Hyperscalers Face Trillion-Dollar Conundrum as AI Bubble Risks Grow

MIT Technology Review has warned about the potential risks of an AI bubble bursting, estimating that hyperscalers such as Alphabet, Microsoft, Amazon, Meta, and Oracle may need to nearly triple their productivity by 2030 just to break even on their trillion-dollar infrastructure bet. According to a study led by Wharton finance professor Jessica Wachter, who previously served as chief economist at the Securities and Exchange Commission (SEC), hyperscalers have a lot of ground to make up. The calculation accounts for the cost of capital, a 15% return, and depreciation of the assets.

The current buildout will be the largest misallocation of capital in history if hyperscalers fail to achieve this growth rate, Wachter and her coauthor concluded. Alphabet posted a $5.9 billion free cash flow deficit last quarter, its first since going public in 2004, and investors are increasingly viewing AI spending as a risk to markets. The concentration of debt among hyperscalers is growing, with Morgan Stanley calculating that they will finance over half of their planned $2.9 trillion in data center spending through 2028 through external capital instead of cash reserves.

Private-credit firm Blue Owl Capital recently acquired an 80% stake in Meta's Hyperion data center, and Columbia Business Schools' Stijn Van Nieuwerburgh warns that debt like this increasingly flows through pension funds and private credit vehicles. Few people realize how deeply their own retirement and insurance savings are exposed to these risks.

Crypto strategist Arthur Hayes has floated a related scenario, arguing that an AI credit bust could force the Federal Reserve to print money, pushing Bitcoin (BTC) toward $1 million. Former SEC chair and MIT Sloan School professor Gary Gensler expects a retrenchment eventually, though its timing remains uncertain. The outcome will depend on whether it arrives gradually or abruptly.