he recent boom in generative AI models relies on intense computational resources, built on power-hungry chips in server farms around the world. That demandby 2026, compounding the increasing need for electricity that’s already taxing the grid.
Utah-based startup Zanskar has developed machine learning models to solve one of geothermal’s biggest problems: finding ideal locations to drill. The exorbitant cost of drilling has largely kept geothermal from competing with other technologies like wind and solar, so much so that less than 1% of the U.S.’s electricity is produced with it.
Geothermal power is promising because it relies on heat from under the Earth’s surface, a theoretically endlessly renewable resource. To tap it, you drill a very deep hole, then bring up hot, pressurized water that turns to steam as it rises, powering the plant. Compounding the expense is that in the process of exploration, drillers don’t always hit the right spot. That’s because there are a lot of factors that go into determining an ideal location, from the mineral composition to the accessibility of water to avoiding the presence of too much natural gas. “The exploration part is what is most difficult,” Maria Richards, geothermal lab coordinator at Southern Methodist University, told.