Hon Chung Lau and Steve C. Tsai had a question: as AI drives a massive boom in data centers across America, where will all the electricity come from — and what happens to the air we breathe?
The two researchers at Rice University and Low Carbon Energies LLC set out to answer that. Their study, published in the journal Energy & Fuels, looked at how fast data centers are growing and how much carbon dioxide they could release into the atmosphere — and whether something practical could be done about it.
Their findings are striking. American data centers currently use about 40 gigawatts of power — roughly the output of 30 large power plants. By 2030, that number could reach 169 gigawatts. Without new strategies, the carbon dioxide released from powering those centers could jump from 90 million metric tons per year today to more than 404 million metric tons per year in just five years.
But here is the hopeful part. The researchers found a natural solution already underground: structures called saline aquifers, which are layers of porous rock soaked in salty water, buried deep beneath the surface. These aquifers can safely trap carbon dioxide permanently. The team mapped 34 states where these aquifers have enough room to store more than 100 years of projected data center emissions — not just for one year, but for generations. By 2030, those aquifers could store enough to capture roughly 74 percent of all data center-related emissions. If nearby states share storage capacity, that number climbs above 90 percent.
Texas, one of the biggest growth areas, would need to add roughly 25 gigawatts of power capacity to keep up with demand. But many of the states seeing the fastest data center growth happen to sit above the very formations that could store the carbon their power plants release.
The researchers suggest that natural gas power plants equipped with carbon capture technology could be a practical near-term option. Natural gas burns cleaner than coal, and the infrastructure already exists. Combined with underground storage, it could buy time while renewable energy scales up.
The team acknowledges their numbers are conservative — they only counted data centers that have been publicly announced, not future projects still in planning. Even so, Lau says the work offers a useful roadmap for states trying to grow their digital economies without worsening climate change.
"The AI economy will require enormous amounts of energy," Lau said. "Our study helps identify where that demand is growing, where emissions are likely to rise and where carbon storage could help."