YouTube2Text

Why the AI Bubble Is Finally Popping — Transcript

by Paper Billions · 4,406 words · 634 segments · language en · Watch on YouTube

Full transcript

  1. 0:00This is Sam.
  2. 0:00And Sam has a huge problem.
  3. 0:03For three years, everyone on the internet has told him his balloon is a bubble.
  4. 0:08And for three years, it just kept getting bigger.
  5. 0:11The people shouting "bubble" started to look silly.
  6. 0:14But what a difference a few months makes.
  7. 0:16Last quarter, Google, the company that
  8. 0:19makes money every time you type a question,
  9. 0:21spent nine billion dollars more than it brought in.
  10. 0:24Wait.
  11. 0:25Google is losing money?
  12. 0:27We're not losing money.
  13. 0:28We're spending it faster than it comes in.
  14. 0:30That's what losing money means for the rest of us.
  15. 0:33So the richest companies on Earth started doing something they haven't needed to do
  16. 0:37in decades: borrowing.
  17. 0:39About half a trillion dollars of it, this year alone.
  18. 0:42When a home loan goes bad, the bank gets a house.
  19. 0:46When some of these loans go bad,
  20. 0:48the bank gets a warehouse full of used computer chips.
  21. 0:51And here's Sam's real problem.
  22. 0:53His plans call for about eight hundred and fifty billion dollars of computers.
  23. 0:58Nobody has that kind of money.
  24. 0:59Not even him.
  25. 1:00But one of the biggest piles of money in America isn't in a bank.
  26. 1:04It's sitting in 401(k)s.
  27. 1:06Ten point eight trillion dollars of it.
  28. 1:09And every payday, it tops itself up.
  29. 1:11Automatically.
  30. 1:12Whether you're paying attention or not.
  31. 1:14Hang on.
  32. 1:15I'm in this?
  33. 1:16Everybody's in this.
  34. 1:17Even Warren Buffett.
  35. 1:26This year, four companies, Amazon, Google, Meta and Microsoft,
  36. 1:31plan to spend about seven hundred billion dollars building AI.
  37. 1:35That's about two thousand dollars for every person in America.
  38. 1:39Seven hundred billion.
  39. 1:41What do you even spend that on?
  40. 1:42Buildings the size of a small town.
  41. 1:45Chips to fill them.
  42. 1:46Power to run the chips.
  43. 1:48And then more buildings, because the first ones are already full.
  44. 1:52For years, nobody panicked.
  45. 1:54Big Tech had piled up cash the way a dragon piles up gold.
  46. 1:58And on paper, building a data center doesn't make you any poorer.
  47. 2:01Okay.
  48. 2:02I just spent three thousand dollars of my savings on this.
  49. 2:06You didn't spend it.
  50. 2:07You swapped it.
  51. 2:09Yesterday you had three thousand dollars in the bank.
  52. 2:11Today you have a computer worth three thousand dollars.
  53. 2:15So on paper, I'm exactly as rich as I was.
  54. 2:18On paper.
  55. 2:19Ask me again in three years what that computer's worth.
  56. 2:23This year, the cash ran short.
  57. 2:25In the first nine months,
  58. 2:26Google's business brought in about a hundred
  59. 2:28twenty-seven billion dollars and spent
  60. 2:31a hundred thirty-two billion on buildings and machines.
  61. 2:35More money went out than came in.
  62. 2:37We're spending more than the business makes.
  63. 2:40How do we pay the rest?
  64. 2:41Three ways.
  65. 2:42We borrow.
  66. 2:43We sell new shares, which means selling little pieces of the company.
  67. 2:47Or we get someone else to do the borrowing for us.
  68. 2:51Which one?
  69. 2:52For now?
  70. 2:52The first two.
  71. 2:53In June, Google announced about eighty-five billion dollars of new stock,
  72. 2:58the largest stock sale ever.
  73. 3:00One of the buyers would surprise you.
  74. 3:02Hold that thought.
  75. 3:03And in February,
  76. 3:04it borrowed with a bond that pays back in the year twenty-one twenty-six.
  77. 3:08So when do I get my money back?
  78. 3:11Twenty-one twenty-six.
  79. 3:13I'll be dead.
  80. 3:14Your great-great-grandkids won't.
