In Today's Issue: The sell-off week and what drove it, why financing matters more than valuations, the three templates tested against this year's numbers, who stands where if losses come, and the strongest case against calling this a bubble. Market figures as of October 16, 2026.

Dear Readers,

CoreWeave rents out Nvidia chips. It belongs to the so-called neoclouds, companies that fill data centres with graphics processors and lease the computing power to AI labs and big tech firms, and it has built that business largely with borrowed money. In the week of October 12, investors lost patience with it. The shares fell about a fifth between Monday and Wednesday (Bloomberg, 10/16/2026), and the yields on its bonds climbed above 15% (CNBC, 10/14/2026; Reuters, 10/15/2026).

By Thursday, credit default swaps, the contracts investors buy as insurance against a borrower failing to pay, priced a default probability of more than 60% within five years, against about 30% a year earlier (Reuters, 10/15/2026; Bloomberg, 10/16/2026). That probability is implied by market prices and rests on an assumption about how much creditors would recover, which neither report discloses.

CoreWeave is a small part of the AI economy, but it is the first major AI borrower in this cycle to be priced at distressed levels. That shifts the bubble debate from share prices to debt, and debt decides who pays when a boom ends. History offers three templates, each with a different bill. The Dotcom crash of 2000 wiped out shareholders and left the banking system largely intact. The telecom and fibre-optic bust of 2001 and 2002 fell on bondholders and on the equipment makers who had lent money to their own customers. The American railway crash of 1873 brought down the banks that had underwritten bonds they could not sell.

If the AI boom is a bubble, which of these three does it resemble most, and who would carry the losses if it burst: shareholders, bondholders, private credit funds, banks or insurers?

All the best,

Kim Isenberg

Wednesday was the worst day. The shares fell 15.46% (Fortune, 10/16/2026), the unsecured bonds due 2032, sold in June with a 9.625% coupon, dropped below 80 cents on the dollar (Bloomberg data, via Reuters, 10/14/2026), and by Thursday the default swaps cost around 1,000 basis points a year, while the yield on the notes due 2031 rose above 16% (Reuters, 10/15/2026).

Three explanations circulated, and the calendar supports each of them in part. One is interest rates. On Wednesday, US inflation for September came in at 3.7% against 3.6% expected, and the ten-year Treasury yield touched 5.3%, its highest level since 2002 (Trading Economics, 10/14/2026). CoreWeave funds its data centres with bonds, so rising yields raise its costs quickly, which was Bloomberg's reading of the slide. Thomas DiFazio of the fund firm Roundhill described the mood: "The macro backdrop has created a sell-first, ask-questions-later market" (Bloomberg, 10/16/2026).

Another is demand. A week earlier the Financial Times had reported OpenAI's annualised revenue at about 50 billion dollars at the end of September, against roughly 70 billion previously reported; according to a CNBC source cited by AFP, the earlier figure had counted partner revenue gross (FT, as reported by AFP, 10/09/2026). On the same Wednesday as the inflation data, the research firm Gartner forecast that more than 3.5 trillion dollars of AI infrastructure spending could face write-downs by 2030 (Gartner, 10/14/2026). Then there is contagion. In a memo to staff, CoreWeave chief executive Michael Intrator wrote that rivals' IPO and financing troubles were "bleeding into the narrative" and that the company's order backlog "more than supports all of our debt obligations" (Reuters, 10/14/2026).

The timing separates these channels only partly. Chip stocks started falling on the day of the OpenAI report, six days before the inflation figure, which points to demand worries in semiconductors. CoreWeave's steepest fall came on inflation day, which was also the day of Gartner's forecast and the Intrator memo, so for CoreWeave the effects of rates and demand cannot be pulled apart. The selling ran along the supply chain: on Wednesday Oracle fell 8.8% and Dell about 11.3% (SiliconANGLE, 10/14/2026), the Philadelphia semiconductor index lost 11.6% over six sessions, about 1.6 trillion dollars of market value (Reuters, 10/15/2026), and Dell and Hewlett Packard Enterprise ended the week down about 20% (Bloomberg, 10/16/2026).

On Thursday afternoon Nvidia's chief executive Jensen Huang, whose company supplies CoreWeave's chips and owns a stake in it, spoke up in Stockholm: "CoreWeave's strength is improving every day. I'm optimistic. Their cost of capital is actually decreasing." He said there was no need for a bailout, and the shares closed 5.9% higher (Bloomberg, 10/16/2026).

The strain had been building for months. In July S&P cut Oracle's credit rating to BBB-, one notch above junk (FactSet, 07/23/2026). Since April, the asset manager Blue Owl has limited withdrawals from private credit funds whose investors wanted their money back (CNBC, 04/02/2026). In September, according to the Financial Times, about 18 billion dollars of bank loans for Project Jupiter, a giant Oracle data centre campus in New Mexico, were quoted at 89 to 91 cents on the dollar (FT, as reported by Reuters, 09/18/2026). What October added was the first distressed-level pricing of a major AI borrower in this cycle. A Deutsche Bank chart published that week shows where CoreWeave now sits in the bond market.