OpenAI's Australian Data Centre Drops Water Recycling Plan as Infrastructure Gaps Force a Hard Choice Between Water and Energy
The 612-megawatt S7 facility in Sydney will now rely on energy-intensive fans instead of recycled sewage water for cooling, a pivot that exposes the widening gap between the AI sector's breakneck expansion and the slower pace of local utility investment.
SYDNEY: The developers behind a massive data centre that OpenAI wants to build in Australia have abandoned plans to cool it with recycled water, switching to a method that uses far more electricity after discovering that the pipeline infrastructure needed to ship treated sewage did not exist (Reuters, 2026). The decision will intensify scrutiny on a project already under pressure from regulators trying to curb the resource demands of an AI construction boom that economists say will top $1 trillion in Asia by 2030.
The 612-megawatt S7 facility, planned for western Sydney by OpenAI's local partner NEXTDC, had originally proposed tapping the city's supply of recycled water to chill its servers without drawing on drinking water or adding to the strain on the energy grid. That plan fell apart when the company could not secure planning permission for a pipeline to carry the water. "Recycled water has real merit and it can support lower PUE and better overall sustainability and NEXTDC considers it wherever that infrastructure exists," CEO Craig Scroggie told Reuters. "In Western Sydney, the enabling infrastructure is not yet in place."
With recycled water off the table, the data centre will now use a closed-loop liquid cooling system paired with fans to expel heat. The method uses no water, but experts say it consumes substantially more energy. Sarvaesh Mohan, an analyst at data centre consultancy DCByte, said the switch illustrated "the wider gap between the current AI infrastructure boom and the slower deployment cycles of local utilities." NEXTDC declined to say whether the new cooling design would raise the facility's overall power draw, and OpenAI referred questions back to its partner.
The energy implications matter because Sydney's grid is already stretched to its limit. A spokesperson for Transgrid, the city's grid operator, said the company had "limited capacity to connect new large data centres to the transmission network in Sydney without augmenting the bulk network," adding that data centre operators would need to fund any expansion themselves. In May, a Transgrid executive told a state parliamentary inquiry that existing regulatory arrangements were never designed for this kind of clustered, large-scale load growth.
Australian authorities are scrambling to catch up. The federal government this month said it would impose energy and water use limits on data centres, though that move trails new state-level rules in New South Wales. A Reuters review found that since late 2025, state authorities have begun requiring data centre developers to disclose water usage figures and commit to a fixed power usage effectiveness ratio. The OpenAI-NEXTDC project is one of six facilities facing the new conditions and must now commit to a PUE limit of 1.3, a figure at the lower end of global averages. NEXTDC has not yet disclosed a target PUE for S7.
David Pocock, an independent senator who has raised concerns about data centre developments, said new federal laws should include mandatory requirements for water usage, renewable energy, and planning controls. The tension is particularly acute in Australia, the driest inhabited continent, where residents can face water restrictions during drought. NEXTDC says it has deployed the waterless cooling method before, cooling 50 megawatts of capacity, but that is one-twelfth the scale proposed for the OpenAI project.
BuiltWorld AI Operational Take: Losing access to recycled water because the pipeline does not exist is a problem that will repeat in any jurisdiction where data centre timelines outpace utility planning cycles. For project leads, the lesson is that water and power infrastructure feasibility needs to be locked down before a cooling technology is selected, not after. A forced switch to energy-intensive air cooling late in the design phase can trigger a cascade of second-order problems: higher electricity costs, a bigger backup generation requirement, and a PUE target that suddenly looks very hard to hit.