Fed's AI focus signals growing concern over financial stability and looming debt risks

Federal Reserve chair Kevin Warsh’s decision to launch a task force on artificial intelligence marks an acknowledgement that the technology is no longer just a productivity story, but a policy problem. Yet the Fed’s stated focus on employment and inflation still leaves out a risk that many officials have already begun to flag: financial stability.

In April, his predecessor Jerome Powell joined US Treasury Secretary Scott Bessent in a meeting aimed at assessing how advanced AI models could affect cybersecurity in the banking system. More recently, the Fed’s May 2026 Financial Stability Report listed AI among the near-term risks facing the US financial system, alongside cyberattacks and geopolitical tensions. The report said respondents were citing AI-related concerns more often than in earlier surveys, while market participants also pointed to private credit, persistent inflation and higher-for-longer interest rates.

That broader caution is well placed. The central bank is trying to understand an economy in which AI is altering both supply and demand, while also distorting the data policymakers rely on. The Bank for International Settlements has warned that the boom is already affecting investment patterns in semiconductors and data-centre components, complicating estimates of the natural rate of interest, unemployment and other unobservable variables that help shape monetary policy. Axios reported that the Fed is now forming task forces to study AI’s effects on labour markets, productivity, inflation measurement and data collection.

The more immediate concern, however, is not simply whether AI proves transformative, but whether the money being poured into it can ever be repaid. As Brian Judge writes in Project Syndicate, the US financial system and asset markets have become heavily exposed to the AI buildout. He argues that outstanding AI data-centre debt could exceed mortgage debt by the end of the decade if current trends persist, raising a basic question about whether the future cash flows from those assets will be enough to service creditors.

That distinction matters. History shows that a technology can be genuinely revolutionary and still produce destructive financial excess. Railways transformed the 19th century even as overinvestment helped trigger the Panic of 1873. Fibre-optic cable and the dotcom economy were equally real innovations, but they still ended in severe losses for lenders and investors. The key issue is not whether AI works; it is whether the pace of investment is running far ahead of plausible returns.

The arithmetic is sobering. Judge cites Sequoia’s David Cahn as estimating that this year’s roughly US$750 billion in hyperscaler AI capital spending would need to generate about US$1.5 trillion in end-user revenue over the life of the equipment just to break even. Bain & Co has separately calculated that meeting expected AI demand by 2030 will require US$2 trillion in fresh annual revenue. Even the profits of the sector’s leading names remain far below that scale, with Anthropic’s annualised revenue rumoured at around US$60 billion.

Funding for the boom is also shifting. According to Judge, the tech giants’ own cash flow is no longer enough to support the pace of construction, so capital markets are taking on a larger role. That is creating circular financing patterns, in which chipmakers back AI labs, cloud providers support start-ups that rent their capacity, and valuations rise alongside capital expenditure. Nvidia has become a key source of support for so-called neoclouds, while private credit funds are increasingly lending against projects linked to their own sponsors.

KPMG’s summary of the Fed’s Spring 2026 reports suggests officials are already uneasy about this broader mix of leverage and valuation pressure. The report described elevated asset prices, vulnerabilities in business and household borrowing, and risks tied to financial leverage and funding. It also noted that market participants are increasingly naming AI and private credit among the most salient threats.

For now, the biggest exposure appears to be in credit rather than equities. Judge argues that modern finance is no longer dominated by banks alone, but by bond markets, securitisation structures and non-bank lenders. That makes the relevant stress scenario less like the banking panics of the 1930s and more like a shadow-banking run: doubts about the quality of the underlying credit force lenders to pull back, short-term funding dries up, and asset sales feed a downward spiral.

The Fed has also heard warnings from within its own ranks. In a May speech, Governor Lisa Cook said AI-driven algorithmic trading could increase correlated behaviour, model collusion, manipulation and market concentration. She also noted that companies are increasingly using debt to fund AI infrastructure, which could create new vulnerabilities if the sector fails to deliver the returns investors expect.

There is, of course, another possibility: that the investment wave is justified, the revenues arrive and the debt is comfortably serviced. But even that outcome would bring its own upheaval. Michael Barr, another Fed governor, has outlined scenarios ranging from gradual labour adjustment to severe disruption, including the possibility that rapid AI adoption could leave large parts of the workforce struggling to adapt.

That is why the Fed’s AI agenda cannot stop at inflation or jobs. If the boom falters, the central bank may be forced to confront a credit event; if it succeeds, the economic dislocation could be even more profound. Either way, financial stability belongs at the centre of the debate.