AI Debt Slowdown: September Borrowing Fell Off a Cliff

The AI boom has a new problem: lenders are getting nervous. Global borrowing tied to artificial intelligence infrastructure fell sharply in September, with AI debt issuance totaling about $23 billion for the month — less than half of August's level, according to the Financial Times, citing data from Morgan Stanley. This AI debt pullback arrives just as Wall Street was treating AI financing as a bottomless well.

The pullback reflects a broader reassessment of the spending spree behind the AI buildout. Investors are scrutinizing borrowing costs, the viability of massive data-center projects, and whether the customers signing up for AI capacity can sustain demand. Developing AI infrastructure requires enormous upfront investment, while the timing of future returns remains uncertain — and lenders are no longer willing to fund that gap on faith alone.

None of this means AI demand is collapsing. A dip in new AI debt can reflect earlier heavy borrowing — companies that already loaded up on bonds earlier in the year simply have less reason to come back to market — alongside plain caution among lenders. But after a year in which the AI infrastructure trade felt unstoppable, the bond market is sending an early warning about the price of money.

Why lenders are asking harder questions

The AI debt market is suddenly behaving like a market that has seen this movie before. Credit has been repricing AI risk for months: bond spreads for major technology borrowers widened through September, and investor appetite for hyperscaler debt has cooled — one market analysis notes that demand at big tech bond offerings has thinned dramatically since the spring, as reported by StartupFortune. Oracle, one of the most aggressive AI spenders, has become a poster child for the squeeze: S&P downgraded the company to BBB-, one notch above junk, citing the pace of its AI spending, and its shares have fallen sharply since June, according to IBTimes reporting on the credit fears.

The financing structures are also getting more creative — and more opaque. Tens of billions in AI debt have moved into off-balance-sheet vehicles, special-purpose entities that keep the debt away from the parent company's books. Analysts have compared the concentration of AI-linked issuance in the high-yield market to telecom bonds in 1999, a comparison that makes bond veterans reach for antacids. Add in a Federal Reserve rate hike in September — its first in three years — and the cost of capital is moving in the wrong direction for a capital-hungry industry.

What this AI debt moment actually means

The counterargument is real and worth taking seriously: unlike the telecom companies of 1999, today's hyperscalers generate enormous free cash flow. The question is whether that cash flow covers the ambition — capital spending across the big AI players is expected to surge by more than four-fifths this year, which would push aggregate free cash flow close to zero. The bond market doesn't need a crash to send a message; it just needs to keep raising the price of money.

That distinction matters for where this goes next. Differentiation between borrowers is becoming the story: companies with strong earnings coverage are still getting funded, while the most leveraged spenders face tougher terms. For investors who have been buying AI exposure through tech stocks, the AI debt market's skepticism is worth watching — bond investors get paid first, and they are the first to flinch.

The week ahead will test that caution. With major bank earnings and fresh inflation data on the calendar, every new data-center financing will be read as a verdict on the whole trade. The analysts watching this market expect issuance to continue — but at wider spreads and with tougher covenants, as lenders price in the reality that AI's buildout phase is the most capital-intensive part of the story.

For now, the September dip looks like a pause rather than a verdict. Some analysts argue that rising rates won't derail borrower demand and that large issuance will continue. But the direction of travel is clear: lenders are no longer rubber-stamping every AI bond. For an industry that has been building the future on AI debt, the era of easy financing is over — and what comes next depends on whether the data centers fill up as fast as the pitch decks promised.

Related reading: how tech billionaires' wealth soared in 2026's AI boom, TypeSafe AI's $870M bet on the future of AI, and the Ghost Core personal AI computer.