Warning Signs: Frédéric Lordon on the Financial Risks of the AI Boom

Summary

Frédéric Lordon argues that the AI infrastructure boom is being sustained by financial commitments whose scale may exceed the underlying businesses’ ability to pay. He focuses on a $500 billion arrangement announced by NVIDIA and six financial institutions, noting that the announcement consisted of memorandums of understanding rather than a completed investment plan. Lordon says OpenAI and Anthropic have promised hundreds of billions of dollars in future payments to hyperscalers through take-or-pay contracts, while their reported revenues and losses remain opaque because they are private companies. The contracts create stepped obligations that rise sharply from 2027 onward, when payment is required even if computing capacity is no longer needed. At the same time, hyperscalers have spent heavily on data centers and are turning to borrowing, using projected future orders as evidence of solvency. Lordon portrays the model as dependent on two expectations: rapid revenue growth at the AI labs and continued investor support for ever-higher valuations. He compares this dynamic with a financial bicycle that must accelerate to remain upright, arguing that a slowdown in valuation growth could expose the system’s weaknesses. The article also describes synthetic risk transfers, special-purpose vehicles, private credit, guarantees and securitization as mechanisms that move AI-related exposure away from traditional balance sheets and into less transparent parts of finance. It cites Goldman Sachs’s estimate that $1 trillion of $1.5 trillion in identified AI-related IOUs is not recorded in financial statements. Lordon warns that pension and mutual-fund savings may increasingly be exposed to these structures, and notes the US regulator’s authorization of hyperscaler-debt securitization. His conclusion is an argument, not a forecasted fact: the AI boom may be entering a latency period in which confidence remains intact while signs of financial fragility multiply.