  81. 3:16Tell them we said hi.
  82. 3:17We'll get to the third way.
  83. 3:18First, the people who own these companies noticed.
  84. 3:21If you have an S&P five hundred index fund, that's you.
  85. 3:25Of every hundred dollars in it,
  86. 3:27thirty-three dollars and sixty-four cents sits in just seven companies: Apple,
  87. 3:32Nvidia, Microsoft, Amazon, Alphabet, Broadcom and Meta.
  88. 3:37It says five hundred companies.
  89. 3:40It is.
  90. 3:41Seven of them just get a third of your money.
  91. 3:46Most of the time, a falling stock doesn't hurt the company at all.
  92. 3:50If I sell you my Meta share, how much does Meta get?
  93. 3:54Nothing.
  94. 3:54The money goes from you to me.
  95. 3:56Meta never touches it.
  96. 3:58So the stock price is just a scoreboard.
  97. 4:01Pretty much.
  98. 4:02A company can watch it drop all week and still pay every bill.
  99. 4:05Unless it needs to sell new shares.
  100. 4:07Then a lower price means selling more of itself for the same money.
  101. 4:11Like pawning your watch the week watch prices crash.
  102. 4:15Exactly like that.
  103. 4:16And that's where Big Tech is now.
  104. 4:18On October twenty-first,
  105. 4:20Meta reported it spent two billion dollars more than it brought in last quarter.
  106. 4:24Five days later, it sold thirty-seven and a half billion dollars of new stock.
  107. 4:29Two days after that, Google reported its nine-billion-dollar gap,
  108. 4:33and its stock fell more than nine percent.
  109. 4:36On November second, Oracle raised twenty billion dollars: half in bonds,
  110. 4:41half in new stock sold at a discount.
  111. 4:43The stock fell almost nine percent.
  112. 4:46Larry Ellison lost about eleven billion dollars in a day.
  113. 4:50Why is the stock falling?
  114. 4:52Because you're selling more of it.
  115. 4:54And who's buying?
  116. 4:55The people who just watched it fall.
  117. 4:57Then came Thursday, November fifth, two days after the midterms.
  118. 5:02The Nasdaq lost one thousand four hundred twenty-one points,
  119. 5:06the biggest point drop in its history.
  120. 5:08About two and a half trillion dollars of stock value disappeared in a day.
  121. 5:13Is this it?
  122. 5:14Is this the pop?
  123. 5:15If this is all it is, you got lucky.
  124. 5:17Stocks fall, stocks come back.
  125. 5:19A slow leak is the best way a bubble can end.
  126. 5:22So the stock isn't the real problem.
  127. 5:24I am.
  128. 5:25A share can fall to any price it likes.
  129. 5:28A loan can't.
  130. 5:29It's due on the date it's due, whether the AI ever pays off or not.
  131. 5:33That same day, the President posted about the drop.
  132. 5:36"Your 401(k)s have, therefore, dropped."
  133. 5:39Yep.
  134. 5:40They have.
  135. 5:41Okay.
  136. 5:42I need a second income.
  137. 5:44Funny you say that.
  138. 5:45You've been watching one for five minutes.
  139. 5:47The cartoon?
  140. 5:48This video.
  141. 5:49The script, the voices, your kitchen.
  142. 5:52Three hundred fifty-eight shots and more than twelve hundred sound effects,
  143. 5:56from a topic to a finished four K film in about eighteen hours.
  144. 6:00We make every video on this channel with a studio called Overnight.
  145. 6:05And they've just opened the app they use to the public.
  146. 6:08You give it an idea.
  147. 6:09Your own Claude researches it, writes it, voices it,
  148. 6:13builds every shot and mixes the sound, on your own Mac or PC.
  149. 6:17So I could make this.
  150. 6:19Myself.
  151. 6:20At this quality.
  152. 6:22Yes.
  153. 6:22With the charts?
  154. 6:23Yes.
  155. 6:24And the robots?
  156. 6:25And the robots.
  157. 6:27There are more than twenty-five styles in it.
  158. 6:29Money stories like this one.
  159. 6:31Three D news documentaries.
  160. 6:33Three D cartoon Shorts.
  161. 6:35Minecraft Shorts.
  162. 6:36It's usually a hundred twenty-seven dollars a month.
