Barry Eichengreen warns that opacity prevails in the private debt markets that are funding the AI boom. Investors need to watch out because regulators are not making information easy to obtain

By Barry Eichengreen*

For months, if not years, AI has dominated speculation about the future, which has fueled speculation in the present. AI-related ventures are largely responsible for the rapid increase of the S&P 500 stock index. Investment in data centers is a major contributor to the rise in US interest rates, as borrowing for construction competes for credit with a deficit-prone federal government.

With so much investment riding on a fast-developing new technology, what happens if the cash flow and profits of the hyperscalers building data centers, and of AI firms renting that infrastructure (two categories that overlap), turn out to disappoint? What would this disappointment mean for financial markets? Would a crisis be in the offing?

This scenario could materialize for any number of reasons. The productivity benefits of AI could be overhyped, in which case firms contemplating the adoption of frontier models will curtail their demand for tokens, the basic units of data processed by such models. Or the availability of open-weight models, which enable users to tailor a model’s parameters to their needs, may allow those users to build small language models that run on their own infrastructure or even on laptops like the one I am using to write this commentary. Either way, the income accruing to investors in data centers will fall short of expectations.

The most immediate and visible impact would be a sharp drop in the lofty stock prices of the “Magnificent 7,” the seven ultra-large-cap tech and tech-adjacent firms that have driven the S&P 500. But stock-market corrections are no guarantee of crises or even of recessions. If the price of Meta shares falls, Mark Zuckerberg will buy fewer castles in Ireland. There would be a slowdown in consumption and investment spending. There would be pain for investment funds, like Situational Awareness, specializing in AI-related ventures. But a financial crisis would follow only if banks, insurance companies, and other mainstream institutional investors were implicated. And there is little evidence that they are heavily exposed to AI equity risk.

Developments in debt markets are more worrisome. With the cost of data-center construction exceeding their sponsors’ free cash flow, and with income from potential tenants still a distant prospect, much of this investment is being financed by borrowing from private credit funds (nonbank firms making privately negotiated loans). If demand for data-center services is lower than expected, these loans will fail to perform (a polite way of saying that borrowers will default).

The question then becomes who is invested in private credit funds. The popular image is of the same family offices and high-net-worth individuals who invest in venture capital and can afford occasional losses. But the reality is that the vast majority of private credit is provided by institutional investors: pension funds, insurance companies and, dare I say, university endowments.

Moreover, private credit funds have been taking steps to tap this institutional money by acquiring insurance companies, whose asset portfolios they can load up with loans. This invites the specter of an insurance industry crisis, which could play out in several ways. Insurance companies’ private credit parents could recapitalize them, assuming they have the funds. Alternatively, state guarantee funds, to which healthy insurance companies contribute, could fund the recapitalization, although there will be no healthy companies if the entire industry comes tumbling down. Lastly, the federal government could step in with a bailout, as it did for American International Group in 2008.

In addition to borrowing from pension funds and insurance companies, private credit firms borrow from commercial banks, using their portfolio of loans as collateral, thereby leveraging their commitments and juicing their returns. Even if banks don’t lend to or invest in AI firms and data centers directly, they still may be on the hook. We can only hope that their regulators are watching.

Finally, private credit firms originally built on a buy-and-hold model increasingly securitize their loans. They package those loans through special purpose vehicles that issue bonds backed by the associated cash flows, where the bonds are divided into risk tiers or tranches claiming first, second, and third dibs on debt-service payments. The so-called “mezzanine tranche” of medium-risk bonds is then sold on to other asset managers. To assess the immediacy of crisis risk, we would need to know more about the riskiness of this mezzanine tranche and who holds it. We don’t. If you are reminded of the roles of securitization, special purpose vehicles, tranching, and opacity in the subprime-mortgage crisis of 2007-08, you are not alone.

Just as debt markets fueled the global financial near-meltdown that followed from that crisis, their exposure to AI will drive developments now. The problem is that we possess only very limited public information about those markets, and regulators are not making that information any easier to obtain.

*Barry Eichengreen, Professor of Economics at the University of California, Berkeley, is the author, most recently, of In Defense of Public Debt (Oxford University Press, 2021). Project Syndicate, (c) 2026, published here with permission.