near term, burning more fossil fuel. Data centres run at high utilisation around the clock, which makes them a baseload-shaped load — and on most ASEAN grids today, baseload means coal.
Renewable procurement helps, but the mechanism matters enormously. There is a wide gap between:
- Additional generation — a developer signing a power purchase agreement that finances a new solar or wind farm that would not otherwise have been built. This physically displaces fossil generation.
- Unbundled certificates — buying renewable energy certificates from existing generation elsewhere on the grid. This reallocates an accounting claim without adding a single megawatt-hour of clean supply.
Both let an operator publish a “100% renewable” headline. Only one changes what comes out of a smokestack. When you read a data centre sustainability claim in this region, the first question worth asking is which of the two it describes.
Grid interconnection offers genuine promise. The ASEAN Power Grid concept — moving hydro from Laos, geothermal from Indonesia, and solar from wherever the sun is strongest to wherever the load happens to be — would meaningfully improve the picture. Progress has been slow, and it is measured in decades rather than in the six years the demand curve is giving us.
Beyond Carbon: Water and Land
Emissions dominate the discussion, but they are not the only cost.
Large data centres in tropical climates consume substantial volumes of water for cooling. In a region that already experiences seasonal water stress and is projected to see rainfall patterns become less predictable under continued warming, siting high-consumption facilities in water-constrained areas creates a direct conflict with agricultural and municipal use. Some operators have moved to closed-loop or air-cooled designs that dramatically reduce water draw, generally at the cost of higher energy consumption. There is no free option, only a trade-off to be made explicitly.
Land use and community impact receive less attention still. Industrial-scale facilities require significant footprints, often on the outskirts of growing cities, and they generate comparatively few permanent local jobs relative to the land and power they consume. Whether that trade is worthwhile is a legitimate question for the communities hosting them, and it is one that rarely gets asked before approval.
What Reasonable Policy Would Look Like
The build-out is happening. Treating it as something to prevent is not realistic — the demand is real, the economic case is strong, and refusing it in one jurisdiction simply exports it to a dirtier one, as Singapore’s experience demonstrates. The useful question is what conditions should attach.
Require additionality in renewable procurement. New load should be matched with new clean generation, not with certificates from existing plants. This is the single highest-impact rule available to regulators, and it is straightforward to write.
Mandate transparent reporting. Actual energy consumption, actual water withdrawal, actual carbon intensity of supply — published annually, on a consistent methodology, comparable across facilities. Most of this data is currently either unreported or reported in formats designed to be flattering rather than useful.
Set efficiency floors. Singapore’s criteria-based approach shows that efficiency standards can be applied to approvals without killing investment. Power usage effectiveness thresholds, waste heat recovery requirements, and cooling efficiency minimums are all well-established and readily enforceable.
Coordinate regionally. The Singapore-to-Johor migration demonstrates that unilateral national standards mostly relocate emissions. Common regional minimums would prevent the race to the least-regulated jurisdiction.
Accelerate grid decarbonisation as the priority. Everything else is marginal compared with the fuel mix. A perfectly efficient data centre on a coal grid emits more than an average one on a clean grid. The infrastructure question and the generation question are the same question, and only one of them is being addressed at the necessary speed.
The Bottom Line
Southeast Asia is building the physical infrastructure of its digital economy right now, in a compressed window, and the choices being locked in over the next three to four years will determine the region’s emissions profile well into the 2040s. Data centres have thirty-year lifespans. The grid connections serving them are longer-lived still.
Getting this right does not require choosing between digital growth and climate goals. It requires recognising that the two are the same decision, made at the same time, by the same institutions — and that pretending otherwise is how a region ends up with a sevenfold increase in electricity demand and no plan for where the clean power comes from.