A 612-megawatt data center, one of the largest artificial intelligence facilities in the world, has abandoned its planned water-recycling cooling system. The Star reported on August 16 that the OpenAI project, a scale of computing that places it among the biggest such facilities ever built, will shift to a different cooling approach. The change follows a similar decision at OpenAI’s data center in Australia, which earlier dropped a plan to use recycled water for cooling.
The pattern across the two projects is the story. AI infrastructure is being built so quickly, and at such scale, that the traditional assumptions about cooling no longer hold. Data centers of this size generate heat in volumes that overwhelm conventional air conditioning, which is why designers turned to water-based systems in the first place. But water is scarce in many of the regions where AI projects are being built, and the permitting, environmental review, and community opposition that accompany water use have become as much of a constraint as power supply.
The 612-megawatt figure puts the project in perspective. A facility of that size would consume electricity equivalent to a small city, and its cooling needs would strain any local water supply. The decision to drop the water-recycling plan suggests the developers concluded that air-based or other cooling methods could handle the load, or that the water requirements were not obtainable on the terms originally planned. People familiar with the project say the change was driven by engineering review rather than by a specific regulatory rejection, and that the facility’s power design is unchanged.
The broader shift is visible across the industry. Hyperscale data center operators, including the cloud arms of Microsoft, Amazon, and Google, have all been rethinking water use as they expand into regions with limited water. Several have signed pledges to reduce water consumption per unit of computing, and some have shifted designs to favor dry cooling or closed-loop systems that use far less water. The trade-off is efficiency: water cooling removes heat more effectively than air cooling in warm climates, and abandoning it can raise energy costs or force the facility to operate at lower density.
The tension between power and water defines the new phase of the AI buildout. For the past two years, the binding constraint on AI expansion has been electricity, with data center developers competing for grid connections, power purchase agreements, and new generation capacity. Water is emerging as the second constraint, and projects that cannot secure water for cooling face the same delays that power shortages once caused. The result is that site selection for AI facilities now weighs water availability as heavily as electricity prices.
OpenAI’s global construction program makes it a test case. The company has announced or begun data center projects in the United States, including major campuses in Ohio and Texas, as well as in Australia and other countries, and each project has had to negotiate local power, water, and environmental rules. The company has said it wants its facilities to be sustainable, and its cooling decisions are being watched as precedents for the hundreds of facilities that other companies will build in the next decade.
The environmental scrutiny is intensifying. Community groups and regulators have begun to question the resource consumption of AI data centers, and several jurisdictions have introduced requirements for water-use reporting or efficiency standards. The concerns extend beyond water: data centers also raise questions about land use, noise, and the emissions associated with backup power generation. The industry’s answer has been a mix of efficiency improvements, renewable power, and new cooling technologies, but the pace of construction is outpacing the pace of innovation in some areas.
The economics push in the same direction. Water rights are expensive, water infrastructure is costly to build, and the permitting process for water use can take years. For a project that must come online quickly to serve the AI boom, avoiding the water question entirely can be the fastest path. The trade-off is that dry cooling systems may consume more electricity, raising operating costs over the facility’s life, and may not be able to cool the highest-density racks that AI workloads prefer.
The cooling decision at the 612-megawatt project will be studied by every developer with a similar facility in planning. If the alternative system performs well, the industry will follow, and water use per unit of AI computing could fall across the board. If it underperforms, the industry will face a harder choice between water-intensive designs and less efficient cooling, at a time when both power and water are becoming scarcer relative to demand.
For OpenAI, the change is one detail in a much larger program, but it is a telling one. The company’s expansion is no longer just a story of capital and chips; it is a story of public resources, local approvals, and physical infrastructure. The decision to drop the water plan shows that even the most well-funded technology company in the world must bend to the constraints of water, power, and community, and that the AI buildout will be shaped as much by environmental limits as by engineering ambition.
