Fed's Schmid Asks Whether the AI Ecosystem Is Becoming Too Big to Fail

For most of the AI boom, central bankers have described artificial intelligence as a productivity story, an inflation story, or an asset-price story. On Thursday, Kansas City Fed President Jeff Schmid gave it another label, one with painful associations for the Federal Reserve: a possible too-big-to-fail problem.

"Where we have to start to really synthesize what's happening in the AI and the data center build-out is are we moving to a too-big-to-fail AI ecosystem," Schmid said on September 25, according to Reuters. He was invoking the public bailouts of major financial institutions during the 2007 to 2009 financial crisis, when banks had grown so large and so entangled with the rest of the economy that letting them collapse was judged too dangerous.

Schmid did not say the AI sector has reached that point. His concern was more basic: that policymakers cannot yet see clearly enough inside the web of companies and contracts to know. "You worry a little bit about how do we understand what's inside. ... Is there anything systemic?" he asked.

A web of contracts, not a single bank

The comparison matters because the AI industry's exposures do not look like a traditional banking system. The largest build-out commitments run through long-term compute and cloud contracts, chip supply deals, equity stakes and data-center leases. Arrangements linking OpenAI, Oracle, Nvidia and SoftBank have drawn particular scrutiny because money often flows in circles: a supplier invests in a customer that then commits to buy the supplier's products, while the physical infrastructure is financed with borrowed money. If one link weakens, the question is how many others weaken with it.

That is the gap Schmid appeared to be pointing at. Reuters did not report him naming any company, but his framing, a "network of firms and contracts," maps closely onto the structure analysts have been flagging all year. Coverage of his remarks by The Economic Times noted that AI's linkages now reach well beyond technology companies into semiconductor makers, cloud providers, data-center operators, utilities, construction, real estate and lenders.

Schmid also pointed to AI as a source of demand the Fed is still trying to size. "There's a lot of data out there that says the technology boom, the AI element is creating demand, especially in things like commodity level prices," he said, in comments reported alongside his financial-stability remarks.

The Fed has been building toward this

Schmid's framing is unusually blunt, but he is not the first Fed official to raise AI-linked leverage. In a May 27 speech at Stanford, Fed Governor Lisa Cook noted that companies had announced "more than $1.5 trillion in data-center plans," only a small portion of which had been realized. She warned that hyperscalers were issuing large investment-grade bonds while smaller developers were tapping private debt funds and asset-backed credit markets, and concluded that "a sustained boom in debt issuance could eventually represent a financial-stability concern." Cook added that even under ambitious projections, leverage was unlikely to return to pre-2008 peaks.

The Chicago Fed has tried to measure how much of that risk sits on bank balance sheets. Its researchers estimated that the average bank's outstanding exposure to AI-adjacent industries is around 0.8% of total assets, and that delinquency rates on those loans are in line with overall portfolios. But they flagged "notable tail risk": about 26% of large-bank commercial and industrial commitments to software companies, roughly $50 billion, are rated B or below. Many of those borrowers are unprofitable and rely on investors to fund their AI spending, meaning a pullback could ripple into chipmakers, energy firms and data centers at once.

Why It Matters

"Too big to fail" is not casual language at the Fed. The phrase carries the memory of the Lehman Brothers collapse and the emergency rescues that followed, and it implies that private risk-taking could end with public costs. When a sitting Reserve Bank president applies it to AI, even as a question, it signals that the debate inside the central bank is shifting from whether AI valuations are frothy to whether the industry's financial plumbing could transmit a shock to the broader economy.

The timing is notable. This month alone, Reserve Bank of Australia Governor Michele Bullock weighed in on AI bubble risk, SoftBank delayed the IPO of its SB Energy unit, and Brookings put the total AI build-out price tag at an estimated $10.3 trillion. Numbers of that size, financed increasingly with debt and bound together by multiyear contracts, are exactly the kind of concentration regulators failed to map before 2008.

The harder issue is jurisdiction. The Fed supervises banks, not AI labs, chip designers or data-center developers. Much of the new financing is flowing through private credit and off-balance-sheet structures that sit outside traditional supervision. Schmid's "what's inside" question is partly an admission that the Fed's current tools were not built to see this kind of risk.

What to Watch

The next test is whether Schmid's question turns into formal work. Watch for AI-linked leverage to feature prominently in the Fed's next Financial Stability Report, for other FOMC members to echo or push back on the too-big-to-fail framing in the coming weeks, and for any move by the Fed or the Financial Stability Oversight Council to collect data on private-credit and bond financing tied to data centers. On the market side, the pace of AI-related debt issuance, the terms on the next round of mega compute contracts, and whether delayed listings like SB Energy's come back will show whether investors share Schmid's unease.