Four Cities Get AI Data Centres. Here Comes the Water Bill

Four sites are already on the list. What nobody is discussing is the bill that comes with them.

Talk about AI data centres sounds like a technology story. It is really an energy and water story. Four Indonesian regions have announced projects totalling about 2.6 gigawatts of capacity. That number is not the problem. How much water arrives with it is.

• Batam, Riau Islands — a 360 MW AI Factory with 170,000 GPUs, built with Nvidia. EGS is developing 30 hectares in Nongsa, going from 140 MW to 580 MW by 2030
• Jatiluhur, Purwakarta, West Java — a 640 MW campus with direct-to-chip liquid cooling up to 500 kW per rack. It sits about five kilometres from the Jatiluhur Dam
• Batang, Central Java — Zankore, involving Indosat, Ooredoo, Nokia and Nvidia, targeting 1 gigawatt in three years
• Surabaya, East Java — DCI Indonesia opened E2 with 9 MW of capacity. The smallest of the four, and the one most often lumped in with the rest

The upside is real

2.6 gigawatts is a serious number. For Southeast Asia that is a strong position, and not many people expect us to have it.

The advantages are not just slogans:

• AI data centres need round-the-clock power. That suits a country with battery capacity and submarine cable links to Singapore and the United States
• The work is skilled, not factory work. It needs engineers, security analysts, cloud architects and AI specialists
• It pulls an ecosystem with it. Vendors, contractors and software companies grow around the site
• Free trade status. Batam is already duty-free and carries more than 80 percent of Indonesia's internet traffic

And this is not just talk. Government is already moving: Komdigi and IDPRO are drafting a sustainable data centre standard, and the industry is targeted to grow tenfold in five years.

The downside: water

This is the part that gets skipped, and the numbers can be worked out. EPRI published water benchmarks for a 150 MW data centre in May 2025. The useful figure is 2.8 litres of water per kilowatt-hour for evaporative cooling.

For 580 MW in Batam that works out to:

• 14.2 million cubic metres of water a year
• the equivalent of 51,284 households using 760 litres a day

For 1 gigawatt in Batang on the same cooling, it becomes 24.5 million cubic metres a year. And that is still a plan on paper.

Compare Surabaya, at 9 MW today: 0.22 million cubic metres a year, or 796 households. Separating what is already running from what is only announced matters, because people easily treat them as the same.

The downside: where the power comes from

The second question is harder than water. At the end of 2024, 86.36 percent of Indonesia's installed capacity was still coal. The 23 percent renewable target has been pushed back to 2030.

A data centre needs steady power around the clock. Solar and wind do not provide that. The 2025-2034 supply plan does add 42.6 GW of renewable capacity, but that suits loads that can be predicted rather than ones that cannot.

The arithmetic is worth sitting with. For 1 GW in Batang, if cooling takes 40 percent of the electricity, that is roughly 400 MW spent purely on keeping chips cool. It is more than 0.4 percent of national installed capacity, and it arrives through a grid that is mostly coal.

What usually gets left out

Two details are rarely mentioned, and they matter more than the headline.

First, the MW figure has no stated basis. Is it server load or total facility load? The answer changes everything, and almost nobody says which. The difference is the PUE, roughly 1.1 to 1.4 for liquid-cooled facilities.

Second, the water assumptions differ. EPRI reports site WUE ranging from 0.1 to 9.0 litres per kilowatt-hour. Numbers in articles often take the lowest figure without naming the assumption behind it.

And the vendor claim about "near zero water" cooling only holds where the outside air is reliably cool. Batang and Surabaya are hot, humid tropics. Outside air is never reliably cool there, so the real figure has to be higher than the brochure suggests.

Questions worth asking

If these are going ahead, five things should be requirements rather than details:

• What is the water-per-kWh benchmark? Without it, a water-saving claim means nothing
• Is the MW figure server load or facility load?
• Where does the water come from, and is the facility drawing from the same source as nearby residents? Sharpest question for Jatiluhur, five kilometres from the dam
• Is the design genuinely liquid-cooled, or just an evaporative tower? The difference is eightfold
• How much water is used during construction? Nobody publishes it, and it can exceed the first full year of operations

Checking it yourself

The numbers above should be possible to verify, and that is the point. Open AI Data Center Water & Power on Loonix, enter the announced MW figure, and see the cubic metres per year for yourself.

The tool deliberately shows an uncertainty range, because figures that come out of it and figures from other sources will not match exactly. What it does not count is also worth asking about: construction water, off-site water from electricity burned at the point of generation, and whether the design actually matches what was promised.

So the answer to the original question is simple. Yes, these four locations can be built, and the upside is real. But the cost does not show up in the investment column. It shows up on the PLN electricity bill, and in the water that nearby residents drink.