Banks are pulling back as AI's massive debt pile disconnects from economic reality

Wall Street appears to be hitting its limit on AI, and financiers are starting to say the quiet part out loud: All of this AI spending doesn't look like it can pay off, at least not for everyone. Why it matters: The gap between what it costs to build AI infrastructure and what companies are actually paying to use it is widening. Wall Street's anxiety is surging as the debt piling up to fund data centers and chips looks increasingly disconnected from economic reality. Driving the news: A massive gap is opening up between AI infrastructure spending and expected returns. While AI providers are projected to spend $7 trillion on infrastructure through 2030, research from Gartner forecasts they will generate no more than $3.6 trillion in related revenue over that same period. More than $4 trillion in AI infrastructure spending could be at risk of write-downs by the end of 2032 if companies maintain their current investment pace. Gartner also predicts that more than 80% of specialized AI cloud providers, known as neoclouds, will cease operating independently by 2030. Case in point: CoreWeave, one of the largest neoclouds with among the closest ties to Nvidia — the AI boom's financial and spiritual foundation — has faced a multi-day selloff of its stock and debt. Yields pushed above 15% and its stock tumbled 15.5% before rebounding after Nvidia CEO Jensen Huang spoke up on the company's behalf. Follow the money: Sources on Wall Street used to tell Axios that as long as the AI buildout was paid for in cash, there was nothing to worry about. But today, a third of the capex being spent on AI is coming from debt, and capital markets are showing signs of severe fatigue. It's getting harder to secure financing. Global AI-related debt issuance plunged to $23 billion in September 2026, down sharply from $113 billion in June, per the Financial Times. Banks are actively pulling back to avoid concentration risk, with JPMorgan CEO Jamie Dimon confirming his bank is tightening up on credit. Infrastructure projects and providers are struggling: approximately $18 billion in bank loans for an Oracle data center campus recently slid into stressed territory, while Nvidia-backed AI cloud provider Firmus shelved an IPO that sought a valuation north of $30 billion. Zoom in: The constant spending relies on the belief that compute demand will never falter — a belief that has structural cracks. Reflection AI co-founder Ioannis Antonoglou told Semafor that businesses may need "a decade or two decades" to fundamentally redesign their operations around AI. That's 10 to 20 years of companies slowly adopting the technology that's being built (expensively) today. Between the lines: This creates a mismatch: Data centers and chips require enormous investment today, but the returns that could justify that spend will take years to materialize. Put bluntly, there is a massive gap between AI spending and AI revenue, one that won't close for many, many years. At the same time, companies that are adopting AI quickly are getting smarter about using tools more efficiently, turning to open-weight models that are less compute-intensive to cut costs. Reality check: Look at Anthropic's latest financials for a counterpoint to the bears. The company is raking in cash, turning free cash flow positive in the third quarter and growing revenue nearly 50% compared to the April-to-June period. But even Anthropic's staggering growth has slowed, albeit expectedly because trees can't grow to the sky. The bottom line: The debt-fueled AI buildout is not an isolated problem for tech investors alone to worry about. The stock market and economy have become increasingly tied up with the AI boom. If the AI bet doesn't work, markets and the economy could falter as well.

Read it at Axios ↗

← Front page