When Jessica Wachter, a finance professor at the University of Pennsylvania’s Wharton School, wanted to understand how artificial intelligence (AI) might affect the economy in the coming years, she looked at the spending habits of hyperscalers—large tech companies like Alphabet, Microsoft, Amazon, Meta, and Oracle. These firms are investing heavily in AI data centers, which are massive facilities that power AI technologies. Wachter estimated that by 2027, these companies will have spent nearly $1.1 trillion on AI infrastructure. To justify this spending, she and her collaborator calculated that these companies would need to boost their productivity by about 2.7 times by 2030, considering costs, returns, and the wear and tear of their assets. If they don’t meet this goal, they might struggle to cover their debts or even face bankruptcy. These hyperscalers are already spending around $750 billion this year on AI data centers, with total investments potentially reaching over $5 trillion in the next four years. However, their revenue from AI is currently much lower, estimated at between $150 billion and $200 billion this year. This creates a significant gap between what they’re spending and what they’re earning. Gary Gensler, a former head of the U.S. Securities and Exchange Commission (SEC) and now a professor at MIT, points out that the real question is whether these large investments will eventually pay off. The financial risks are increasing as these companies borrow more money to fund their infrastructure. For example, Alphabet recently reported a free cash deficit—meaning it spent more than it earned—of $5.9 billion, the first such deficit since 2004. Although these companies currently have the financial strength to handle their debt, the long-term success of their AI investments depends on future revenue growth and the productivity gains from AI. Maintaining and upgrading AI infrastructure is also getting more expensive. Graphics processing units (GPUs), which are critical for AI and make up about 60% of the costs, are improving rapidly. Their performance doubles roughly every two years, meaning data centers must constantly invest in new, more powerful chips to remain competitive. Mihir Kshirsagar, from Princeton’s Center for Information Technology Policy, warns that without these updates, data centers could become outdated, like “hulks,” or stranded assets scattered across the country with no market value. For these investments to be sustainable, AI needs to drive broader economic growth. Stijn Van Nieuwerburgh, from Columbia Business School, estimates that by 2032, AI companies will need to generate about $3.7 trillion in annual revenue to justify their spending, assuming a 10% return. This growth depends on AI’s ability to boost productivity across the economy, which has yet to be fully realized. Public opinion and the impact of AI on jobs are also important factors. Many business leaders believe AI will increase sales while reducing the number of employees needed. However, this could lead to public resistance if people perceive AI as taking away jobs. Daron Acemoglu, an MIT economist, notes that without real productivity gains, public support for AI investments may decline. The financial risks of AI investments extend beyond the hyperscalers. Complex financing arrangements for data centers have involved financial institutions, pension funds, and even life insurance policies. For example, Meta’s Hyperion data center in Louisiana is part of a joint venture with Blue Owl Capital and includes a series of intricate lease agreements. These arrangements increase the financial risk for all parties involved if the data centers don’t generate the expected returns. The future of AI investments remains uncertain. While some predict a potential slowdown as the market adjusts, others see the current spending as essential for long-term innovation. Whether these investments pay off will depend on AI’s ability to deliver on its promises of productivity and economic growth, as well as the financial resilience of the companies and institutions involved.