Could the AI Boom Become Too Big to Fail?

Kansas City Federal Reserve President Jeff Schmid’s concern is not simply that an AI company might lose money. It is that the firms, contracts and financing behind the AI buildout could become interconnected enough for trouble in one part of the network to spread. On September 25, 2026, he asked whether the industry was moving toward a “too-big-to-fail AI ecosystem,” invoking the public bailouts of major financial institutions during the 2007–2009 crisis.1

Why the network matters

Schmid had raised the issue in August, saying the scale and finances of AI development warranted attention as a potential systemic problem.2 The distinction is between an investment loss that its owners can absorb and losses that reach lenders, insurers or other businesses through shared obligations.

One possible route is data-center project finance: if expected demand fails to produce enough revenue, a project could struggle to repay debt backed by its cash flows. If multiple projects depend on similar assumptions, losses could affect several financiers at once. That is a risk scenario, not evidence that such a chain reaction has begun. Reuters reported, citing Nikkei Asia, that Nippon Life planned 2 trillion yen ($12.75 billion) in infrastructure financing including U.S. data-center construction; the figure is not an amount committed exclusively to AI data centers.17

The financial-crisis comparison therefore points to a question for supervisors: can they see how exposures connect across companies and institutions? It does not establish that today’s AI financing resembles the pre-2008 financial system in scale or fragility. Reuters reported in August that some Fed officials favored vigilance while doubting that a housing-crisis-style collapse was imminent.3

What policymakers are doing

Japan’s Financial Services Agency is increasing scrutiny of AI data-center financing by major banks and life insurers, including their risk-management practices for projects mainly in the United States, according to a September report.23 Its stated policy priorities also include examining data-center and other project-finance exposures alongside real-estate lending risks.20 That is a response to potential financial losses, not a finding that those loans are failing.

In the United States, a bipartisan group of 26 attorneys general led by New York’s Letitia James has urged Congress to establish comprehensive federal regulation and safety protocols for frontier AI.3349 Their focus is principally on risks from advanced AI systems, rather than the repayment of data-center loans. The two lines of oversight are complementary: safer models would not make every infrastructure investment sound, and stronger lending practices would not resolve every model-safety concern.

Schmid’s warning leaves the central question open. Before calling the AI ecosystem too big to fail, policymakers need to understand who owes what to whom—and whether a setback could travel beyond the companies that took the original risk.12