NEXTDC, the Australian company building a data centre that OpenAI wants to anchor, has quietly abandoned a plan to cool the facility with recycled sewage water. The replacement method uses no drinking water, which sounds like a win. It uses more electricity, which is the problem, because Sydney's energy grid is already full.

The facility in question is the S7 project in Western Sydney, a 612-megawatt complex that would rank among the largest data centres on the planet. OpenAI intends to help plan, build, and run it, and to buy computing power from it. When the project's planning documents were filed in April, they described a cooling system that would pipe treated sewage water to the servers, reducing the demand on both drinking water supplies and the electricity grid. That plan is now dead.

NEXTDC told Reuters the recycled water pipeline was abandoned because the infrastructure to deliver it simply does not exist in Western Sydney yet. The permits were not in place, the pipes were not there, and building them would have required approvals that could not be obtained in time. The company's CEO, Craig Scroggie, said recycled water "has real merit" but acknowledged the "enabling infrastructure is not yet in place."

Why this matters beyond one data centre

The fallback cooling method circulates liquid coolant around the chips and then fans the heat away. It avoids drinking water entirely, which matters in Australia, the driest inhabited continent, where residents already face periodic water restrictions during drought. But the process draws more power from a grid that, by Sydney's own energy operator's account, is at capacity due to data centre connection requests.

The grid operator for Sydney, Transgrid, told a state parliament inquiry in May that "existing regulatory arrangements were not designed for this level of large, clustered load growth." In plain terms: the rules were written for a world where nobody was trying to plug a dozen supercomputers the size of warehouses into the same stretch of suburbia simultaneously.

That leaves the S7 project, and others like it, caught between two resource constraints at once. Water restrictions push developers toward electricity-heavy cooling. Grid congestion pushes them toward water. Neither escape hatch is fully open.

Sarvaesh Mohan, an analyst at data centre consultancy DCByte, told Reuters the situation would be "more energy intensive" and said the abandoned recycled water plan illustrated "the wider gap between the current AI infrastructure boom and the slower deployment cycles of local utilities." The gap is real and growing. Data centres make infrastructure commitments that last decades. Pipelines, substations, and regulatory frameworks move on timelines measured in years, sometimes longer.

Kurt Dahl, the CEO of a recycled water company named in the original planning documents as a possible supplier, did not comment on the specific decision but said connecting data centres to recycled water systems should be treated "with urgency, as data centres are currently making long-term choices between energy and water."

The pattern behind the project

Australia is competing aggressively for a share of the global AI construction wave. Economists project AI infrastructure investment in Asia will exceed $1 trillion by 2030, and Sydney is positioning itself as a regional hub. But the resource math is proving harder than the ambition. Every data centre that defaults to energy-heavy cooling because the recycled water grid isn't ready is a compounding load on a system that regulators and grid operators admit they did not design for this moment.

For ordinary Australians, the near-term consequence is less abstract than it sounds. A strained electricity grid in a city where housing costs are already punishing, where summers are getting longer, and where air conditioning is not optional is a grid that will face hard questions about who gets priority when demand peaks. A single data centre does not crash a grid. A cluster of them, each making the same default choice, begins to look like a policy problem that arrived before the policy did.