  163. 6:39Right now it's seventy-seven, and the price goes up as more people join.
  164. 6:43Whatever you join at stays yours.
  165. 6:45And at three hundred members, Overnight closes the doors to the public again.
  166. 6:50Nobody wants the internet flooded with these.
  167. 6:53Three hundred.
  168. 6:54Got it.
  169. 6:54Make fourteen videos in your first thirty days.
  170. 6:57If you're not happy with them, you get your first month back.
  171. 7:00made overnight dot com.
  172. 7:02It's on screen, and it's the first link in the description.
  173. 7:05Now, back to the bubble.
  174. 7:06To see how your 401(k) got tied to it, go back one year,
  175. 7:10to a single day in September.
  176. 7:15September tenth, twenty twenty-five.
  177. 7:18Oracle tells Wall Street it has four hundred
  178. 7:20fifty-five billion dollars of signed
  179. 7:22future contracts.
  180. 7:24The stock jumps thirty-six percent in one day,
  181. 7:27its best day since nineteen ninety-two.
  182. 7:29And for part of that day,
  183. 7:31Larry Ellison passes Elon Musk as the richest person on Earth.
  184. 7:35Larry.
  185. 7:36What did you sell?
  186. 7:37Nothing yet.
  187. 7:38Then how are you richer than me?
  188. 7:40Somebody promised to buy.
  189. 7:42Most of that promise came from one customer: OpenAI.
  190. 7:46About three hundred billion dollars of computing over five years,
  191. 7:50starting in twenty twenty-seven.
  192. 7:52That's about sixty billion a year.
  193. 7:54OpenAI's entire revenue in twenty twenty-five was thirteen billion.
  194. 7:59Sixty billion a year.
  195. 8:01Your whole revenue is thirteen.
  196. 8:03For now.
  197. 8:04So you're promising me almost five times everything you make.
  198. 8:09Every year.
  199. 8:10Where do I initial?
  200. 8:11So where does OpenAI get the money?
  201. 8:13Investors.
  202. 8:14In March it raised a hundred twenty-two billion dollars.
  203. 8:18About thirty billion of it came from Nvidia.
  204. 8:20Now watch where Nvidia's money goes.
  205. 8:23An investment.
  206. 8:24For you.
  207. 8:25For the data centers.
  208. 8:26For the chips.
  209. 8:27Record quarter.
  210. 8:28In October, the Bank for International Settlements, the bank for central banks,
  211. 8:33studied more than a thousand AI firms.
  212. 8:36Of every ten dollars invested in them,
  213. 8:38about five and a half came from other AI firms.
  214. 8:41To be fair, this is old, and it's legal.
  215. 8:45A car dealer will happily lend you the money for the car.
  216. 8:48But it only works while everyone in the circle keeps paying.
  217. 8:51And now the circle is opening a door to the public.
  218. 8:55Anthropic, the maker of Claude, is about to sell its shares on the stock market,
  219. 8:59and its filing says its own chip and cloud suppliers will buy twenty-six billion
  220. 9:04dollars of them.
  221. 9:05So who's buying Anthropic's new shares?
  222. 9:08Ten billion.
  223. 9:09Me.
  224. 9:10Ten billion.
  225. 9:11Me.
  226. 9:12Aren't you two the ones selling Anthropic its chips?
  227. 9:15We like our customers to do well.
  228. 9:17The rest goes to the public,
  229. 9:19and up to a fifth of it is set aside for ordinary people,
  230. 9:22on apps like Robinhood and Schwab.
  231. 9:24So anyone can join the circle.
  232. 9:27Always room for one more.
  233. 9:31The one they all lean on is OpenAI.
  234. 9:34In twenty twenty-five,
  235. 9:35it brought in thirteen billion dollars and lost almost twenty-one billion running
  236. 9:40the business.
  237. 9:41For every dollar it made, it lost about a dollar sixty.
  238. 9:45The year before, it was two thirty-seven, so it's getting better.
  239. 9:49It's also a lemonade stand that sells a glass for a dollar and spends two sixty
  240. 9:54making it.
  241. 9:54Busiest stand on the street.
  242. 9:57How much do you make on each glass?
  243. 9:59Make is a strong word.
  244. 10:01Then how do you stay open?
