Artificial intelligence’s effect on the environment is clouding the indicators officials commonly use to set interest rates, according to a Bank for International Settlements study.

Changes in productivity, climate resilience and energy constraints driven by the technology are altering potential output and make capacity utilization and price pressures harder to gauge, researchers Leonardo Gambacorta and Salvatore Polizzi said in a paper released on Thursday.

Key challenges they identified include how AI may push up energy prices before productivity gains lower costs, complicating how officials assess demand and supply. The ultimate climate impact will depend on whether efficiency gains outweigh rising data-center power use, and how that electricity is produced, they said.

“Ultimately, AI is neither inherently a climate solution nor inherently a climate problem,” the authors wrote, while adding that “The coming decade will be critical in determining whether AI becomes a force that accelerates the transition to a more sustainable economy or one that intensifies existing environmental pressures.”The academics also flagged potential financial stability risks from the technology, saying that large investments in data centers and electricity grids could cause sudden corrections if expectations turn out to be overly optimistic.

Such concerns echo repeated warnings of the Basel-based institution that a crash after the AI boom has the potential to cause large-scale damage. In a speech in September, BIS chief Pablo Hernandez de Cos said valuations of AI companies rest on “ambitious expectations” and have caused an “arms race” between them.