What Must Happen for AI’s Trillion-Dollar Gamble to Pay Off

Summary

AI hyperscalers are making an unprecedented infrastructure bet: their data-center expenditures could approach $1.1 trillion by 2027, while combined AI revenue is estimated at only $150 billion to $200 billion this year. Research by Wharton professor Jessica Wachter and a coauthor estimates that the companies would need to raise their own productivity 2.7-fold to break even by 2030, assuming the cost of capital, a 15% return, and asset depreciation. Other estimates suggest that if roughly 183 gigawatts of planned capacity are built between 2025 and 2032, the sector could require about $3.7 trillion in annual revenue by 2032 to deliver a 10% return. The investment case therefore depends on three linked outcomes: hyperscalers must generate much larger revenues, AI must produce broad productivity gains for customers, and expensive frontier models must remain competitive with cheaper models that may be good enough for many businesses. Current economy-wide data show little or no productivity gain from AI, although executives surveyed across four countries expect future improvements and plan substantial additional spending. The article also warns that AI infrastructure is increasingly debt-financed; Morgan Stanley estimates that more than half of $2.9 trillion in hyperscaler data-center spending from 2025 to 2028 could come from external capital. Debt, rapid GPU depreciation, and uncertain future demand could leave data centers as stranded assets if companies cannot keep upgrading them or fill their capacity. Meta’s Louisiana project illustrates how these risks spread beyond company balance sheets: its planned campus grew from $10 billion and two gigawatts to a reported $50 billion and five gigawatts, alongside expanded gas-fired power plans. Complex leases and guarantees could expose lenders, utilities, and residential ratepayers if demand changes or Meta exits. The article argues that a market retrenchment is likely, but that a financial correction and the underlying AI technology could have different futures. AI may survive a bubble burst, while the long-term value of the massive data centers and the economic obligations tied to them remain uncertain.