YouTube2Text

The Smart People Have Started Looking for the Exit — Transcript

by Casual Finance · 3,273 words · 495 segments · language en · Watch on YouTube

Full transcript

  1. 0:00Last month, the Bank for International
  2. 0:01Settlements, which is basically the
  3. 0:03central bank for central banks,
  4. 0:05published a study of more than a
  5. 0:06thousand AI companies. The study focuses
  6. 0:09on one question. Using new data on AI
  7. 0:11firms, investment deals, and supply
  8. 0:13chain linkages, the study examined
  9. 0:15investment patterns among the largest AI
  10. 0:17firms globally, focusing on circular
  11. 0:19investment relationships. What they
  12. 0:21found is that between 2021 and 2025, 55%
  13. 0:26of the investment money flowing into AI
  14. 0:28companies came from other AI companies.
  15. 0:30So more than half the money that's
  16. 0:32funding the AI bubble is coming from
  17. 0:34inside the bubble, and it only gets
  18. 0:35messier as the checks get bigger. Some
  19. 0:38of these investors are also doing
  20. 0:39business with the companies they invest
  21. 0:41in, like a chip maker putting money into
  22. 0:43an AI company and that same AI company
  23. 0:45turning around and buying its chips.
  24. 0:47Deals like that make up only about one
  25. 0:49in six of the deals between AI
  26. 0:51companies, but when you look at the size
  27. 0:52of those deals, they're nearly half of
  28. 0:54all the money, which means the bigger
  29. 0:56the check, the more likely the investor
  30. 0:58is also the supplier or the customer.
  31. 1:00The study calls these relationships
  32. 1:02circular. The Bank for International
  33. 1:04Settlements flagged the problem in one
  34. 1:05very polite sentence. These circular
  35. 1:07arrangements tend to be opaque, making
  36. 1:09monitoring and supervision difficult,
  37. 1:12which is basically the same as the
  38. 1:13referees telling you they can't see the
  39. 1:14field. And until recently, that's where
  40. 1:17this story stopped. Everybody knew about
  41. 1:18the circular financing problem in AI,
  42. 1:21but it stayed mostly a so what type of
  43. 1:24hypothetical, the kind of thing that
  44. 1:25gets a thoughtful nod at a conference
  45. 1:28right before everyone heads to lunch.
  46. 1:30But recently, that's changed. Now we're
  47. 1:32seeing people starting to react to the
  48. 1:34risks in the AI bubble, and you can
  49. 1:35actually watch them react because
  50. 1:37there's one corner of Wall Street where
  51. 1:39fear has a price tag. That corner trades
  52. 1:41a product called a credit default swap,
  53. 1:44which is insurance on a company's debt.
  54. 1:46And if that sounds familiar, it's
  55. 1:47because this is the instrument Michael
  56. 1:49Burry used in the big short to bet
  57. 1:51against the housing market. It works
  58. 1:53like any other insurance. You pay a
  59. 1:54premium, and if the company stops paying
  60. 1:56its lenders, the insurance pays you,
  61. 1:58meaning the more worried the gets, the
  62. 2:00more that premium cost. Most of the
  63. 2:02time, nobody talks about this market.
  64. 2:04It's a quiet corner of finance that only
  65. 2:06makes noise when something's about to
  66. 2:07break, which means when it does start
  67. 2:09making noise, it's worth paying
  68. 2:11attention to. And as you can see from
  69. 2:13this chart, the CDS market is waking up.
  70. 2:15This shows the average amount of default
  71. 2:17insurance on AI companies traded each
  72. 2:19month over the last five years. Since
  73. 2:22the summer, the people who lent the AI
  74. 2:23bubble its money have been buying
  75. 2:25insurance against it at a record pace.
