Zankore AI factory plans for Batang, on the north coast of Central Java, would create one of the largest AI computing sites in Southeast Asia. The project is backed by Indosat Ooredoo Hutchison, Ooredoo Group, Nvidia and Nokia. It aims for 1 gigawatt (GW) of computing capacity, starting with about 200 megawatts (MW) in 2027. Indonesia’s government says it will make the country a regional AI hub.
The question two Indonesian academics asked this week is whether Central Java can supply the water. Writing in The Conversation on 9 October 2026, Imam Salehudin of Universitas Indonesia and Usman Arief of Universitas Gadjah Mada estimate the first phase could evaporate 4.8 million to 9.1 million litres a day. That is in a province that has just come out of a four-month drought alert, on a coastline that is sinking.
This article sets out what is known about the Zankore AI factory, checks the water arithmetic against other published benchmarks, and explains why the choice of cooling technology decides almost everything. We also look at who owns the project, who would bear the water cost, and what Zankore has not yet disclosed. For background on how data centre water is measured, see our explainer on water usage effectiveness in AI data centres.
Table of contents
- What Is the Zankore AI Factory?
- The Case for the Zankore AI Factory
- The Zankore AI Factory Water Estimate
- Why Cooling Choice Decides the Zankore AI Factory’s Water Bill
- Central Java’s Water Crisis and the Zankore AI Factory
- Would the Zankore AI Factory Strain Batang’s Water Supply?
- Who Pays and Who Benefits From the Zankore AI Factory?
- What the Zankore AI Factory Has Not Disclosed
- What the Zankore AI Factory Means for Businesses Buying AI
- Zankore AI Factory: Frequently Asked Questions
- References
What Is the Zankore AI Factory?
The Zankore AI factory was launched in Jakarta on 6 August 2026 as “Zankore by Indosat”. Nokia’s launch release describes a platform “targeting 1 gigawatt of NVIDIA DSX AI Factory capacity”, built for hyperscalers, enterprises, governments and AI developers across the Asia-Pacific region.
| Item | Detail |
|---|---|
| Launched | 6 August 2026, Jakarta |
| Partners | Indosat Ooredoo Hutchison, Ooredoo Group, Nvidia, Nokia |
| Location | Batang, Central Java (Industropolis Batang Special Economic Zone) |
| Target capacity | 1 GW; about 200 MW in the first half of 2027 |
| Hardware | Nvidia GB300 NVL72 systems, Nvidia DSX reference design |
| Financing | Senior term loan of up to US$3.1 billion (September 2026) |
| Cooling system | Not disclosed |
Southeast Asia’s largest AI computing project
Indonesia’s Communications and Digital Affairs Minister, Meutya Hafid, told Antara that the Zankore AI factory should reach 1 GW within three years. The ministry calls it Southeast Asia’s largest AI computing infrastructure project. “The presence of the AI Factory will strengthen Indonesia’s position as a regional AI hub because it will be able to serve both domestic and regional demand,” she said.
She said construction would begin in Batang in the first half of 2027. Nokia’s release says about 200 MW of capacity will be delivered in that same period. The two statements do not quite match, and neither company has published a site plan.
More than a data centre
An “AI factory” is Nvidia’s term for a data centre built around training and running AI models. Zankore also plans to host Sahabat-AI, an open-source large language model with 70 billion parameters that supports Bahasa Indonesia, Javanese, Sundanese, Balinese and Batak. That is a genuine natural language processing gap for a country of several hundred languages, and it is part of the political case for the project.
The Case for the Zankore AI Factory
The water question does not mean the project lacks merit. Indonesia’s government and the backers make a serious economic and strategic case, and it deserves to be stated fairly before weighing the costs.
A sovereign AI ambition
Minister Meutya Hafid said the investment “reflects growing global confidence in Indonesia’s digital economy”. She told Antara that the government had backed the Zankore AI factory since it was first proposed two years ago and had presented it to President Prabowo Subianto. She also pointed to Indonesia’s first GPU manufacturing facility, under construction in Cikarang, West Java, and due to open in December 2026.
Cheaper AI for Indonesian users
Zankore plans a “Token-as-a-Service” offer, which Antara says will let businesses, government agencies, developers and the public use AI computing “at lower cost”. Combined with Sahabat-AI’s support for regional languages, the Zankore AI factory could give Indonesian organisations AI tools built for their own languages, hosted in their own country.
