Indosat, Nvidia and Nokia Are Building the Zankore AI Platform in Central Java

Four companies, Indosat Ooredoo Hutchison, Ooredoo Group, Nokia and Nvidia, plan to develop Zankore, an AI infrastructure platform for the region. The source material gives no figures: no megawatts, no GPU count, no budget. What it does have is the question from its headline, which usually becomes the central one for projects of this kind: can Central Java afford the water footprint of an AI factory?

What we know about the project

The participants' roles can be inferred from their profiles, although the source contains no official division of responsibilities. Indosat is a major Indonesian operator with customers and licenses. Ooredoo Group is its shareholder and international partner. Nokia usually handles the network side, and Nvidia the accelerators and software stack. That is our interpretation, not a statement from the companies.

What is confirmed:

• Name: Zankore
• Format: an AI infrastructure platform for the region
• Participants: Indosat Ooredoo Hutchison, Ooredoo Group, Nokia, Nvidia
• Location: Central Java (per the headline of the source article)
• Capacity, number of accelerators, investment, launch timeline: not disclosed

Until those parameters appear, it is too early to judge the project's scale. An "AI factory" can mean a campus drawing tens of megawatts or a modest cluster for local customers. The difference in water and energy consumption would be a multiple.

Why the conversation comes down to water

Nvidia's latest-generation accelerators put out a lot of heat per rack, and air cooling copes with it less and less well. That is why new campuses increasingly install liquid cooling loops. How much water they use depends on the design. Evaporative cooling towers consume water continuously, and at large sites that can mean millions of liters per day. Closed loops and dry coolers use noticeably less water but more electricity.

In the tropics the trade-off is harsher. High temperature and humidity reduce the efficiency of free cooling, so giving up evaporation almost always means a higher electricity bill. And coal still makes up a notable share of Java's power grid, so "we saved water" can easily turn into "we increased emissions."

Central Java also has its own history. Coastal areas such as Semarang have long suffered from land subsidence, one cause of which is groundwater extraction. Any large water consumer in such a place immediately lands in a sensitive zone, and drawing from wells here will raise questions with regulators and residents before it does with investors.

What to ask the project

While Zankore exists only as an announcement, the most useful thing to do is demand specifics. For local authorities, customers and investors, the questions are:

• What cooling design is planned, and what is the target WUE (liters of water per kWh of compute)?
• Where will the water come from: the municipal network, surface sources, wells or treated wastewater?
• What is the site's energy balance, and are there agreements for renewable generation?
• How much capacity will launch in the first phase, and how will it grow?
• Will actual water consumption data be published, and not just design figures?

For Indonesian companies that need local compute for model training and inference, Zankore could become an alternative to foreign clouds, primarily on latency and data-storage requirements. But migration should be planned only after the technical parameters and access terms are published. If the participants disclose the water budget in advance, that will be a strong sign of the project's maturity. If they do not, the question of the price of water will go unanswered until the first complaints about shortages.