  76. 2:27In July, about $2 billion a week of
  77. 2:30default insurance was trading on the
  78. 2:31seven big AI linked names. By September,
  79. 2:34it was over 9 billion a week, more than
  80. 2:37four times the volume in just two
  81. 2:38months, and the prices followed the
  82. 2:40volume. A few weeks ago, insuring Meta's
  83. 2:43debt hit a record of about 119 basis
  84. 2:46points. That means investors are paying
  85. 2:48more for protection against a potential
  86. 2:50debt default, and investors don't pay up
  87. 2:52for protection unless they're starting
  88. 2:53to get worried. Then Nvidia, Broadcom,
  89. 2:56Alphabet, Oracle, and others all hit
  90. 2:58record highs as well. And at the riskier
  91. 3:00end of AI, the numbers stop looking like
  92. 3:02insurance and start looking like a
  93. 3:04prediction. But before we get into that,
  94. 3:06let's take a quick pause to talk about
  95. 3:08something else. According to Microsoft,
  96. 3:10the average worker gets 117 emails a
  97. 3:12day. It starts before breakfast. Of the
  98. 3:14people online at 6:00 a.m., 40% are
  99. 3:17already in their inbox. Then the workday
  100. 3:19gets in the way and the replies pile up.
  101. 3:21By 10 p.m., 29% of active workers are
  102. 3:24back in their boxes. That's hours of
  103. 3:26your week disappearing into a keyboard,
  104. 3:28which brings us to the sponsor of
  105. 3:30today's video, Whisper Flow. Whisper
  106. 3:32Flow is a voice-to-text tool that
  107. 3:34actually works. Here's an email that
  108. 3:36would usually take me about two and a
  109. 3:37half minutes to type. With Whisper Flow,
  110. 3:40the same email took me 30 seconds and
  111. 3:42there's nothing to fix. Whisper Flow
  112. 3:44isn't just for email either. Here's the
  113. 3:45same text spoken into my phone twice.
  114. 3:48Whisper Flow on the left, built-in voice
  115. 3:50on the right. One of these I can send,
  116. 3:52the other one I have to fix first.
  117. 3:54Whisper Flow turns your speech into
  118. 3:55clear, polished writing across any app
  119. 3:58and any device. If you're interested,
  120. 4:00Whisper Flow is free to use and you can
  121. 4:02get one month of Whisperflow Pro with
  122. 4:04unlimited words for free using the link
  123. 4:06below. Thank you Whisperflow for
  124. 4:07sponsoring this video. Now back to what
  125. 4:09the market's showing us with the AI
  126. 4:11bubble. The five-year CDS spreads for
  127. 4:13CoreWeave climbed past 1,100 basis
  128. 4:16points, while the yield on the company's
  129. 4:172031 bonds rose to 16 %. At those
  130. 4:21levels, the market is implying about a
  131. 4:2364% risk of default for CoreWeave over
  132. 4:26the next five years. That's more than a
  133. 4:28coin flip's chance. The market is
  134. 4:29pricing in a serious probability that
  135. 4:31CoreWeave won't be able to pay back its
  136. 4:33debt. And this is a company the AI
  137. 4:35bubble treats as critical
  138. 4:37infrastructure. What's weird is that
  139. 4:38nothing big in the AI bubble has
  140. 4:40actually defaulted yet. There isn't a
  141. 4:42single massive bankruptcy to point to,
  142. 4:44but the price of insuring these
  143. 4:46companies against one is hitting new
  144. 4:48highs anyway. And every insurance
  145. 4:50contract has two sides, someone
  146. 4:52collecting the premium and someone
  147. 4:53paying it. So who's actually moving this
  148. 4:55market? Well, you can't buy a credit
  149. 4:57default swap through your brokerage and
  150. 4:59there's no buy button for it on Robin
  151. 5:01Hood. In the US, this market is
  152. 5:03effectively closed to everyday investors
  153. 5:05by law. To trade over-the-counter swaps,
  154. 5:07you need more than $10 million in
  155. 5:09investments just to qualify. So the
  156. 5:11market is almost entirely made up of
  157. 5:13banks, hedge funds, and institutional
  158. 5:15money managers trading hundreds of
  159. 5:17billions of dollars a day. Part of the
  160. 5:19reason why comes down to one simple
  161. 5:20rule. Banks cap how much they'll lend to
  162. 5:23any one industry. And last year, the
  163. 5:25Chicago Fed found that the average large
  164. 5:27American bank had promised loans to
  165. 5:29AI-adjacent industries worth about a
  166. 5:31quarter of their Tier 1 capital. That's
  167. 5:33the bank's core capital, which is
  168. 5:35supposed to act as a safety net. It's
  169. 5:37there to make sure a bank can absorb
  170. 5:39large losses without threatening the
  171. 5:40stability of the institution itself.
