Size a data hall to the island's water, not to the tenant's request

Batam, August 2024. The design review for the second phase of our data-centre campus took place in a meeting room that overlooks a reservoir. That was a coincidence, but it shaped the afternoon. Our tenant, a global cloud provider already running 24 megawatts of IT load in our first building, wanted 60 megawatts more. Its reference design cooled the halls by evaporation, an approach that saves a great deal of electricity in the right climate. At full load it would have drawn about 1,400 cubic metres of water a day.

Since Singapore paused approvals for new data centres between 2019 and 2022 and then reopened under stricter efficiency conditions, capacity has spread to Johor and to the Riau Islands across the strait. Batam, an hour away by ferry, has land, a free-trade zone and subsea cable routes to Singapore. It also has a limited supply of fresh water. My argument in this note is that campuses in places like Batam should be sized by the local water and grid budget first and the tenant's request second, and that the industry's favourite efficiency figure makes that harder to see.

A campus shares its reservoir with its neighbours

Batam draws its water from a small number of reservoirs fed by rain on a small island, and dry spells have tightened supply before. Our park's allocation from the island's water operator covers 47 manufacturers that between them use about 6,900 cubic metres a day. The evaporative design would have taken the park to within a few per cent of its allocation, leaving almost no margin for those manufacturers to expand or for a dry year.

The pattern of demand matters as much as the volume. A factory's water use rises and falls with its shifts and its order book. A data centre's cooling load runs every hour of every day, and in an evaporative design it rises in the hottest, driest weeks, which are exactly when a reservoir is lowest. A campus that takes the margin in a wet year takes it from people who will need it in a dry one. Some of those people are our tenants. Others are households who never signed anything with us.

The efficiency figure that hides a trade

Data-centre efficiency is usually reported as power usage effectiveness: the total energy a facility uses divided by the energy that reaches the IT equipment. A figure close to one means little energy is spent on cooling and power conversion. Tenants have public commitments built around it, and it is the number their engineers defend most fiercely.

Evaporative cooling lowers power usage effectiveness because it uses water to reject heat that would otherwise need electricity. For our second phase, the evaporative design came out at an expected 1.27 over a year. The design we chose, air-cooled chillers on a closed loop topped up with recycled water from our own wastewater plant, came out at 1.38. On 36 megawatts of IT load, that difference means about four megawatts of additional average demand, or roughly 35 gigawatt-hours a year from a grid that still burns mostly fossil fuel.

I do not pretend that cost away. It is a real environmental and financial price, and our tenant's engineers were right to raise it. What I object to is a measure that shows only one side of the exchange. The evaporative design would have consumed close to one litre of water for every kilowatt-hour reaching the servers. Ours is designed to use about a seventh of that, most of it recycled. An owner that reports power usage effectiveness alone will always appear to have chosen badly when it saves water, and will be rewarded for spending it.

The industry already has a companion measure for water, expressed in litres per kilowatt-hour of IT energy. Our board now sees both figures for the campus every quarter, side by side, with the park's water allocation and its remaining margin. I would like every data-centre operator in the region to publish both, and every board to ask for both.

Water is also harder to add back than electricity. If Batam's grid grows tight, new generation and storage can be built within a few years, at a price. A reservoir cannot be enlarged so quickly on a small island with limited catchments, and desalination carries its own energy cost and its own waste brine. That asymmetry is a further reason, when the two compete, to spend electricity before water.

What I ask before we offer megawatts

The tenant signed for 36 megawatts in March 2025 and placed the other 24 in Johor. We gave up roughly US$11 million a year of rent we might have earned, and we spent about US$14 million more on chillers than the evaporative design would have needed, most of which the rent will recover over the lease. The first 18 megawatts were energised in July 2026.

The more lasting change is in how we sell capacity. Before we market a plot on the campus, we now set out four budgets for it: water, grid capacity with its redundancy, wastewater and the cooling design those three allow. A tenant can see from the first conversation what the plot can bear, and can design to it rather than discover our limits in a design review.

I also ask four questions before we offer megawatts to anyone. How much water will the design consume in the driest month we have on record, not the average one? What margin does that leave for the manufacturers already in the park, through the life of their leases? Has the island's water operator seen the projection, not merely the request? And if the answer to any of those is uncomfortable, what smaller offer could we make honestly?

None of this argues against data centres on Batam. They bring investment, skilled work and a reason for subsea cables to land on the island, and our campus is one of the better businesses we own. It argues for treating a small island's water as the scarce input it is, and for being as precise about litres as we are about megawatts. The tenant's request is one number in the plan. The reservoir is the limit on all of them.