Johor has become the fastest-growing data centre market in Asia. The construction is visible from the highway — shell after shell going up in Nusajaya, Sedenak, and along the corridor toward Singapore, each one rated in the tens of megawatts.
The electricity system underneath it was not planned for this. That gap is now the most consequential constraint on where Asian compute gets built.
The scale of the build
Johor’s appeal is straightforward. Land is available and cheaper than Singapore. The crossing to Singapore is short enough that latency-sensitive workloads can serve the financial hub from Malaysian soil. Power, until recently, was assumed to be abundant.
- Announced and under-construction capacity in Johor has grown from under 100MW in 2022 to well over 1,000MW in committed projects
- A single hyperscale campus can draw 100MW or more continuously — comparable to a mid-sized city
- AI training and inference workloads have power density per rack five to ten times conventional cloud
- Construction timelines run 18 to 30 months; grid reinforcement timelines run longer
That last bullet is the whole problem. A data centre can be built faster than the transmission capacity to feed it can be permitted, engineered, and constructed.
Compute follows power. Power follows policy. Policy, right now, is sprinting to catch both.
What the grid is actually doing
Malaysia’s grid operator has been managing this as a queue problem. New large-load connections are assessed against available capacity, and where capacity does not exist, the applicant either waits or funds reinforcement.
Three responses are visible across the market:
On-site generation. Developers are installing gas generation behind the meter, sometimes as primary supply rather than backup. This sidesteps the grid queue entirely and shifts the constraint from transmission to fuel supply and emissions permitting.
Load scheduling. Designing facilities to flex consumption — shifting batch inference and non-urgent training to off-peak windows. This is genuinely difficult for workloads sold as always-available, and easier for those with slack.
Location shift. Moving projects to jurisdictions where grid headroom exists. This is already happening, and it is the mechanism by which power availability is redrawing Asia’s compute map.
The regional competition
Johor is not alone. The same dynamic is playing out across the region, with different starting positions:
Singapore effectively closed new data centre approvals for three years and has reopened them under strict efficiency and sustainability criteria. The constraint there was explicit and policy-driven.
Indonesia has grid headroom but a thinner transmission network outside Java, and permitting timelines that developers find unpredictable.
Vietnam has capacity and competitive power pricing but a grid that has already shown strain during demand peaks.
Japan has both grid capacity and a mature industrial base, at a cost premium that only certain workloads can absorb.
Why policy is the deciding variable
None of these outcomes is determined by engineering. Every one of them turns on decisions that governments are making right now, often without a clear framework:
- Whether large-load connections get priority over other industrial demand
- Whether behind-the-meter generation is permitted, and on what emissions terms
- Whether renewable procurement obligations apply to data centres, and how they are verified
- Whether transmission investment is funded by the operator, the developer, or the state
The jurisdictions that answer these questions quickly and coherently are the ones that will host the next wave of capacity. The ones that leave them ambiguous will watch projects go elsewhere — not because of cost, but because of timeline.
The reckoning
The build will not stop. Demand for compute in Asia is growing faster than any supply constraint has yet slowed it. What the power situation determines is where that demand lands.
Johor’s boom is the clearest test case in the region: a market with land, location, and demand, discovering that the binding constraint is the one nobody priced in at the start. The resolution — whether through grid investment, on-site generation, or a shift in project location — will set the pattern for every other market that follows.
The next trillion dollars of Asian compute will be built where the power already is, or where policy makes it arrive fastest. Increasingly, those are the same place.
