The AI selloff of mid-October 2026 came from the bond market, not from a sudden lack of buyers for AI chips. Lenders started charging much more to fund data centers, and stocks tied to that borrowing fell hard, even as the biggest chipmaker in the world said it still can’t make enough.
A Dow Jones Market Talk column on 20 October 2026 summed it up in its headline: credit concerns triggered the selloff, but demand remains strong. Here is what that means and why the gap between the two matters.
Key takeaways
- On 16 October 2026 the Nasdaq Composite fell 2.1% to 26,140, ending its worst week since June.
- CoreWeave, a “neocloud” that rents out AI computing power, lost about 24% that week as its bonds slid toward distressed levels.
- US 10-year Treasury yields held at 5.1%, near their highest level in more than two decades, which raised the cost of every dollar borrowed for AI.
- On 15 October TSMC forecast fourth-quarter revenue of $52.3 billion to $54.3 billion and said AI demand still exceeds its supply.
- The fight is over who pays for the AI build-out, and at what interest rate. Nobody is arguing that AI demand has gone away.
What triggered the AI selloff?
Three pressures hit at once. Inflation data came in hotter than expected, oil reached $110 a barrel, and Treasury yields pushed above 5%. The Federal Reserve had already raised its benchmark rate in September to a range of 3.75% to 4%, CBS News reported.
Higher rates hurt most where companies borrow the most. Adam Crisafulli of Vital Knowledge pointed to neoclouds such as CoreWeave, which fund much of their spending through debt and new shares rather than cash from operations. When borrowing gets expensive, that model gets fragile fast.
The damage spread quickly. US chipmakers shed nearly $2 trillion in market value in about a week, according to Reuters. On 16 October alone, Micron fell 5.6%, AMD 6% and Broadcom 4.5%.
Why did credit markets turn on AI?
Credit investors care about one thing: getting paid back. The bonds behind several flagship data center projects fell well below par (par is the 100 cents on the dollar a bond repays at maturity).
According to a roundup by EnterpriseAM drawing on Financial Times and Bloomberg reporting, these were the weak spots:
| Debt | Where it traded (mid-October 2026) | Starting point |
|---|---|---|
| CoreWeave unsecured bond | Below 80 cents, yield near 15% | Issued at par |
| Meta-anchored Hyperion project debt | Just under 92 cents | Above par in early July |
| Oracle Michigan campus debt | 87 cents | Par in July |
| QTS Project Excalibur (leased to Microsoft) | Below 88 cents | n/a |
Allianz chief investment officer Ludovic Subran told the FT that “data center bonds are having their Darwin moment, where the weakest credits struggle to survive.” It fits. The stress falls hardest on younger, heavily borrowed builders, while the cash-rich giants are under far less pressure.
Parth Talsania of Equisights Research called the episode “a rates event in credit clothing, set off by a hot CPI print.” Put plainly, the trigger was the price of money, not a shortage of customers.
If demand is strong, why did AI stocks fall?
Think of a restaurant with a queue around the block. The owner built three new branches on a credit card, and the card’s interest rate just jumped. The food is still selling. The question is whether the profits arrive before the bills do.
That is AI in October 2026. TSMC reported record third-quarter revenue of NT$1.494 trillion on 15 October, up 50.9% from a year earlier, and chief executive C.C. Wei said AI demand is still above what the company can supply. Kunjal Gala of Federated Hermes said robust AI demand continues to support the sector, and Bloomberg reporting found little evidence that financing pressure had cut chip orders.
Even so, a Goldman Sachs basket of Asian AI hardware stocks dropped 11% that week. Investors were pricing in the risk that expensive money could delay projects later. Current orders were not the worry.
Who is most exposed?
- Neoclouds. CoreWeave, Nebius and similar firms rent out GPUs and borrow heavily to buy them. Nebius chief executive Arkady Volozh said in September that his 5-gigawatt plan needs about $250 billion and “so far, we financed maybe 20% of that.”
- Project-finance lenders. Banks and private credit funds hold construction loans they now struggle to sell on.
- Suppliers further from the chip. Kevin Net of Financière de l’Échiquier warned that companies further from core chipmaking face more risk from project delays.
- AI labs heading to IPO. Reuters reported on 15 October that investors were discussing a $1.5 trillion valuation for Anthropic, down from a $2 trillion target.
The long-run maths is also under question. Gartner estimates AI data center owners need $8.5 trillion in revenue by the end of 2030 to cover costs and earn a healthy return, against $3.6 trillion in expected AI spending. If you want the chip-side view of that race, our piece on Amazon TorchNeuron and Nvidia’s CUDA moat shows how hyperscalers are trying to cut their dependence on one supplier.
Is this the AI bubble bursting?
Not yet, on the evidence so far. Bret Kenwell of eToro said CoreWeave’s slide “does not mean CoreWeave is bound for bankruptcy,” and Nvidia’s Jensen Huang said, according to Bloomberg, that CoreWeave’s “cost of capital is actually decreasing.” On 19 October CNBC reported that stock buyers were ready to move on from the rout while credit investors kept demanding higher yields.
My view, clearly opinion: this looks less like a bubble popping and more like a sorting. Strong balance sheets will keep building. Thinly financed players may have to sell assets, merge or slow down.
What happens next
- Fed meeting, 28-29 October 2026. Most traders expect no change, but CME FedWatch put the odds of a December hike at 74.3% as of 16 October.
- Big Tech earnings in late October. Capital spending guidance from the hyperscalers will show whether the build-out is slowing.
- Bond prices on project debt. If the Hyperion and Oracle-linked bonds climb back toward par, the panic is fading. Further drops would mean the stress is spreading.
- Energy prices. Oil at $110 feeds inflation and keeps rates high. Our look at Europe’s diesel reserves explains part of that squeeze.
FAQ
What caused the October 2026 AI selloff?
Rising borrowing costs. Hot inflation data, 10-year Treasury yields above 5% and a September Fed rate hike made the debt behind AI data centers more expensive, and heavily leveraged firms such as CoreWeave fell hardest.
Is demand for AI chips falling?
Not according to the latest data. On 15 October 2026 TSMC said AI demand still exceeds its supply and guided fourth-quarter revenue to $52.3 billion to $54.3 billion.
What is a neocloud?
A neocloud is a company that rents out GPU computing power for AI, like CoreWeave or Nebius. Unlike Amazon or Microsoft, it usually funds its hardware with large amounts of debt, which makes it sensitive to interest rates.
Your next step: when the hyperscalers report in late October, look at their capital spending guidance and at data center bond prices, not just the share moves. Those two numbers will tell you whether this was a scare or a turning point.
This article is for information only and is not investment advice.
4 comments