According to Columbia professor Stijn Van Nieuwerburgh, America may need to spend roughly $3.5 trillion annually on AI services by 2032, equivalent to 8.8% of GDP. This expenditure is necessary to justify today’s massive data-center investment. He warns that expanding compute capacity could drive AI prices and margins sharply lower, potentially leaving revenues unable to support current investment. While AI could still deliver major productivity gains, economic benefits do not necessarily translate into investor returns.