Backed by international lenders
In September, DCD reported that Zankore had secured a senior term loan of up to US$3.1 billion for an initial 100 MW build. Citi, ING, Natixis CIB, Qatar National Bank and United Overseas Bank arranged it. “AI is creating a defining opportunity for Indonesia and Southeast Asia,” said Vikram Sinha, Zankore’s chairman.
The Zankore AI Factory Water Estimate
The Conversation estimate for the Zankore AI factory starts from a published benchmark. A 2025 study by University of California researchers, building on Microsoft’s projected data-centre designs, found Indonesia had the highest cooling water use among the countries it examined: about 1.9 litres for every kilowatt-hour (kWh) used by the servers.
The authors explain why in the Indonesian version of their article. Indonesia’s humid climate makes water evaporate less readily, so cooling towers must evaporate more of it to remove the same amount of heat.
How the 9.1 million litre figure is reached
The arithmetic is straightforward. A 200 MW load running all day uses 200,000 kW × 24 hours = 4.8 million kWh. At 1.9 litres per kWh, that is 9.12 million litres a day. The authors’ lower figure of 4.8 million litres allows for part of the 200 MW going to equipment other than servers.
“Based on those projections, Zankore’s 200-MW first phase could evaporate between 4.8 million and 9.1 million liters of water per day, assuming continuous operation,” the authors write. At 1 GW, every figure is five times larger.
Checking it against other benchmarks
We compared the Zankore AI factory estimate with two other published numbers. Nvidia says conventional cooling-tower systems use “roughly 2.6 million gallons per megawatt per year”. At 3.785 litres per gallon, that is about 9.84 million litres per MW a year, or 5.39 million litres a day for 200 MW.
Amazon Web Services publishes water usage effectiveness for its Indonesian data centres. In its Indonesia sustainability fact sheet, the AWS Asia Pacific (Jakarta) region averaged 2.85 litres per kWh in 2025, up from 2.75 in 2024. Applied to 200 MW, that rate would mean 13.68 million litres a day.
Bar widths are each estimate divided by 13.68. Every figure assumes 200 MW running around the clock, and every one assumes water-based cooling. None of them is a Zankore disclosure.
Putting the volume in context
AWS says it withdrew 405,720,604 litres of water in Indonesia in 2025. At the upper Conversation estimate, the first phase of the Zankore AI factory would use 9.12 million × 365 = 3.33 billion litres a year, about eight times that. Withdrawal and evaporation are not identical measures, so the comparison is rough, but the order of magnitude is clear.
Why Cooling Choice Decides the Zankore AI Factory's Water Bill
All four estimates above share one assumption: that the Zankore AI factory cools its servers by evaporating water. That is not settled. The Conversation notes that Zankore “has not disclosed which cooling system it will use”.
Nvidia’s case for near-zero water
Zankore will be built on Nvidia’s DSX reference design, and Nvidia makes a strong claim for it. In a June 2026 blog post, Nvidia’s director of data centre cooling, Ali Heydari, said: “The NVIDIA DSX reference design for AI factories has zero water consumption.” Chips are cooled by liquid at up to 45°C in a closed loop, so outdoor dry coolers, essentially large radiators, can reject the heat without evaporating water.
The catch: a favourable climate
The same post adds a condition. Chiller-free operation with dry coolers is possible “in favorable climates”, described as “somewhere with reliably cool outdoor air”. Heydari also allowed for “maybe 1% of the year when we might need chillers in some climates”. Batang sits on Java’s hot, humid tropical coast.
Whether Batang counts as a favourable climate for the Zankore AI factory is exactly the question the company has not answered. If it does, the water estimates fall sharply, with some cost in electricity for fans and occasional chillers. If it does not, the 1.9 litres per kWh assumption looks reasonable, or even optimistic next to AWS’s Jakarta figure.
Seawater cooling and its trade-offs
Batang is on the coast, so seawater cooling is another option. The Conversation says it “could reduce freshwater consumption” but warns of “other environmental risks, including impacts on marine ecosystems”, such as warm water discharged back into the sea.
| Cooling option | Freshwater use | Main trade-off |
|---|---|---|
| Evaporative cooling towers | High: the basis of the 4.8m to 13.7m litre estimates | Competes with farms and households in a drought |
| Closed-loop liquid with dry coolers (Nvidia DSX) | Near zero, says Nvidia, in favourable climates | More electricity for fans; may need chillers in hot weather |
| Seawater cooling | Low freshwater use | Warm discharge and possible harm to marine life |
Central Java's Water Crisis and the Zankore AI Factory
The timing of the Zankore AI factory is awkward. The Zankore AI factory was announced in the middle of one of Central Java’s worst dry seasons in recent years.