  245. 10:03Investors.
  246. 10:04And a much bigger stand next year.
  247. 10:06OpenAI's yearly revenue today is somewhere between fifty and seventy billion,
  248. 10:11depending on who's counting.
  249. 10:12So that sixty-billion-dollar Oracle bill went from almost five times its revenue to
  250. 10:18about all of it.
  251. 10:19Some of the people funding the gap are borrowing to do it.
  252. 10:22In September, Masayoshi Son's SoftBank sold eleven billion dollars of junk bonds,
  253. 10:28paying up to nine point seven five percent,
  254. 10:31to pay the last ten billion it promised OpenAI.
  255. 10:34Let me get this straight.
  256. 10:35You're borrowing money to invest money.
  257. 10:38I'm borrowing money to believe.
  258. 10:40At nearly ten percent.
  259. 10:42Belief is expensive.
  260. 10:44To serve OpenAI, Oracle became a landlord.
  261. 10:48Last quarter it spent about five billion dollars more than it brought in.
  262. 10:52Over its last full year, about twenty-four billion.
  263. 10:56Mister Ellison, we're cutting your credit rating.
  264. 10:59To what?
  265. 11:00One notch above junk.
  266. 11:02About half your future contracts come from one customer.
  267. 11:05It's a wonderful customer.
  268. 11:08That's the one we're worried about.
  269. 11:09That was July.
  270. 11:11A credit score one point above the line where the bank says no.
  271. 11:15You can't buy OpenAI on the stock market.
  272. 11:17But you can buy the companies tied to it.
  273. 11:20Nvidia, which owns a piece of OpenAI,
  274. 11:23is six dollars and thirty cents of every hundred in an S&P five hundred fund.
  275. 11:28So the balloon isn't in my fund.
  276. 11:31No.
  277. 11:32Just the hands holding the string.
  278. 11:36In November last year, Michael Burry,
  279. 11:38the investor who bet against the housing market before two thousand eight,
  280. 11:42accused Big Tech of something specific.
  281. 11:45To follow it, you need one boring word: depreciation.
  282. 11:49I bought a fifty-thousand-dollar van that lasts five years.
  283. 11:53Do I count all fifty thousand as a cost this year?
  284. 11:56No.
  285. 11:57You spread it out.
  286. 11:58Ten thousand a year for five years.
  287. 12:00And if I said the van lasts ten years?
  288. 12:03Then it's five thousand a year,
  289. 12:05and your profit suddenly looks five thousand dollars bigger.
  290. 12:08Without selling one more thing.
  291. 12:11Burry's charge was that Big Tech says its AI chips last longer than they really do.
  292. 12:16In his words:
  293. 12:17"Understating depreciation by extending useful life of assets artificially boosts
  294. 12:22earnings – one of the more common frauds of the modern era."
  295. 12:26By his count,
  296. 12:27that leaves a hundred seventy-six billion dollars of costs out of the big spenders'
  297. 12:32books over three years.
  298. 12:33CNBC couldn't confirm it.
  299. 12:35Remember the gaming computer?
  300. 12:37Chips get old fast.
  301. 12:39In early twenty twenty-four,
  302. 12:41renting one of Nvidia's H100 chips cost about eight dollars an hour.
  303. 12:45By October twenty twenty-five, about a dollar seventy.
  304. 12:49Then demand got so big that the rent on those old chips went back up thirty-eight
  305. 12:54percent, while the chips got older.
  306. 12:56In September,
  307. 12:57Oracle said it extended contracts on
  308. 12:59four-year-old chips at twenty percent more than
  309. 13:02the original price.
  310. 13:03Picture selling a four-year-old car for more than you paid for it new.
  311. 13:07That chip is four years old.
  312. 13:10And it just got a raise.
  313. 13:12That's not supposed to happen.
  314. 13:14Nobody told the customers.
  315. 13:16So the bears were right on paper and wrong in real life.
  316. 13:20So far.
  317. 13:21The other place to look is off the books.
  318. 13:24In October twenty twenty-five,
  319. 13:26Meta put its giant Hyperion data center in Louisiana into a separate company:
  320. 13:31twenty-seven billion dollars, mostly owned by an investment firm, Blue Owl.
  321. 13:36It sold bonds that pay back in twenty forty-nine.
