AI infrastructure investments raise systemic financial risks, study warns

A new study warns that the scale of U.S. investments in artificial intelligence infrastructure could surpass historical projects like railroads and highways, raising systemic financial risks. Professor Stijn Van Nieuwerburgh from Columbia University’s Business School highlights that AI investments, initially funded by tech giants like Amazon, Meta, and Alphabet, are now expanding rapidly. By 2032, the U.S. is expected to allocate 3.6% of its GDP annually to AI, totaling over $10 trillion. This surpasses the 2.2% GDP share of railroad investments in the late 19th century. Van Nieuwerburgh warns that the complexity of AI financing, similar to the subprime mortgage crisis, could lead to significant risks if market expectations shift. The professor estimates that by 2032, AI will need to generate $3.7 trillion annually to meet current investment returns. Currently, OpenAI and Anthropic combined earn about $100 billion, requiring an 80% annual growth rate to meet this target. He also notes that the U.S. will add approximately 183 gigawatts of data center capacity over the next seven years, with existing capacity at 57 gigawatts. Concerns over energy and water usage have led some local governments to resist hosting large data centers, while the Federal Reserve is examining potential inflation impacts from this trend.