  172. 5:43Loan promises are now worth a quarter of
  173. 5:45that safety net, all tied to a single
  174. 5:47theme. And when these banks started
  175. 5:48pressing against their own limits, they
  176. 5:50started looking for somebody else to
  177. 5:52hold the risk. In May, JPMorgan Morgan
  178. 5:55Stanley and one of the largest banks in
  179. 5:57Japan started looking for ways to
  180. 5:59offload their rapidly growing pile of
  181. 6:01data center debt. That same month, a
  182. 6:03survey of credit insurers who absorb
  183. 6:05risk from banks found several nearing
  184. 6:07their sector limits, or quote, are
  185. 6:09already full on data center exposure. By
  186. 6:12September, HSBC was working on a deal to
  187. 6:15hedge the default risk on more than a
  188. 6:16billion dollars in data center loans by
  189. 6:19transferring it to outside investors.
  190. 6:21Then, last week, the largest AI chip
  191. 6:23financing deal on record was signed. 63
  192. 6:26billion dollars raised to put chips in
  193. 6:28front of Anthropic, the AI lab behind
  194. 6:30the chatbot Claude. And the structure
  195. 6:32behind that deal is the strangest part.
  196. 6:35The money doesn't even go to Anthropic.
  197. 6:37It goes to a shell company that buys the
  198. 6:38chips and Anthropic leases them, which
  199. 6:40is a key detail we'll come back to. This
  200. 6:43was also a deal that almost didn't even
  201. 6:45close. Some of the banks underwriting it
  202. 6:47felt they weren't being paid enough for
  203. 6:48the risk and considered walking away.
  204. 6:51Broadcom's CEO spent the weekend before
  205. 6:53the announcement personally talking them
  206. 6:55off the ledge, but only in exchange for
  207. 6:57higher interest rates and more
  208. 6:58protection around Broadcom's guarantees.
  209. 7:00And two days after that deal closed,
  210. 7:02Bloomberg reported that Barclays is
  211. 7:04looking for default protection on more
  212. 7:06than 10 billion dollars of AI
  213. 7:08infrastructure debt. The stated reason
  214. 7:10is to create capacity for more lending.
  215. 7:12So the banks aren't pulling back from
  216. 7:14AI. They're still writing the loans and
  217. 7:16closing the deals. They just don't want
  218. 7:17to be the ones holding all the risk when
  219. 7:19they do. And we've seen a version of
  220. 7:21this from Wall Street before. It's the
  221. 7:22same playbook from the last great
  222. 7:24bubble, and we all know how that one
  223. 7:26ended. In 2005, Michael Burry started
  224. 7:29buying credit default swaps on mortgage
  225. 7:31bonds. He later said he was astounded at
  226. 7:33how willing the big investment houses
  227. 7:35were to sell him swaps on highly risky
  228. 7:37packages of mortgages. But a year and a
  229. 7:39half later, Wall Street caught on and
  230. 7:41started to agree with him. And we know
  231. 7:43exactly what that looked like from the
  232. 7:44inside because a Senate investigation
  233. 7:47later reconstructed everything using the
  234. 7:49bank's own emails. It started on
  235. 7:51December 14th, 2006. Goldman Sachs's
  236. 7:54mortgage desk had just lost money 10
  237. 7:56days in a row. That was enough for the
  238. 7:58chief financial officer, David Viniar,
  239. 8:00to pull the mortgage team into his
  240. 8:01conference room. His instruction was
  241. 8:03simple. Get closer to home. By closer to
  242. 8:06home, Viniar meant for the mortgage
  243. 8:08department to assume a more neutral risk
  244. 8:10position. The next day, he summed up the
