University of Texas Study Proposes Using Oilfield Wastewater for AI Data Center Cooling

Researchers at the University of Texas at Austin have proposed a novel solution to the increasing water demands of AI data centers by modeling the use of oilfield wastewater and desalinated brackish groundwater for cooling purposes. The study, published in the journal Water Research, indicates that utilizing these water sources could significantly reduce the overall water footprint of data centers.

Led by mechanical engineering professor Vaibhav Bahadur, the research explores established desalination techniques, particularly reverse osmosis, to determine their effectiveness in generating cooling water from less desirable sources. The findings suggest that the energy and cost associated with treating these water sources would constitute only a small fraction of a data center's total energy consumption.

A key innovation from the study is the introduction of the Saline Water Utilization Intensity (SWUI) metric, which assesses local water stress under various treatment scenarios. The researchers found that a proposed data center project in Longview, Texas, which would utilize brackish water with a two-stage reverse osmosis system, scored favorably in terms of water efficiency.

Notably, the Permian Basin generates approximately 8 billion gallons of produced water daily, equivalent to the volume of 1,200 Olympic-sized swimming pools. In contrast, a large data center may require up to 5 million gallons of water each day for cooling.

The timing of this study is particularly critical, as Texas Governor Greg Abbott recently paused all new data center permits amid heightened scrutiny over water usage by existing facilities. This research could provide data center developers with a compelling argument to regulators, advocating for the use of brackish or produced water as a more sustainable alternative to freshwater.

However, the study remains a modeling exercise, and practical challenges such as brine disposal from desalination processes will need to be addressed in any large-scale implementation.