SINGAPORE, 30 AUG 2026 — A market forecast published this week puts Southeast Asian data centre investment at US$15.72bn in 2025 rising to US$35.08bn by 2031, with Malaysia leading an upcoming pipeline of more than 6GW. A pipeline counts what has been announced, which is not the same as what a utility has agreed to connect.
What the forecast says
The report, from research firm Arizton, describes 306 operational data centres across nine countries in the region with a further 173 in the pipeline. Malaysia leads the upcoming IT load pipeline at over 6GW, with Thailand emerging at around 3.5GW.
The investment figure grows at a compound rate of 14.32 per cent to reach US$35.08bn in 2031.
The numbers are useful for direction, and this is vendor research sold as a product. With that category of source it pays to read the definitions before the conclusions.
What a pipeline is
An IT load pipeline is the sum of the capacity that announced projects say they intend to build. It includes projects at every stage: signed and under construction, land acquired and awaiting approval, memorandum of understanding only, and press release only.
It does not require a grid connection agreement, a completed environmental approval, secured water, or financing. Each of those is a separate gate, and each has caused regional projects to slip or disappear.
So 6GW in Malaysia is best read as a measure of intent and of investor interest. It is not a forecast of what will be energised, nor does the report claim it is. The distortion happens downstream, when the figure is quoted in a news story or a board paper as capacity that is coming.
The constraint is on the other side of the meter
Malaysia's own published analysis is the reason to be careful with the 6GW number.
We have reported the forecast that data centres could take 31 per cent of national electricity by 2035. A pipeline that would require a further several gigawatts of firm supply is bounded not by developer appetite but by generation, transmission and the political question of who is served first when both a data centre and a residential estate want the same connection.
Neighbours face the same arithmetic with different numbers. The proposed Luzon hub in the Philippines would need 16 per cent of the island's grid. Singapore has been importing rather than building, conditionally approving 900MW of Johor solar. Thailand has used a grid-connection letter as the gate on investment incentives, which is a regulator saying explicitly that the constraint is electrical.
Read together, those are four governments treating power as the binding limit while a pipeline figure counts intent. Both descriptions are accurate, and only the second one constrains what actually gets built.
Nine countries is doing quiet work too
The other figure worth unpacking is the count of 306 operational facilities across nine countries, because a regional aggregate hides a distribution that is not close to even.
Southeast Asian data centre capacity is heavily concentrated in Singapore and, increasingly, Johor immediately across the strait, with Jakarta and Bangkok as the next tier. A single regional total treats a hyperscale campus in Johor and a small colocation facility in a secondary city as one unit each, when their capacity differs by orders of magnitude.
Facility counts and gigawatt figures therefore describe different things, and it is the gigawatt figures that bear on the grid. Counting buildings tells you how many operators are present, whereas counting load tells you what the region has to find electricity for.
It also means a country's position in a facility count says little about its position in capacity. A market with many small facilities and one with few large ones can appear similar in one metric and nothing alike in the other.
Why the number still matters
None of this makes the forecast worthless. Dismissing a pipeline figure is as unhelpful as quoting it uncritically.
A pipeline is a real measure of where capital wants to go, and capital wanting to go somewhere is the precondition for anything happening. Malaysia leading it over Singapore, Indonesia and Vietnam is a genuine signal about land, cost and policy posture, and it is consistent with what has actually been built in Johor.
The 306 operational facilities are the harder number and the more interesting one. That is a count of things that exist, and set against 173 in the pipeline it describes a region that would roughly double its facility count if everything announced were completed. Whether it does is the question the pipeline figure cannot answer.
Malaysia's utility publishes the harder number
The useful thing about Malaysia specifically is that the gap between pipeline and reality is not a matter of inference. Tenaga Nasional reports it.
As at the first quarter of 2026, TNB described a data centre pipeline of 8.3GW. It had signed electricity supply agreements with 49 projects totalling about 7,100MW. Twenty-three projects were under construction, with a maximum demand of 3.8GW. And actual data centre load being drawn stood at 1.05GW.
Those four figures are the same industry measured at four different stages of commitment, and they descend by roughly a factor of eight from end to end. A pipeline number and a connected-load number are both true, and quoting the first as though it were the second overstates the position by that factor.
The direction is not in doubt: that 1.05GW was up 117 per cent year on year, which is a very fast ramp by the standards of electricity demand. TNB has also compressed grid connection timelines for data centres from 36 months to as little as 12 through a dedicated pathway, and is spending heavily on the grid to keep up.
Malaysia is therefore building quickly, from a small base, against a pipeline several times larger than what has been energised. That sentence carries more information than the 6GW figure does on its own.
How to read a regional forecast
Three questions separate a usable number from a quotable one, and none of them requires access to the report.
What is being counted — announced projects, projects under construction, or energised capacity. Over what period, since a compound growth rate to 2031 is a modelled projection rather than an observation. And who commissioned it, because infrastructure market research is frequently sold to the participants whose market it describes.
Applied here: this is announced capacity, projected to 2031, from a commercial research provider. Momentum is what that combination can tell you. It is not a basis for planning.
One check is available to any reader here, and it beats all of the above. Regional grid operators and energy regulators publish connection queues and approved load. Where those exist, they are a harder measure of what is coming than any pipeline figure, because a project in a connection queue has had to ask a utility for power rather than issue a statement.