  245. 8:12meeting in an email to Tom Montag, one
  246. 8:15of the firm's most senior executives. On
  247. 8:17Goldman's big mortgage bet, he wrote,
  248. 8:19the position is reasonably sensible, but
  249. 8:21is just too big. And then he followed up
  250. 8:23with, let's be aggressive distributing
  251. 8:25things because there will be very good
  252. 8:27opportunities as the markets go into
  253. 8:29what is likely to be even greater
  254. 8:30distress. The translation for this, sell
  255. 8:33it now even at a loss because the real
  256. 8:36money will be made when everyone else is
  257. 8:37forced to sell. And whatever they
  258. 8:39couldn't sell, they insured. But Goldman
  259. 8:41didn't stop at just insuring its
  260. 8:43holdings. By the end of the first
  261. 8:45quarter of 2007, Goldman's mortgage
  262. 8:47department had swung from a $6 billion
  263. 8:49net long position in December 2006 to a
  264. 8:52$10 billion net short position in late
  265. 8:55February 2007, a $16 billion reversal. A
  266. 8:59senior Goldman executive later described
  267. 9:01a $3 billion net short position in
  268. 9:03subprime mortgage-backed securities as
  269. 9:06huge and outsized. Goldman's net short
  270. 9:08position in February 2007 was $10
  271. 9:11billion,
  272. 9:12more than triple that size. That summer,
  273. 9:15as the market started collapsing, Viniar
  274. 9:17got an update on the mortgage desk
  275. 9:18numbers. His response tells you what
  276. 9:20might be happening to people who don't
  277. 9:22have the big short. By the end of 2007,
  278. 9:25those bets against the mortgage market
  279. 9:26had brought in about $3.7 billion.
  280. 9:29The positions were so large and risky
  281. 9:31that the mortgage department repeatedly
  282. 9:33breached its risk limits, and Goldman's
  283. 9:35senior management responded by
  284. 9:36repeatedly giving the department new,
  285. 9:38higher temporary risk limits to
  286. 9:40accommodate its trading. But here's the
  287. 9:42part that matters for where we are today
  288. 9:44with AI. While Goldman was betting
  289. 9:46against mortgages, they were still
  290. 9:48selling them. In 2006 and 7 alone,
  291. 9:51Goldman built and sold about $100
  292. 9:53billion of mortgage deals. One of them
  293. 9:56was a $1 billion deal called Timberwolf,
  294. 9:58a hybrid cash and synthetic CDO squared.
  295. 10:01It closed in March 2007, around the same
  296. 10:03time Goldman was flipping its bet to the
  297. 10:05downside, and within 5 months, the deal
  298. 10:08had lost 80% of its value. and Tom
  299. 10:10Montag, the same executive who received
  300. 10:13Viniar's email in December, later gave
  301. 10:15his opinion on the deal in an email that
  302. 10:17was read aloud at a Senate hearing. Boy,
  303. 10:19that timberwolf was one shitty deal. So
  304. 10:22that's what the inside looked like last
  305. 10:24time around. The firms closest to the
  306. 10:25risk saw it first. They didn't announce
  307. 10:27it. They didn't stop. They sold what
  308. 10:29they could, insured what they couldn't,
  309. 10:31and kept the deals moving. In July 2007,
  310. 10:34Citigroup's CEO, Chuck Prince, described
  311. 10:37this all perfectly when he was asked why
  312. 10:39his bank was still making risky loans.
  313. 10:41His answer became one of the most famous
  314. 10:43lines in finance. When the music stops
  315. 10:46in terms of liquidity, things will be
  316. 10:47complicated, but as long as the music is
  317. 10:49playing, you've got to get up and dance.