  322. 13:39Meta's lease runs four years.
  323. 13:41This bond runs until twenty forty-nine.
  324. 13:44Your lease runs four years.
  325. 13:46With options to stay.
  326. 13:47And if you don't take them?
  327. 13:49Then we guarantee the building's value for sixteen years.
  328. 13:52And after that?
  329. 13:54It's a really nice building.
  330. 13:55None of it is hidden.
  331. 13:57It's legal, and it's in the filings, on pages almost nobody reads.
  332. 14:01And the biggest buyer of those bonds, about eighteen billion dollars of them,
  333. 14:06was reportedly PIMCO, a bond giant whose funds sit on plenty of 401(k) menus.
  334. 14:13Those bonds have reportedly slipped to around ninety-four cents on the dollar.
  335. 14:19So how do you tell when lenders get scared?
  336. 14:22You check what it costs to insure the loan.
  337. 14:24Wall Street calls that insurance a credit default swap.
  338. 14:28I lent Oracle ten million dollars.
  339. 14:30What if it can't pay me back?
  340. 14:32Then you pay me a little every year.
  341. 14:34If Oracle ever defaults, I cover your loss.
  342. 14:37How much is a little?
  343. 14:39Depends how scared everyone is this week.
  344. 14:41Here's Oracle's dial.
  345. 14:43At the start of this year,
  346. 14:44insuring ten million dollars of Oracle's debt cost about a hundred forty-four
  347. 14:48thousand dollars a year.
  348. 14:50By July, about two hundred thousand.
  349. 14:52Now, about three hundred thousand.
  350. 14:54That's what people pay to insure a junk-rated company,
  351. 14:58and Oracle is still officially rated five steps higher.
  352. 15:01A driver with a clean license, charged like he's had three crashes.
  353. 15:05For CoreWeave,
  354. 15:07the price says there's about a sixty percent
  355. 15:09chance it defaults within five years.
  356. 15:11Worse than a coin flip.
  357. 15:13Wait.
  358. 15:14If lenders are this scared, why lend at all?
  359. 15:17And if they love AI so much, why not just buy the stock?
  360. 15:20Because I'm not betting they win.
  361. 15:22I'm betting they last.
  362. 15:23Last how long?
  363. 15:25Four years.
  364. 15:26That's when my bond pays me back in full, with interest.
  365. 15:29If they go under in year five...
  366. 15:31You've already been paid.
  367. 15:33And I'm waving from a beach.
  368. 15:34In October,
  369. 15:35CoreWeave's bonds due in twenty thirty-one yielded more than sixteen percent a year.
  370. 15:41So why doesn't everyone lend to them?
  371. 15:43Because the US government pays me about five percent with no risk at all.
  372. 15:48So the scared money goes there, and the brave money buys the stock.
  373. 15:52And the middle?
  374. 15:53Empty.
  375. 15:53Which is why whoever's left in it gets to charge more.
  376. 15:56In late October,
  377. 15:58Elon Musk said debt markets are "almost
  378. 16:00entirely shut off" for AI firms that live on
  379. 16:04data centers alone.
  380. 16:05And he made a prediction.
  381. 16:07"Two hundred plus neoclouds are going to see guppy-level death and consolidation."
  382. 16:12And the brave money that buys the stock?
  383. 16:14Less of it is brave than you'd think.
  384. 16:16Funds that simply copy an index now own more than a fifth of the S&P five hundred.
  385. 16:21They don't read warnings.
  386. 16:23They buy when the money comes in.
  387. 16:27So if lenders won't lend to AI companies on their own,
  388. 16:30how did the biggest AI loans of the year still happen?
  389. 16:34Someone signed underneath.
  390. 16:36On October twenty-sixth,
  391. 16:37big banks and Blackstone lined up sixty-three
  392. 16:40billion dollars to buy Broadcom chips
  393. 16:43for Anthropic.
  394. 16:44But Anthropic isn't the one borrowing.
  395. 16:46A separate company borrows the money, buys the chips, and rents them to Anthropic.
  396. 16:51If the chips end up worth less than the loan, Broadcom covers the gap,
  397. 16:55up to about forty-three and a half billion dollars.
  398. 16:58With an earlier deal, Broadcom backs about seventy-two billion.
