AI's Trillion-Dollar Loop, and the Last Time It Happened

Broadcom told investors that one of its customers may borrow up to $42 billion from it, and that the money can only be used to lease Broadcom's chips. Reuters says the customer is Anthropic.

It isn't the only deal like this. Nvidia has guaranteed up to $105 billion of lease and power payments for an OpenAI data center campus in Ohio. It has promised to rent back $36 billion of computing capacity from cloud companies if they can't sell it to anyone else. AMD gave OpenAI and Meta the right to buy up to 160 million AMD shares each for one cent a share, unlocked as they buy AMD chips. Oracle reports $664 billion of contracted future revenue and $125 billion of borrowings.

Telecom equipment makers did a version of this in 1999 and 2000. By September 2000 Lucent had promised its customers $8.1 billion in loans and guarantees. Nortel had promised $5.2 billion by December. When the start-up phone companies ran out of money, Lucent set aside reserves of $2.1 billion against the $2.96 billion its customers had drawn.

We measured the same thing in each company's own filings: the support a seller promised its customers, divided by a year of its revenue. Lucent in 2000 was 28 percent. Nortel was about 18 percent. Nvidia today is 54 percent, counting guarantees, capacity it promised to rent back and data center leases it took on for others. Adding its equity stakes takes it to 95 percent, but not every company in that number is a customer.

This time could end differently. Nvidia's revenue is about ten times Lucent's, many of its buyers are profitable, and most of its guarantees only start in 2028 or later. The Ohio guarantee ends if OpenAI earns a strong enough credit rating.

Chapters:
0:00 The strange promise
0:43 How the loop works
1:28 Nvidia
2:20 AMD, Oracle, CoreWeave
3:41 1999 and 2000
4:48 How it broke
6:01 What we measured
6:55 Why this time could be different
7:55 What to watch

Sources:
Broadcom 10-Q, Aug 2, 2026:
https://www.sec.gov/Archives/e....dgar/data/1730168/00
Reuters via CNBC, Oct 1, 2026:
https://www.cnbc.com/2026/10/0....1/broadcom-lending-a
Nvidia 10-Q, July 26, 2026:
https://www.sec.gov/Archives/e....dgar/data/1045810/00
AMD 10-Q, June 27, 2026:
https://www.sec.gov/Archives/e....dgar/data/2488/00000
Oracle 10-Q, Aug 31, 2026, and 10-K, fiscal 2026:
https://www.sec.gov/Archives/e....dgar/data/1341439/00
https://www.sec.gov/Archives/e....dgar/data/1341439/00
CoreWeave 10-Q, June 30, 2026:
https://www.sec.gov/Archives/e....dgar/data/1769628/00
Lucent 10-K fiscal 2000 and 2001, 10-Q June 2001:
https://www.sec.gov/Archives/e....dgar/data/1006240/00
https://www.sec.gov/Archives/e....dgar/data/1006240/00
https://www.sec.gov/Archives/e....dgar/data/1006240/00
Nortel 10-K 2000 and 2001:
https://www.sec.gov/Archives/e....dgar/data/72911/0000
https://www.sec.gov/Archives/e....dgar/data/72911/0001
Federal Reserve, Chairman Warsh press conference, Sept 16, 2026:
https://www.federalreserve.gov..../mediacenter/files/F
The Information, on Nvidia's investment in OpenAI and OpenAI's first quarter results (via Yahoo Finance):
https://finance.yahoo.com/tech....nology/ai/articles/o
Share prices for Lucent and Nortel are market data as reported at the time.

Photos from Wikimedia Commons:
"Jensen Huang (cropped) (2024)" by Peter Dasilva, CC BY 4.0
"AMD CEO Lisa Su speaking at ORNL 2019-05-07" by Genevieve Martin, ORNL, CC BY 2.0
"Larry Ellison on stage" by Oracle Corporate Communications, CC BY 2.0
"Sam Altman, June 2023 (GPOABG244) (cropped)" by Amos Ben Gershom, CC BY-SA 3.0
"Broadcom Headquarters San Jose" by Coolcaesar, CC BY-SA 4.0
"Lucent HQ" by Blaxthos, CC BY-SA 3.0
"Nortel Carling Campus" by Iouri Goussev, CC BY-SA 2.0
"TPU v4" by Norman P. Jouppi and others, CC BY 4.0
Video of Chairman Warsh: Federal Reserve, U.S. government work. Stock video from Pexels (Nicola Narracci, Tom Fisk, cottonbro studio, Alireza Akhlaghi, ALL IZ Well, David Pickup, Pixabay). The loop maps and charts are our illustrations. Music and sound: Apple Loops and our own sound design.

#Nvidia #AIBubble #BusinessHistory

VREDBOR tells the stories of companies, markets and systems that became too big to question, and what happened when the numbers stopped working.

Business. Finance. Power. Failure.