The AI Loop Meets the Lenders: What the Past Week Says About Who Pays
In The Ouroboros Economy, I argued that the AI buildout is not a snake eating its own tail in a sealed loop. Outside money is flowing in, and the biggest question is who supplies it. Last week gave a live test. The people who supply it, the bond buyers and banks, started asking harder questions.
What happened
Oracle went back to the market. On November 2, Oracle priced a $20 billion package: about 86.2 million new shares at $116 each, plus $10 billion of senior notes. The notes carry coupons from 7.00% on debt due 2029 to 8.65% on debt due 2046. Oracle’s stock fell about 8.8% that day. The raise followed Oracle’s own statement that it would cap new capital-markets funding at $20 billion through fiscal 2030, so it was expected. What stands out is the price: a company funding a huge AI buildout was paying high-single-digit rates and selling shares at a discount to do it.
The stock market flinched. Reuters reported that the Nasdaq 100 lost about $2.5 trillion in market value on Thursday, November 5, the same day my paper was dated.
Then the lenders did. On November 9, Reuters reported that some banks and credit investors are showing reluctance to take on more AI-related debt, even when large technology companies guarantee it. A law firm partner quoted by Reuters described the past year as one of “unbridled enthusiasm” for lending to data center builders and said debt investors’ books have become full. Reuters tied the shift to the selloff and to the midterm election results, which it said are raising questions about possible oversight of data centers. The bankers quoted did not say they were leaving the sector. They said they would wait and see.
The guarantees came under a spotlight. The same day, Aletheia Capital cut Nvidia from Buy to Hold and withdrew its price target, estimating that Nvidia’s financial commitments to partners in its AI Computing Partner program have grown more than tenfold, to $165 billion, with about $105 billion tied to a single Ohio data center project. This is one analyst’s estimate, and most analysts still rate the stock favorably, but it points at the exact mechanism in my paper: a chip supplier standing behind the borrowing of its own customers.
A big customer relationship wobbled. Reuters, citing Bloomberg, reported on November 6 that Microsoft is letting an arrangement to rent AI computing capacity from Oracle expire rather than renew it. Oracle responded that demand for its AI cloud remains strong. Separately, Bernstein estimated that OpenAI’s identified infrastructure contracts through 2030 total about $614 billion, below the roughly $750 billion figure often repeated, with about $300 billion of it tied to Oracle. That is a smaller number than the headlines, and still an enormous one.
Why this matters for the loop
My paper’s main point was that the weak link is the gap between what AI labs have promised to spend and what they earn, and that the risk lands on whoever lent against those promises. Debt is the pipe through which that risk travels. Goldman Sachs Research projects global hyperscaler debt issuance of $420 billion in 2027, more than 60% above its $250 billion estimate for 2026, according to Reuters. If lenders become pickier while borrowing needs keep rising, one of two things has to give: the price of the debt, or the pace of the buildout.
That spread detail is worth a pause. A wider spread means lenders are demanding more extra yield to hold hyperscaler debt than they did, and more than they demand from comparable non-AI companies. It is an early sign of repricing, not a crisis.
How this reaches ordinary households
The paper’s second key judgment was that stress can move into credit markets and into pension, insurance, and bank portfolios. That is the part of this story that reaches people who never buy a share of an AI company. Three routes are worth knowing.
- Stock index funds. The big AI spenders are among the largest holdings in the broad index funds that most 401(k) and IRA investors own. A drop in those shares shows up in retirement balances without the investor choosing an AI bet.
- Bond funds and insurers. The paper notes that pension funds, insurers, and banks are among the buyers of this debt. Hyperscaler bonds can sit inside investment-grade bond funds and insurance portfolios.
- Credit conditions. If lenders pull back from one sector, they can tighten elsewhere. That is more speculative, and it is the pathway I will watch rather than assume.
What this does not show
One bad week is not a trend, and the evidence has real limits. The Reuters report relies on unnamed bankers, and it says plainly that banks have not abandoned AI lending. Aletheia’s $165 billion is an estimate, not a company disclosure. The Microsoft and Oracle report is a single relayed report that Oracle disputes. Demand for AI services is also real and growing, and the strongest companies still generate large cash flows. The paper’s “growth catches up” scenario has not been ruled out by any of this.
My assessment in MM-2026-003 was a roughly 25 to 35 percent chance, over 24 months, of an AI-linked repricing severe enough to force contract cancellations or restructurings and noticeably tighten credit. Last week’s events are consistent with the early stages of that path, and also with an ordinary wobble. They are not enough, on their own, to change the estimate. I will revise it if the indicators below move.
What to watch next
- Nvidia’s earnings on November 17. The questions are customer financing, guarantees, and any change to demand forecasts, not just the revenue number.
- Oracle’s credit pricing and its OpenAI-linked backlog. Whether its borrowing costs keep rising, and whether the Microsoft report is confirmed.
- The next big hyperscaler bond deal. How much extra yield buyers demand, and whether deals get done on schedule.
- OpenAI’s revenue against its targets. This remains the clearest test of the gap at the center of the paper.
Three questions for your own money
This is general information, not personal financial advice.
- What are the top holdings in the funds you own? Fund companies publish them. Many people are more exposed to a handful of AI-linked companies than they realize.
- What is in your bond or target-date fund? A fund’s fact sheet lists its largest issuers and credit quality. Know whether corporate bonds from large tech borrowers are part of it.
- Does your time horizon match your exposure? Money you need within a few years is a different problem from money you will not touch for decades. A diversified plan set before a selloff will usually serve better than decisions made during one.
Sources
The research paper, MM-2026-003, has the complete reference list. Key sources for this memo:
- Oracle Corporation, “Oracle Announces Pricing of $20 Billion of Common Stock and Senior Notes,” November 2, 2026 (investor.oracle.com and PR Newswire)
- Reuters, “Banks, investors grow wary of AI debt, spooked by market selloff,” November 9, 2026 (read through syndicated copies and summaries; the Nasdaq 100, Goldman Sachs, Bank of America, and law firm figures and quotations are as reported there)
- Reuters, “Microsoft is winding down Oracle deal to rent AI cloud capacity,” November 6, 2026, citing Bloomberg (as summarized by secondary sources; I did not read the original)
- Bernstein estimate of OpenAI infrastructure obligations, as reported by secondary sources, November 8 and 9, 2026
- Aletheia Capital downgrade of Nvidia, November 9, 2026, as reported by Investing.com and others; Nvidia’s scheduled earnings date of November 17 from the same reporting
- NVIDIA Corporation, Form 10-Q for the quarter ended July 26, 2026, risk factors on guarantees and commercial arrangements (SEC)
Cite this article
For information and education only. Not investment, tax, or legal advice.