  399. 17:02In August, Nvidia did the same for OpenAI.
  400. 17:06It agreed to back up to a hundred five
  401. 17:08billion dollars of OpenAI's twenty-year lease
  402. 17:11on a data center campus in Ohio.
  403. 17:13Young company, very big apartment.
  404. 17:16Does the tenant have income?
  405. 17:18It's growing really fast.
  406. 17:20That's not what I asked.
  407. 17:22Who signs underneath?
  408. 17:23I do.
  409. 17:24And you are?
  410. 17:25The guy who sold him the furniture.
  411. 17:27Google does it too.
  412. 17:28At the end of last year,
  413. 17:30Google had promised to cover about seventeen
  414. 17:32billion dollars of other companies'
  415. 17:34data center rent and power bills.
  416. 17:36By September, forty-eight billion.
  417. 17:39Almost triple in nine months.
  418. 17:41It calls them "credit derivatives." What's that?
  419. 17:44Promises.
  420. 17:45If a customer can't pay the rent on its data center, we cover it.
  421. 17:49So you're selling insurance against your own customers going bust.
  422. 17:53When you put it like that, yes.
  423. 17:55Add it up,
  424. 17:56and the chip and cloud companies have co-signed about two hundred twenty-five
  425. 18:00billion dollars.
  426. 18:02And this isn't Enron.
  427. 18:03Nobody is hiding anything.
  428. 18:05Is any of this even allowed?
  429. 18:07It's all legal.
  430. 18:09And it's all in the footnotes.
  431. 18:11Who reads the footnotes?
  432. 18:12Exactly.
  433. 18:13Nvidia earns enough to afford it.
  434. 18:15But co-signing changes what can go wrong.
  435. 18:18Before, if your customer's stock fell, that was your customer's problem.
  436. 18:22Now, if your customer can't pay, you owe its rent.
  437. 18:26So who's actually safe in all this?
  438. 18:28Me.
  439. 18:29My loan is backed by the chips, the way a mortgage is backed by a house.
  440. 18:33And the chips hold their value because AI companies keep renting them.
  441. 18:37And we keep renting because investors keep funding us.
  442. 18:41And some of those investors are me.
  443. 18:44So everyone is safe because of the next guy.
  444. 18:47Now run the chain backwards.
  445. 18:49What if people use AI a little less next year?
  446. 18:53Then we rent fewer chips.
  447. 18:55Then my racks sit empty, and I sell the chips.
  448. 18:59Then used chips flood the market, and the price drops.
  449. 19:02Then my loan isn't backed by much.
  450. 19:04Then I'm the one paying out.
  451. 19:06And then I'm paying the rent.
  452. 19:08And one of the lenders in that earlier Broadcom deal is Athene,
  453. 19:12the insurer owned by Apollo that sells retirement annuities.
  454. 19:15It holds about seven hundred million dollars of that debt,
  455. 19:19and has promised four point nine billion more.
  456. 19:22So the next guy is me.
  457. 19:26Nobody in that line is a fool.
  458. 19:28And bankers have stood in this exact spot before.
  459. 19:31July twenty oh seven.
  460. 19:32Housing loans are starting to go bad, and the head of Citigroup, Chuck Prince,
  461. 19:37gives an interview to the Financial Times.
  462. 19:39"As long as the music is playing, you've got to get up and dance.
  463. 19:43We're still dancing."
  464. 19:44By November, he was out of a job.
  465. 19:47A year later, the music stopped for everyone.
  466. 19:50This isn't two thousand eight.
  467. 19:52Back then, the loans went to people who couldn't pay.
  468. 19:55Today the borrowers are some of the richest companies in history.
  469. 19:59What's the same is the dance.
  470. 20:01Everyone sees the risk, and nobody sits down first,
  471. 20:04because whoever stops spending first loses the race.
  472. 20:08So who stops first?
  473. 20:10Not me.
  474. 20:11Not me.
  475. 20:12Definitely not me.
  476. 20:14Shouldn't somebody check the exits?
  477. 20:16Later!
  478. 20:17What's changing is the music.
  479. 20:19Rates went up instead of down.
  480. 20:21The Fed raised rates in September,
  481. 20:23and after a hot inflation report on November tenth,
  482. 20:26markets fully expect another hike in December.