A four-month drought alert
Central Java declared a drought alert from 5 June to 28 September 2026, affecting 2.34 million people in 784 villages, including parts of Batang. As of 30 September, Metro TV News reported that 29 local areas had declared drought emergency alerts. The 784 villages lay in 255 districts across 34 regencies and cities. The provincial government prepared 75.46 million litres of clean water for residents.
Days of drought relief per facility
That relief figure gives the most striking comparison in The Conversation piece. At 4.8 million to 9.12 million litres a day, the first phase of the Zankore AI factory would use the province’s entire drought relief volume in 75.46 ÷ 9.12 = 8.3 days to 75.46 ÷ 4.8 = 15.7 days. At 1 GW, using 24 million to 45.6 million litres a day, it would take 1.7 to 3.1 days.
Bar widths are each figure divided by 15.7 days. The 1 GW figures are the 200 MW estimates multiplied by five.
A coastline that is sinking
The second pressure is land subsidence. The authors cite Indonesia’s Geological Agency and the Bandung Institute of Technology, which have recorded the ground sinking by around 6 to 20 centimetres a year in the Pekalongan-Batang area. Groundwater extraction is widely cited as a main cause of subsidence along Java’s north coast. As land sinks, it stores less water and seawater pushes further inland, worsening tidal flooding.
“These changes are difficult to reverse and could eventually force communities to relocate,” the authors write. For an industrial user in this location, where the water comes from matters as much as how much is used.
Would the Zankore AI Factory Strain Batang's Water Supply?
Indonesia’s government has already built water infrastructure for the Batang zone. The question is how much of it one tenant could take.
The raw-water system built for the zone
According to Kompas, the Ministry of Public Works spent Rp409.7 billion on raw-water infrastructure for the Batang zone. It can supply up to 285 litres a second for industry and for communities around the zone. That works out at 285 × 86,400 seconds = 24.6 million litres a day.
Share of capacity
At the lower estimate, the first phase of the Zankore AI factory would use 4.8 ÷ 24.6 = 19.5% of that capacity. At the upper estimate it would use 37.0%, matching The Conversation’s “more than 30%”. At 1 GW, demand would be 97.5% to 185% of the system’s entire output, before any other factory in the zone or any household is served.
| Scenario | Water per day | Share of Batang raw-water capacity (24.6m litres a day) |
|---|---|---|
| 200 MW, lower estimate | 4.8m litres | 19.5% |
| 200 MW, upper estimate | 9.12m litres | 37.0% |
| 1 GW, lower estimate | 24m litres | 97.5% |
| 1 GW, upper estimate | 45.6m litres | 185% |
These shares assume the Zankore AI factory draws from this system and uses water-based cooling. Neither has been confirmed. They show why disclosure matters, not what will happen.
Who Pays and Who Benefits From the Zankore AI Factory?
The authors’ economic argument is about who bears the cost. They describe a risk of “negative externality: profits accrue to shareholders, while part of the environmental cost, particularly water consumption, could fall on the local community”.
A foreign-majority ownership structure
Indonesian business site Investor.id reported the shareholding after an analyst meeting with Indosat in September.
| Shareholder | Stake | Home country |
|---|---|---|
| Ooredoo Group | 49% | Qatar |
| Indosat Ooredoo Hutchison | 37% | Indonesia |
| Nvidia | 11% | United States |
| Nokia | 3% | Finland |
Adding up the stakes held outside Indonesia gives 49 + 11 + 3 = 63%. Investor.id also reported that Zankore’s customers sign five-year “take or pay” contracts, and that its targets include AI developers and hyperscalers from overseas.
What Indosat expects to earn
Investor.id’s report also shows where the money is expected to go. A Stockbit analyst, Christian William Munaba, estimated that the first 200 MW of the Zankore AI factory could add up to Rp2.9 trillion to Indosat’s profit. That was up from earlier projections of Rp1.6 trillion to Rp2.4 trillion. Indosat also gets fast access to Nvidia GPUs, in about two to three months against a typical two to three years, the analyst said.
Few permanent jobs
Data centres cost a great deal to build but employ few people once running. The authors cite Georgia Tech research by Daniel Yue and Yiyang Zeng on US data centres. It found metropolitan counties capture most of the economic gains, while facilities in rural areas “tend to employ relatively few permanent workers — often fewer than 100”.