  318. 10:51We're still dancing. That music he's
  319. 10:53talking about is liquidity, and 19 years
  320. 10:55later, the same music is back. Only this
  321. 10:58time, it didn't come from the head of a
  322. 10:59bank. It came from the richest man on
  323. 11:01Earth. A few weeks ago, Elon Musk posted
  324. 11:04on X and said, Debt markets have almost
  325. 11:06entirely shut off access to money for
  326. 11:08any company that doesn't have real
  327. 11:09revenue outside of AI data centers. Then
  328. 11:12he explained why his own company was
  329. 11:13fine. He said, SpaceX just closed $40
  330. 11:16billion of debt for GPUs, but we're
  331. 11:18investment grade and have Starlink and
  332. 11:20launch revenue. Then came the
  333. 11:22prediction. The more than 200 neoclauds,
  334. 11:25which are cloud companies that rent out
  335. 11:26AI computing power, Musk says they're
  336. 11:28headed for, quote, guppy level death and
  337. 11:31consolidation. And anyone who's ever
  338. 11:33owned a fish tank knows how guppy deaths
  339. 11:35go. There's no funeral and no
  340. 11:36announcement. One morning, the tank just
  341. 11:38has fewer fish in it, and the big ones
  342. 11:40look well fed. Musk never used Chuck
  343. 11:42Prince's words specifically, but it's
  344. 11:44exactly what he's describing. A debt
  345. 11:46market slowing its lending on the AI
  346. 11:48story and starting to lend on everything
  347. 11:50else a borrower has. We're three years
  348. 11:53into the most hyped technology bubble in
  349. 11:54history, and the survivors are being
  350. 11:56selected by their non-AI income. That's
  351. 11:58what it looks like when market liquidity
  352. 12:00starts to go because market liquidity
  353. 12:02isn't a number on anyone's balance
  354. 12:04sheet. It's an assumption. The
  355. 12:06assumption that when you need to borrow
  356. 12:08or you need to sell, somebody will be
  357. 12:10standing on the other side at close to
  358. 12:11yesterday's price. And market liquidity
  359. 12:14is there right up until everybody checks
  360. 12:16for it at once, which leaves one
  361. 12:18question. If the debt market really has
  362. 12:20shut its doors on companies that live on
  363. 12:21AI alone, how did the largest AI chip
  364. 12:24financing deal in history close last
  365. 12:26week? Anthropic is one of the most
  366. 12:28important AI companies on the planet,
  367. 12:30but from a lender's perspective, they're
  368. 12:32still exactly the kind of borrower Musk
  369. 12:34is talking about. They're only five
  370. 12:36years old. Nearly every dollar they earn
  371. 12:38comes from AI, the one thing lenders
  372. 12:40have gotten nervous about, and they have
  373. 12:41committed to more than $500 billion
  374. 12:44in future infrastructure spending, which
  375. 12:46means a company like Anthropic walking
  376. 12:48into the debt markets alone for $63
  377. 12:50billion should get an ugly rate attached
  378. 12:53to that debt, if they even get an answer
  379. 12:55from lenders at all. Except Anthropic
  380. 12:57didn't walk into the debt markets alone.
  381. 12:59They tagged in Broadcom with them. It's
  382. 13:01the same reason a landlord will rent a
  383. 13:0322-year-old their first apartment if a
  384. 13:05parent co-signs the lease. The landlord
  385. 13:08isn't approving the kid. He's approving
  386. 13:10the parent's credit score. Broadcom made
  387. 13:12the lenders a promise. They've agreed to
  388. 13:14provide residual value support equal to
  389. 13:16the first lean portion. Basically, if
  390. 13:18Anthropic can't pay whatever the senior
  391. 13:20lenders are owed, then Broadcom will
  392. 13:22cover their obligations up to about $43
  393. 13:25billion. And Broadcom made a nearly
  394. 13:27identical promise on another Anthropic
  395. 13:29deal back in June for up to $29 billion.
  396. 13:33When you add it all up, Broadcom is on
  397. 13:35the hook for as much as $72 billion of
  398. 13:37other people's debt, and they're also
  399. 13:39reportedly weighing more than $50
  400. 13:41billion of financing to help open AI by
  401. 13:43chips. Nvidia has already done the same
  402. 13:46thing for Open AI. They agreed to
  403. 13:48provide up to $105 billion in financing
  404. 13:51for a new artificial intelligence data
  405. 13:53center for Open AI in Ohio. And Google
  406. 13:55does the same thing, one level down.