  483. 20:30In twenty twenty-one, Oracle borrowed at one point six five percent.
  484. 20:34On November second, at seven to eight point six five.
  485. 20:37The same dollar now costs four to five times the interest.
  486. 20:42Picture your mortgage rate quadrupling before you've unpacked.
  487. 20:45And there's one dancer on this floor who never sits down,
  488. 20:48because nobody ever asks it to.
  489. 20:50Your 401(k).
  490. 20:54About seventy million Americans have a 401(k).
  491. 20:57At most big employers, you're signed up automatically.
  492. 21:00If you never pick a fund, part of every paycheck goes into a target-date fund.
  493. 21:05That fund owns the same giant stocks the index does.
  494. 21:08Nobody decides to buy.
  495. 21:10It happens every payday.
  496. 21:12Where did my retirement money go this month?
  497. 21:15A little bit of every big company.
  498. 21:18A bigger bit of seven.
  499. 21:19Did I pick that?
  500. 21:21You picked nothing.
  501. 21:22That's how you got me.
  502. 21:24For a decade, these giants bought back their own shares.
  503. 21:27This year they started selling new ones.
  504. 21:30In the third quarter, Alphabet, Amazon, Microsoft,
  505. 21:34Meta and Oracle bought back just four point
  506. 21:37four billion dollars of stock between
  507. 21:39them, about ninety percent below their peak.
  508. 21:42Google's buybacks went to zero.
  509. 21:44So who ends up with all the new shares?
  510. 21:46When a company issues more shares, I end up holding my slice.
  511. 21:50I don't get to choose.
  512. 21:52Every payday?
  513. 21:53Every payday.
  514. 21:54Then there's Anthropic.
  515. 21:56Around November eighteenth, it plans to sell shares to the public,
  516. 22:00valuing it around a trillion dollars.
  517. 22:02Remember the shares set aside for people on Robinhood and Schwab?
  518. 22:06You don't have to press that button.
  519. 22:07If your fund tracks the whole stock market,
  520. 22:10it owns Anthropic within about a week of listing.
  521. 22:13That includes Vanguard's total-market funds, and through them,
  522. 22:16most of Vanguard's target-date funds.
  523. 22:19Nobody holding those funds gets asked.
  524. 22:21The last door isn't open yet.
  525. 22:23In March,
  526. 22:24the Labor Department proposed a rule that shields employers from lawsuits when they
  527. 22:28offer private credit, the kind of lending that funds data centers, in 401(k)s.
  528. 22:34Most likely as a slice inside those target-date funds.
  529. 22:37And private credit has a catch.
  530. 22:39Early this year,
  531. 22:40investors asked to pull forty-one percent of the money in one of Blue Owl's tech
  532. 22:45lending funds.
  533. 22:46Blue Owl capped withdrawals at five percent.
  534. 22:49I'd like my money back, please.
  535. 22:51Here's some of it.
  536. 22:52And the rest?
  537. 22:53Next quarter.
  538. 22:54Maybe.
  539. 22:55Go back to November fifth.
  540. 22:56The Nasdaq's biggest point drop ever.
  541. 22:59The day the President posted that your 401(k)s had dropped.
  542. 23:03That same day, the final version of that rule reached the White House for sign-off.
  543. 23:09Which sounds alarming.
  544. 23:11Unless the thing inside the balloon is real.
  545. 23:14Here's the builders' side.
  546. 23:15In its summer quarter, Nvidia brought in ninety-six billion dollars,
  547. 23:20more than double a year earlier.
  548. 23:21And it keeps about seventy-five cents of every dollar after making the chips: a
  549. 23:26lemonade stand that sells a glass for a dollar that cost a quarter to make.
  550. 23:31It's a bubble.
  551. 23:32Then why can't anyone buy enough of it?
  552. 23:34Oracle's signed future contracts have grown to six hundred sixty-four billion
  553. 23:39dollars.
  554. 23:40On November tenth, CoreWeave, the one with the scary insurance price,
  555. 23:44raised its forecast, and its shares jumped.
  556. 23:47ChatGPT has about one point two billion weekly users,
  557. 23:51about one in every seven people on Earth.
  558. 23:54And Anthropic.
  559. 23:55Last year it brought in four point six billion dollars.
