Indonesia's proposed ZanKore AI factory makes a familiar infrastructure question unusually visible: who supplies the resources behind a gigawatt of computing ambition? The project could expand domestic access to AI hardware and support local-language services. Its environmental case, however, cannot be established from accelerator specifications alone. Cooling design, water allocation and drought procedures determine what the announced capacity would mean for Batang and surrounding communities.
Distinguish the target from the first phase
ZanKore's partners describe a long-term target of one gigawatt and an initial deployment in Central Java. Public announcements have referred to approximately 200 megawatts in the first half of 2027 and a later initial 100-megawatt infrastructure stage. Those statements should be read as staged plans, not as one gigawatt already operating.
The platform links NVIDIA hardware and infrastructure design with Indonesian AI ambitions, including Sahabat-AI. Domestic computing can offer local processing and serve regional customers. These are plausible benefits, but neither a hardware choice nor a sovereign-computing objective establishes the site's resource efficiency or actual customer demand.
Follow heat beyond the processor
Liquid cooling moves heat from chips into a fluid loop. The building must still release that heat outdoors. Depending on its design, it might use evaporation, dry coolers, chillers or a hybrid system. A closed internal loop therefore does not automatically mean a waterless facility.
Water Usage Effectiveness expresses on-site water consumption relative to computing energy, usually in liters per kilowatt-hour. It is useful only when its boundary and assumptions are clear. Annual averages can conceal demand on the hottest, driest days, exactly when other users may face shortages.
One analysis applied an assumed rate of about 1.9 liters per kilowatt-hour to a continuously loaded 200-megawatt installation, producing roughly 9.1 million liters daily. This is a scenario, not ZanKore's measured or confirmed consumption. Utilization, cooling architecture and weather can change the result substantially. It should prompt disclosure rather than become a headline assertion of inevitable demand.
Assess local supply without misleading comparisons
Central Java reported distributing more than 75 million liters of emergency water during its 2026 dry season. That was aid delivered across many places over months, not a reservoir reserved for this project. Comparing the total directly with a hypothetical daily industrial demand does not prove the facility would take emergency household supplies.
The relevant question is whether industry, residents, farming and ecosystems share constrained sources. Groundwater, surface water, reclaimed wastewater and seawater have different risks. Avoiding drinking water alone does not settle allocation, subsidence, treatment, coastal intake or discharge concerns.
Make expansion conditional on evidence
Before commissioning, publish the cooling architecture, water source, annual budget and peak daily draw for each phase. Separate on-site consumption from water associated with electricity generation. Explain seasonal operating modes, discharge quality and who can require workload reductions during drought.
A smaller first phase can provide evidence for later expansion if measurements and limits are agreed beforehand. Track total water and energy alongside efficiency ratios: better computing per megawatt can still accompany greater total consumption.
ZanKore's opportunity is substantial, but the useful decision is more precise than approving or rejecting AI investment in general. Require a transparent resource plan that protects continuous computing and local water access together, then judge expansion against observed performance.