  407. 13:57They guarantee the rent owed to the data
  408. 13:59center companies building the computers
  409. 14:01that Anthropic runs on. Alphabet's
  410. 14:03latest filing shows $48 billion of these
  411. 14:06backstops. At the end of last year, it
  412. 14:08was about 17. These three companies
  413. 14:10together are standing behind more than
  414. 14:12$200 billion of other companies'
  415. 14:14obligations, which raises the obvious
  416. 14:16question, why would anyone volunteer for
  417. 14:18that? It's because every borrower on
  418. 14:21that list is also a customer. Broadcom's
  419. 14:23backstop keeps Anthropic buying Broadcom
  420. 14:25chips. Nvidia's keeps OpenAI filling an
  421. 14:28Nvidia only campus, and the companies
  422. 14:30doing the backstopping even collect a
  423. 14:32fee for the signature. The parents
  424. 14:34aren't backstopping their debt out of
  425. 14:35love. They're financing their own sales,
  426. 14:38which brings us all the way back to that
  427. 14:40study from the Bank of International
  428. 14:41Settlements at the start of this video.
  429. 14:43They asked where the money that's
  430. 14:45fueling the AI bubble is actually coming
  431. 14:47from and found that more than half of it
  432. 14:49comes from inside the industry. This is
  433. 14:51how the suppliers keep guaranteeing the
  434. 14:53money so their customers can keep
  435. 14:54buying. And when you start to see
  436. 14:56default insurance rising to record highs
  437. 14:58across the industry, that's the market
  438. 15:00getting nervous because a parent can
  439. 15:02only co-sign so many leases before their
  440. 15:04own credit starts looking stretched.
  441. 15:06It's not just three companies holding up
  442. 15:08Anthropic and OpenAI anymore. They're
  443. 15:10holding up something much bigger, the
  444. 15:12American economy. Last year, the
  445. 15:14research firm Renaissance Macro found
  446. 15:16that for the first time ever, AI
  447. 15:18investment added more to US GDP growth
  448. 15:21than all of American consumer spending
  449. 15:23combined. And JPMorgan found that since
  450. 15:25ChatGPT launched in 2022, just 42
  451. 15:29AI-linked companies produced about 65 to
  452. 15:3175% of the market's returns, profits,
  453. 15:34and capital spending. Their conclusion
  454. 15:36was the generative AI investment theme
  455. 15:39has smothered the rest of the US equity
  456. 15:41market. If you take these 42 AI-linked
  457. 15:44companies out, the American stock market
  458. 15:46would have underperformed Europe, Japan,
  459. 15:48and China. So the music hasn't stopped
  460. 15:50yet. None of the big AI borrowers has
  461. 15:52defaulted. The deals keep closing, and
  462. 15:54by every official measure, the dancing
  463. 15:56continues. But you don't want to be the
  464. 15:58one still on the floor when it ends. 4
  465. 16:00months after Chuck Prince said Citigroup
  466. 16:02was still dancing in 2007, he was fired.
  467. 16:04His bank went on to take some of the
  468. 16:06biggest mortgage losses on Wall Street,
  469. 16:08and they needed one of the largest
  470. 16:09bailouts in American history to survive,
  471. 16:1145 billion dollars from taxpayers plus a
  472. 16:14government guarantee on more than 300
  473. 16:16billion dollars of their assets. So the
  474. 16:19institutional money isn't calling the
  475. 16:20top. It's doing something quieter. The
  476. 16:22banks are still writing the loans. The
  477. 16:24chip makers are still shipping the
  478. 16:26chips. The deals keep getting bigger,
  479. 16:27just like we saw in 2007, but the risk
  480. 16:30is being passed around from the banks to
  481. 16:32insurers and credit default swap sellers
  482. 16:35and from the borrowers back onto the
  483. 16:37chip makers and Big Tech who are now
  484. 16:39guaranteeing other companies' bills just
  485. 16:41to keep the deals closing. We even have
  486. 16:43the central bank for central banks
  487. 16:45saying these circular arrangements tend
  488. 16:47to be opaque, making monitoring and
  489. 16:49supervision difficult. More than half
  490. 16:51the money flowing into AI companies
  491. 16:53comes from other AI companies. America's
  492. 16:55economic growth has been leaning on that
  493. 16:57money and the people financing it are
  494. 16:59now quietly buying insurance against it.
  495. 17:02>> [music]

About this transcript

This page contains the full transcript of The Smart People Have Started Looking for the Exit by Casual Finance, generated from the public captions YouTube serves with the video. The transcript has 3,273 words across 495 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.