  560. 23:59In the first nine months of this year, thirty-three billion,
  561. 24:02and it lost only about one point three billion running the business.
  562. 24:06So revenue can catch up with spending.
  563. 24:08One AI lab just showed it.
  564. 24:10And remember this one?
  565. 24:11The four-year-old chip.
  566. 24:13You said it would be junk.
  567. 24:15People are lining up to rent it.
  568. 24:17So far.
  569. 24:18So far is how every company gets built.
  570. 24:21Now, that buyer at Google's record stock sale.
  571. 24:24For decades, Warren Buffett was famous for steering clear of tech.
  572. 24:29Then, in the third quarter of twenty twenty-five,
  573. 24:32while he was still running Berkshire Hathaway,
  574. 24:34Berkshire bought into Google's parent company for the first time.
  575. 24:38In January, Buffett handed the top job to Greg Abel and stayed on as chairman.
  576. 24:43And in June, when Google sold all that new stock,
  577. 24:46Berkshire wrote it a check for ten billion dollars.
  578. 24:49Hold on.
  579. 24:50Berkshire Hathaway?
  580. 24:52Buffett's company?
  581. 24:53Buying the AI build-out?
  582. 24:54Berkshire bought Google while Warren was still running things.
  583. 24:58I just bought more.
  584. 24:59You don't usually buy tech.
  585. 25:00We looked at the business.
  586. 25:02Did you want a special price?
  587. 25:04Same price as everyone else.
  588. 25:06So yes, Buffett bought in.
  589. 25:08Well, his company did.
  590. 25:10Ten billion dollars, at full price, into the biggest build-out in tech.
  591. 25:15And so did your 401(k), a little every
  592. 25:18payday, at whatever the price was that day.
  593. 25:20For a young saver, that's the good news:
  594. 25:23buying every payday means buying every dip.
  595. 25:26The risk, one adviser told Fox Business on the day of the drop,
  596. 25:30is "the sixty-two-year-olds who are three years away from retirement."
  597. 25:36Yes.
  598. 25:37Just not the way anyone pictured.
  599. 25:39There's been no single crash.
  600. 25:41No AI borrower has defaulted, and Oracle is still rated investment grade.
  601. 25:46What popped is an idea: that the richest companies on Earth could pay for AI out of
  602. 25:51their own pockets.
  603. 25:53Google's buybacks went to zero.
  604. 25:55Hundreds of billions got co-signed, all legal, all in the footnotes.
  605. 26:00The price of borrowing, insuring and owning AI debt jumped all at once.
  606. 26:05And Buffett's company bought in anyway.
  607. 26:07So what happens now?
  608. 26:09Now we find out who can pay.
  609. 26:11When?
  610. 26:12Soon.
  611. 26:12The dates are already printed.
  612. 26:14And if they can pay?
  613. 26:15Then this was the deal of the century, and I got paid to watch it.
  614. 26:20Anthropic prices around November eighteenth.
  615. 26:22Nvidia reports again in late November.
  616. 26:25And on December ninth, the Fed decides whether money gets more expensive again.
  617. 26:30If the money coming in catches up,
  618. 26:32this was one of the most ambitious building
  619. 26:34projects in history, paid for on credit.
  620. 26:37The balloon turns into a building.
  621. 26:39If it doesn't, the bills are still due on the dates they're due.
  622. 26:43Nobody knows which, not even the people blowing up the balloon.
  623. 26:46In July twenty oh seven, a banker said they were still dancing.
  624. 26:50Today the music is still playing, and everyone's still on the floor.
  625. 26:54They're just starting to look around for the chairs.
  626. 26:57And on the next payday, the 401(k)s buy in again.
  627. 27:00Automatically.
  628. 27:01Whether you're paying attention or not.
  629. 27:04I'm paying attention now.
  630. 27:05Then pay attention to this.
  631. 27:07The app this video was made in is open to the public right now,
  632. 27:11at made overnight dot com.
  633. 27:13The price goes up as people join, and at three hundred members it closes again.
  634. 27:17The link is first in the description.

About this transcript

This page contains the full transcript of Why the AI Bubble Is Finally Popping by Paper Billions, generated from the public captions YouTube serves with the video. The transcript has 4,406 words across 634 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.