Data Center Site Selection in Indonesia: The Five Constraints That Decide Which Sites Get Built

Indonesia's data center pipeline looks enormous on paper. Installed data center capacity reached roughly 1,717 MW in 2026 and is forecast to more than double to around 4,145 MW by 2031. Planned capacity for upcoming facilities is estimated at almost twice what is already built, and Jakarta alone carries more than 1 GW of announced projects.

Announced capacity and energized capacity are two different businesses.

Grid interconnection in Indonesia runs three to five years, and the binding constraint sits in transmission rather than generation. One analysis has the country heading toward a 1 GW shortfall by 2030, which is a delivery problem rather than a supply problem.

That gap is where the risk lives. It is also the point where data center site selection stops being a real estate exercise and becomes an infrastructure one. Here are the five constraints that decide whether a new site gets built, roughly in the order they kill projects.

1. Power, not land, is the binding constraint on an Indonesian data center site

In most property markets, land and capital are the gating constraints and power is assumed. Indonesia inverts that. A site without a confirmed path to firm power is not a data center site. It is a field with a good story attached.

A credible power position is specific and testable. A named substation with documented available capacity. A defined feed path, ideally dual feeds off a 150 kV connection, which is what the serious campuses in the Cikarang corridor have specified. An interconnection schedule with a transmission plan behind it, not a letter of intent. And increasingly a renewable component, because hyperscaler procurement policy now demands one.

PLN also appears to be consolidating data center power allocation around anchor partners rather than spreading it thinly. So the diligence question is no longer "how many megawatts does this site claim." It is "what has been allocated, by whom, under what document, deliverable in what year."

The 2025–2034 RUPTL helps at the margin, with a much larger renewable target and a serious transmission build-out. It does not shorten the PLN interconnection queue for anyone starting today.

2. Rack density: why most existing Indonesian capacity cannot host AI workloads

Indonesia's headline data center capacity numbers combine two markets that no longer compete with each other.

A hall designed in 2018 for 8 to 10 kW racks cannot host modern AI workloads. The gap is not a cooling upgrade. It is floor loading, slab design, ceiling height, power train, and the plumbing for direct-to-chip or immersion systems.

Look at what is being built now. BDx specified its Jatiluhur AI campus for up to 120 kW per rack with liquid cooling from day one. Digital Edge's Cikarang campus targets an annualized PUE of 1.25 on a design scaled to 500 MW. EDGNEX's Cikarang project, a USD 2.3 billion AI-dedicated build of more than 144 MW, targets 1.32.

None of that is reachable by retrofit. Tier III designs still account for the large majority of live capacity in Indonesia, most of it built for enterprise and BFSI colocation at conventional densities. When a hyperscale or AI tenant goes shopping, the qualifying pool is far smaller than the national totals suggest.

This is unusual in property. Greenfield data center development is structurally advantaged here, because the design assumptions changed faster than the asset life.

3. Water is the second constraint, and it is arriving faster than expected

The chairman of IDPRO, Indonesia's data center provider association, has said it plainly: after energy, water is the next strategic issue in developing data centers here.

Tropical siting makes this harder rather than easier. High ambient temperature and humidity mean mechanical cooling runs at high load year-round, and tropical data center cooling is not simply a warmer version of temperate cooling. Conventional evaporative systems consume somewhere between 2 and 4 liters per kWh, against a regional WUE target of 1.8 to 2.0 liters per kWh by 2030. PDAM supply in many industrial areas is not stable enough to underwrite a 100 MW load. Groundwater extraction in the Jakarta basin carries permitting risk and subsidence risk together.

The practical consequence is that water source, treatment, and discharge permits belong in site selection, not in detailed design. Cooling decisions made before a facility opens set its water profile for decades. Hyperscale tenants now ask for a WUE number alongside a PUE number, and operators are openly favouring locations outside Jakarta on the strength of water supply and utility infrastructure.

4. The 36-month clock: title, zoning, permitting and PLN interconnection

The distance between signing for land and serving first traffic is where Indonesian data center projects quietly die.

The sequence is long and mostly serial. Land title clean-up and understanding exactly what HGB position you are inheriting. RTRW and zoning alignment for the intended use. PBG and environmental permitting. The PLN interconnection queue. Fiber build to at least three diverse routes. Each step has its own queue, and several cannot start until the one before finishes.

"Shovel-ready" is a technical claim, not a marketing adjective. Six questions worth asking any vendor before anything else:

1. What is the current land title status, and what is the path to a title a lender will accept?
2. Is the intended use already permitted under the applicable spatial plan, or does it need an amendment?
3. Which substation, at what voltage, with how much documented available capacity?
4. What is the written PLN position, and the earliest realistic energization date?
5. What is the water source, and what permits does it require at 50 MW and at 200 MW?
6. How many physically diverse fiber routes reach the boundary today, not in plan?

If a seller cannot answer all six in one meeting, the site is at an earlier stage than it is being presented as.

5. Why the hyperscale corridor moved east to Bekasi and Cikarang

The latency argument for Jakarta proper is real, but it has a radius, and the radius is wider than the market assumed five years ago.

Workload type sets the tolerance. Payment processing, trading and real-time financial services need metro proximity and very low latency paths. AI training, batch inference, archival storage and most enterprise workloads do not. Once you separate those two groups, the arithmetic changes. Jakarta still holds around 57 percent of national capacity on the strength of interconnect density and submarine cable landings. But land cost per megawatt, availability of large contiguous parcels, industrial power infrastructure and flood exposure all point east.

That is how Bekasi and Cikarang became Indonesia's hyperscale data center corridor. Industrially zoned land, access to substations, room for multi-building campuses, and an established industrial estate utility base. Keppel, Princeton Digital, Digital Edge, EDGNEX, NTT and the government's own PDN facility are all there or nearby. Construction cost of roughly USD 8 to 9 million per MW keeps Indonesia competitive against Singapore and Malaysia, although that number is rising.

To be fair about the trade-offs: the eastern corridor gives up metro interconnect density, it depends on fiber routes that are newer than Jakarta's, and Batam has genuine advantages for Singapore-adjacent workloads through its SEZ status and latency position. Corridor selection should follow the workload, not the brochure.

What this means for data center land owners and investors

For site owners, the honest version is that land alone is no longer a marketable asset in this sector. What is marketable is a documented power position, a water plan, clean title, confirmed zoning and fiber. Assembling that package takes two to three years, and it is the difference between a land sale and an infrastructure transaction.

For buyers and investors, underwriting should reweight toward secured power allocation as the primary variable, and toward deliverable capacity rather than announced capacity. The market still prices press releases. It does not yet discount properly for the delivery gap.

Nata Advisory works on land and infrastructure transactions across Indonesia's property and energy sectors, including current mandates in the Cikarang corridor. If you are evaluating a site or a corridor, I am happy to compare notes.

Nucky Poedjiardjo is Managing Partner of Nata Advisory.

Sources: Mordor Intelligence, Research and Markets, Global Data Center Hub, IDPRO via The Jakarta Post, Sustainable Digital Infrastructure Accord, Uptime Institute, operator disclosures from BDx, Digital Edge, EDGNEX and Keppel, and the 2025–2034 RUPTL.