OpenAI has scrapped plans to cool its planned S7 data center in Sydney with recycled wastewater, opting instead for a closed-loop liquid cooling system that will consume significantly more electricity. The decision marks a setback for sustainability efforts in the booming AI infrastructure sector, as the company faces regulatory hurdles that prevented construction of a dedicated wastewater pipeline.

Why Did OpenAI Abandon Its Original Water-Cooling Plan?

The S7 facility, a 612-megawatt data hub ranked among the world's largest, was initially designed to tap Sydney's treated wastewater to cool its servers. This approach would have reduced dependence on potable drinking water and significantly lowered the cooling system's energy requirements. However, NextDC, the developer partnering with OpenAI, encountered regulatory roadblocks when attempting to secure permits for a dedicated wastewater pipeline.

"Recycled water has real merit and it can support lower PUE and better overall sustainability, and NEXTDC considers it wherever that infrastructure exists. Potential sources were canvassed in early planning. In Western Sydney, the enabling infrastructure is not yet in place," said Craig Scroggie, CEO of NextDC.

Craig Scroggie, CEO at NextDC

PUE, or power usage effectiveness, is a closely watched metric that measures the energy used for cooling and other overhead costs beyond actual computing. The original recycled water plan would have achieved a lower PUE score, making the data center more efficient overall.

What's the Environmental Trade-Off of the New Cooling System?

The revised cooling approach uses a closed-loop liquid cooling system that circulates coolant directly across processing chips and dissipates heat into the air using large fan arrays. While this system uses no drinking water, it comes with a significant drawback: the fan arrays will consume substantially more electricity than the original water-based design.

The timing of this shift is particularly concerning for Sydney's power infrastructure. The city's grid operator has already warned that transmission capacity is stretched thin by the rapidly growing demand from data centers. Adding a more energy-intensive cooling system to one of the world's largest AI facilities could exacerbate these capacity constraints.

How Does This Fit Into the Broader AI Infrastructure Boom?

The S7 data center represents OpenAI's expanding physical footprint to support its growing AI operations. In December 2025, OpenAI and NextDC announced a Memorandum of Understanding that would give ChatGPT's maker access to some of the data center's capacity, with options to scale over time. The facility is expected to deliver thousands of direct and indirect jobs during construction and ongoing operations, including technical, manufacturing, engineering, and operational roles.

This development underscores a broader tension in the AI industry: the massive computational demands of training and running large language models like GPT-4 and future versions require enormous amounts of energy and cooling infrastructure. As AI adoption accelerates globally, companies face mounting pressure to balance rapid expansion with environmental sustainability.

Steps to Understanding AI's Infrastructure Challenges

• Energy Consumption: Large AI data centers like OpenAI's S7 facility consume hundreds of megawatts of power, comparable to small cities, making cooling and energy efficiency critical operational concerns.
• Water vs. Electricity Trade-Offs: Sustainable cooling methods like recycled water systems reduce electricity use but require supporting infrastructure; when unavailable, facilities must resort to more energy-intensive alternatives.
• Grid Capacity Constraints: Cities like Sydney face transmission bottlenecks as multiple data centers compete for limited power supply, forcing grid operators to manage demand carefully.
• Regulatory Barriers: Permitting processes for infrastructure like wastewater pipelines can delay or derail sustainability initiatives, pushing companies toward less efficient but faster-to-implement solutions.

The AI industry's explosive growth means that infrastructure decisions made today will shape energy consumption patterns for years to come. OpenAI's shift away from recycled water cooling illustrates how regulatory delays and infrastructure gaps can inadvertently push companies toward less sustainable solutions, even when they initially planned otherwise.

As generative AI adoption continues to accelerate, with nearly 53% population-level adoption of generative AI reached within three years according to Stanford University's 2026 AI Index, the infrastructure demands will only intensify. This makes the efficiency of data center cooling systems increasingly important to global energy sustainability.