The investment numbers do not add up neatly
Zankore’s investment is reported at about Rp82 trillion, around US$4.59 billion, more than the Batang zone’s five-year target of Rp75.8 trillion. Separately, the ministry told Antara that a full-scale AI factory needs “about US$50 million in investment for every one megawatt”. By that yardstick, 200 MW would cost about US$10 billion and 1 GW about US$50 billion. Nobody has explained the gap; the Rp82 trillion may cover only part of the Zankore AI factory build.
What the Zankore AI Factory Has Not Disclosed
The authors’ core demand of the Zankore AI factory is transparency. “At minimum, Zankore should be required to publicly disclose its projected water consumption, water sources, cooling technology and contingency plans during droughts,” they write. They also call for “clear rules for allocating water between industrial users and local communities”.
| Disclosure the authors ask for | Public as of 9 October 2026? |
|---|---|
| Projected water consumption | No |
| Water sources | No |
| Cooling technology | Reference design named (Nvidia DSX); cooling system not confirmed |
| Drought contingency plans | No |
| Rules for sharing water with communities | No published allocation rules |
A precedent from West Java
Indonesia has already paused one AI data centre over permits. On 30 September, Detik reported that West Java’s provincial government had temporarily halted an AI data centre at Jatiluhur in Purwakarta. Its environmental impact assessment (AMDAL) and building approval were still being processed. The authors say the same scrutiny should apply there.
Why disclosure helps the operator too
“An AI factory is designed to operate continuously. Its servers cannot simply shut down because water becomes scarce,” the authors note. A clear water plan protects the Zankore AI factory’s own uptime as much as it protects residents. As we reported in our piece on regulating rather than banning data centres, clear conditions can give communities more leverage than outright opposition.
What the Zankore AI Factory Means for Businesses Buying AI
Few of our readers will build a gigawatt AI factory. Many already buy cloud AI services that run in places like Batang, and the water question increasingly reaches their own supply chains and sustainability reports.
Ask providers for water data by region
Water efficiency varies hugely by location. AWS’s own figures show its Jakarta region at 2.85 litres per kWh in 2025. When choosing where AI workloads run, ask providers for water usage effectiveness by region, not just a global average. Our data centre operations specialists can help compare providers on this.
Treat water as a governance risk
Water scarcity can force throttling or relocation, and public opposition can delay capacity. For any business relying on AI capacity in water-stressed regions, that is an availability risk as well as a reputational one. It belongs in supplier due diligence alongside cybersecurity and resilience checks, as part of wider IT governance.
Zankore AI Factory: Frequently Asked Questions
What is the Zankore AI factory?
A planned AI data centre platform in Batang, Central Java, backed by Indosat Ooredoo Hutchison, Ooredoo Group, Nvidia and Nokia. It targets 1 GW of capacity, starting with about 200 MW in 2027.
How much water would the Zankore AI factory use?
Zankore has not said. Economists writing in The Conversation estimate 4.8 million to 9.1 million litres a day for the first 200 MW if it uses evaporative cooling. Our check against Nvidia and AWS benchmarks gives 5.4 million to 13.7 million litres.
Could the Zankore AI factory use no water at all?
Possibly. Nvidia says its DSX design can reach near-zero water use with closed-loop liquid cooling and dry coolers, but only in favourable climates. It has not said whether Batang qualifies.
Who owns Zankore?
Ooredoo Group holds 49%, Indosat Ooredoo Hutchison 37%, Nvidia 11% and Nokia 3%, according to figures reported by Investor.id.
Why is Batang’s location a concern?
Central Java has just ended a four-month drought alert affecting 2.34 million people, and the Pekalongan-Batang coast is sinking by around 6 to 20 centimetres a year.
References
Zankore’s AI Factory: Can Central Java afford the water cost? (The Conversation)
Indosat, Ooredoo Group, Nokia, and NVIDIA launch Zankore by Indosat (Nokia)
AI Factory strengthens Indonesia’s position as regional AI hub (Antara)
Zankore secures $3.1bn senior term loan facility for AI cloud build-out (DCD)
Indosat (ISAT) kuasai 37% Zankore (Investor.id)
Investasi data center AI Rp82 triliun masuk Batang (Warta Ekonomi)
Liquid cooling for AI factories (Nvidia)
AWS Indonesia sustainability fact sheet (Amazon)
Making AI less thirsty: the secret water footprint of AI models (arXiv)
2.3 juta jiwa di Jateng terdampak kekeringan (Metro TV News)
Bangun infrastruktur KEK Batang (Kompas)
Proyek pusat data AI Jatiluhur disetop sementara (Detik)
More AI coverage: explore Progressive Robot's AI Models, Tools & Releases hub — hands-on reviews, setup guides and benchmarks in one place.
