TL;DR · 30-second read
The Short Version
A huge computer center planned for Sydney, Australia, and linked to the maker of ChatGPT was going to cool its machines with recycled sewage water.
That plan is off. The developer could not get permission to build the pipe that would carry the water, so it is switching to a sealed system that keeps reusing the same liquid.
The catch: the new design uses far less water but significantly more electricity. One missing permit changed the project’s environmental trade-off.
Data Center Dynamics reported on July 23, 2026, citing Reuters, that the planned S7 data center in Sydney, Australia, will no longer be cooled with recycled sewage water. The 612-megawatt (MW) facility, developed by NextDC and linked to OpenAI through a December 2025 memorandum of understanding, will instead use a closed-loop liquid cooling system.
The change followed a failure to secure planning permission for a pipeline that would have served the site. The new design is expected to reduce water use significantly but require significantly more energy than the earlier plan.
Executive Summary
S7 is planned for a 255,135-square-meter parcel in Eastern Creek, a logistics hub in Sydney’s western suburbs. NextDC bought the land for AU$353 million (US$244 million) in October 2024 and submitted proposals for the project in mid-March. Under the original design, sewage water would have reduced the site’s dependence on drinking-quality supplies and significantly lowered the energy needed for cooling.
The switch matters because it does not simply make the project better or worse environmentally. It moves the burden from one resource to another: less water, more electricity. Without published figures for either side of that exchange, the net effect cannot yet be judged.
It also offers a practical lesson for AI infrastructure. A cooling design can depend on approvals for infrastructure that sits beyond the main building, and if those approvals do not arrive, the site’s efficiency and resource profile can change with them.
The Permit Decided the Cooling Design
The chain of events is short. Recycled water had to reach the site, reaching the site required a pipeline, and the pipeline required planning permission that was not secured. Once that link failed, the cooling plan for a 612MW facility had to change. The obstacle described was not the performance of the cooling technology, but access to the water supply it relied on.
NextDC CEO Craig Scroggie made a similar point in an email to Reuters: “Recycled water … has real merit and it can support lower PUE and better overall sustainability and NextDC considers it wherever that infrastructure exists.” The final clause is the important one. It treats recycled water as an option that depends on available infrastructure, not as a fixed feature of every design.
This has direct consequences for developers, tenants, planners and water utilities. Anyone relying on a recycled water scheme needs to know whether the supply connection is approved, not only whether the water source exists. One project does not establish an industry trend, but S7 shows how a single off-site approval can alter a large facility’s design.
Less Water, More Electricity
Data centers turn almost all the electricity they consume into heat, which must then be removed. Designs that use water can reduce the electricity spent on cooling. A closed-loop system instead keeps the same fluid circulating and generally relies on other equipment, such as fans or chillers, to reject heat. S7’s specific equipment choices have not been detailed, but the reported result fits that general trade-off: water use falls and energy use rises.
The metric at issue is power usage effectiveness (PUE), the ratio of a facility’s total energy use to the energy reaching its computing equipment. A score closer to 1.0 means less overhead for cooling and other supporting systems. Scroggie’s comment links recycled water with a lower PUE, so a higher energy requirement after the switch would be expected to work against that score.
Whether this is a good trade depends on local conditions and the size of each change. Lower water use may be valuable to a growing metropolitan area, while higher electricity demand adds to the load on the power system and could affect operating costs and emissions. “Significantly” describes both sides of the exchange, but it does not show which effect is larger.
What a Water Plan Is Worth Before It Is Built
S7 tests how much weight an early cooling plan should carry. The recycled water design appeared in a planned project, not an operating one, and nothing reported indicates that it was a binding public commitment. Changing it after losing a key input is an ordinary development decision. Even so, resource claims made at the proposal stage may not describe the facility that is eventually built.
The change has been framed as OpenAI’s retreat, but the sequence described points to NextDC, the primary developer, changing its plans after the pipeline setback. OpenAI’s disclosed connection is a memorandum of understanding, a preliminary agreement, to use some of the capacity with an option to scale. How far OpenAI shaped the cooling decision remains unclear, which matters when assigning responsibility for the site’s water and energy outcomes.
Scale explains why the question will recur. S7 alone would exceed the just under 500MW across NextDC’s six other Sydney data centers in operation or development. It would also equal about 6% of the more than 10 gigawatts of capacity OpenAI has contracted globally, although OpenAI has agreed to use only part of the site. Buyers and local stakeholders can reasonably ask which environmental targets would survive a design change, and whether those targets are tied to outcomes or to a particular technology.
Background
NextDC has six data centers in operation or development in Sydney, totaling just under 500MW and 178,500 square meters. They include S3 and S6 in Artarmon, S1, S2 and S5 in Macquarie Park, and S4 in Horsley Park. S7 in Eastern Creek would be larger in capacity than that group combined.
OpenAI, the maker of ChatGPT, has secured more than 10 gigawatts of contracted AI data center capacity worldwide. Most of it sits within its Stargate portfolio, which includes a flagship site in Abilene, Texas, and locations in Shackelford County and Milam County, Texas; Doña Ana County, New Mexico; and Lordstown, Ohio. Further international sites are also planned. Source: OpenAI data center in Sydney, Australia, abandons plans for recycled water cooling system – report, Data Center Dynamics’ coverage of the cooling change at the planned NextDC S7 facility.Sources