The September number, read carefully
Two numbers landed this week. The first is the dramatic one: September issuance of $23 billion, roughly a fifth of June's $113 billion. A fall of about 80% in three months reads like a door slamming. The second is the one that keeps the first honest: $466 billion raised since January, nearly five times the comparable 2025 figure. Both are Morgan Stanley's, both came through the FT, and they describe the same market. The story is not that AI ran out of credit. It is that the first half borrowed so much the third quarter had nothing left to do.
What the bank actually said
Morgan Stanley did not call it a funding crisis, and anyone writing this story should not either. Its read, as reported: the monthly decline was mostly the arithmetic of front-loading — companies pulled their borrowing forward, and September paid the price. Investor scrutiny of returns got a supporting role. That is a lender saying "we already lent" rather than "we will not lend." The distinction matters because it sets the condition for what comes next: the borrowing resumes when companies can show what the first $466 billion is buying.
The number nobody wanted next to it
JPMorgan's separate finding is the one that will travel furthest with the least context, so put it in context here. Deeply distressed US loans hit $65 billion, up from $40 billion a year ago, the highest since March 2020; loans priced at 80 cents or less approach $140 billion. Tech firms hold about 39% of the distressed total. What the data does not do — and its own write-up says this plainly — is link any of that distress to AI borrowing. Treat it as what it is: a reminder that the credit market has a stress channel, running parallel to the AI boom, not necessarily through it.
What actually changes
The practical effect is a narrower financing window, and it lands unevenly. Hyperscalers with cash flow and diversified revenue can wait out a quarterly dip; they were the ones front-loading anyway. The squeeze falls on the marginal projects — neoclouds, speculative data-center builds, infrastructure bets with no customer yet signed. For them, "investor scrutiny" stops being an analyst phrase and becomes a term sheet. The question that replaced "how much can you raise" is "when does it pay back," and the honest answer in most AI infrastructure decks is still a projection.
One more number for the file: Goldman Sachs has hyperscaler capex at $600–820 billion this year, potentially over $1 trillion in 2027. The debt slowdown does not touch that spend directly — capex plans are bigger than the bond market. But the bond market is where the industry proved it could fund anything. September was the first month the market asked for a reason. How fast lending recovers will tell you more about AI's economics than any model release this quarter.
Sources
- [1] AIImpactLab — “Nvidia eyes a deeper Reflection AI deal as AI borrowing cools” (Oct 11, 2026), citing FT / Morgan StanleyRead source
- [2] Crypto Briefing — “AI companies face borrowing slowdown amid $466B debt concerns” (Oct 11, 2026)Read source
- [3] Cointime — “AI Corporate Lending Shrinks Dramatically, Down $90 Billion in Three Months”Read source
- [4] Unbiased Headlines — “Deeply Distressed U.S. Loans Hit $65 Billion...” (JPMorgan data)Read source
- [5] SignalPlus — “AI Debt Boom Tests Big Tech's Credit Appetite” (Goldman Sachs capex figures)Read source
- [6] Daily Press — “The AI Debt Slowdown: Why Wall Street Is Becoming Cautious About Big Tech Borrowing in 2026”Read source