YouTube2Text

Brad Gerstner: No AI Bubble, Semis Eat the Nasdaq & AI's Take Off Problem — Transcript

by All-In Podcast · 3,118 words · 464 segments · language en · Watch on YouTube

Full transcript

  1. 0:00Our favorite fifth bestie in the world,
  2. 0:03[music] the one, the only Brad Bersner
  3. 0:06from Altimter. Brad has had unbelievable
  4. 0:08[music] career starting five companies.
  5. 0:11So, he's got a very different mentality
  6. 0:12than your sort of classic hedge fund
  7. 0:14guy.
  8. 0:15>> He's an amazingly [music] successful guy
  9. 0:17and he's put up a tremendous amount of
  10. 0:19money.
  11. 0:19>> Every child in America, all 70 million
  12. 0:23kids under the age of 18 deserve to
  13. 0:25[music] have one of these accounts. This
  14. 0:27would not be a law if Brad Gerster did
  15. 0:30not pursue it with absolute dogged
  16. 0:33determination.
  17. 0:34>> This is not a program. This is a
  18. 0:36platform. It is the largest unlock of
  19. 0:39direct philanthropy in the history of
  20. 0:40the country.
  21. 0:41>> I think the antidote [music] to more
  22. 0:43socialism is more capitalism.
  23. 0:46Please welcome Brad Gersner.
  24. 0:49[music]
  25. 0:54Woo. [music]
  26. 0:56Wow.
  27. 0:58Let's go.
  28. 1:00[applause and cheering]
  29. 1:02Let's go.
  30. 1:04Thank you guys. And thank you for so
  31. 1:06much love yesterday, especially on the
  32. 1:08Trump accounts. So many people came up.
  33. 1:12Everybody gets what this means for
  34. 1:14America. We're in a battle for the soul
  35. 1:16of America. 70 million kids are going to
  36. 1:19be made direct owners in America. That
  37. 1:22is how we beat the scourge of capital or
  38. 1:25socialism. We make every child a
  39. 1:27capitalist.
  40. 1:29[applause]
  41. 1:32>> And also
  42. 1:34thank you to all the people yesterday
  43. 1:36who came up who took the CAC scan, the
  44. 1:39heart scan outside from the center for
  45. 1:41heart attack prevention which we
  46. 1:43started. There's no doubt based on these
  47. 1:45results we uh you know we're going to
  48. 1:48save some lives even yesterday. Okay.
  49. 1:51This is the highest ROI thing you can do
  50. 1:54in healthcare. Every cardiologist I talk
  51. 1:57to does this for themselves, their
  52. 1:59family, and their friends. It's a
  53. 2:00hundred bucks, 15 minutes. Get it done.
  54. 2:03If we turn this into the mammogram for
  55. 2:05the heart, we'll save 50,000 lives a
  56. 2:07year in this country. It's like ending a
  57. 2:09Ukraine war in America every year. So,
  58. 2:11go get your CAC scan done if you have
  59. 2:13it. But today is not about those two
  60. 2:14moonshots.
  61. 2:16Today is kind of a throwback to what I
  62. 2:18used to do on the pod with these guys,
  63. 2:20which is a market check, a tech check,
  64. 2:22state of the market. Where are we? Where
  65. 2:24are we going? What do we have to believe
  66. 2:26to be true in order for the market to
  67. 2:28continue to work? We're going to do a
  68. 2:30bit of a speed round here. So, bear with
  69. 2:32me. Get your cameras out. Get your notes
  70. 2:35out. Some good some good chart candy in
  71. 2:37here for you guys. So markets up 15%
  72. 2:41this year, up 39% since January of last
  73. 2:43year, despite all the concerns about
  74. 2:45tariffs, despite all the concerns about
  75. 2:47geopolitics,
  76. 2:49despite all the concerns about AI
  77. 2:51regulation.
  78. 2:52The scoreboard, we have a lot of people
  79. 2:55in the bestie group who said gold was
  80. 2:57going to be off the charts this year.
  81. 2:58It's flat. Bitcoin's down 10%. But look,
  82. 3:01we have Nvidia revenue up 2x.
  83. 3:03Hyperscaler capex up 2x. Open AI and
  84. 3:06Enthropics valuation up 2x. SpaceX up
  85. 3:09two and a halfx in a pretty nasty
  86. 3:11backdrop.
  87. 3:13This is not about multiple expansion.
  88. 3:16This is an earnings driven market
  89. 3:19expansion. We've seen multiple
  90. 3:21contraction this year. Earnings are up
  91. 3:2326% of course driven a lot by AI
  92. 3:26infrastructure, but the multiple on the
  93. 3:28NASDAQ and the S&P is actually down.
  94. 3:31Look at Nvidia trading at 14 times next
  95. 3:34year's fully taxed gap earnings. This is
  96. 3:37no bubble like it was in 2000.
  97. 3:40NASDAQ, S&P, Socks, Nvidia, all trading
  98. 3:44well below their average multiples.
  99. 3:47Okay, MAG 7 basically in line with its
  100. 3:51average multiple, but not everybody is
  101. 3:53winning. Okay, at the bottom here,
  102. 3:55consumer discretionary, software,
  103. 3:57financials, huge sections of the market
  104. 3:59have barely moved, right? This is a
  105. 4:02market that is being driven by the
  106. 4:04largest capex buildout, the largest
  107. 4:06super cycle in the history of
  108. 4:08technology.
  109. 4:11Semiconductors are 70% of the NASDAQ's
  110. 4:14return.
  111. 4:1670% of the return. That's both good and
  112. 4:18bad, and we'll get into that.
  113. 4:22So, who's making the money? The makers
  114. 4:23of the tokens are making the money and
  115. 4:26the buyers of the tokens are basically
  116. 4:28going along for the ride
  117. 4:33because of the tightness in the
  118. 4:35infrastructure market. We have massive
  119. 4:38public companies that look like venture
  120. 4:40capital returns.
  121. 4:42Dell up 5x up 9x in just 18 months.
  122. 4:49I love this chart. In blue you have the
  123. 4:52hyperscaler capex.
  124. 4:54In orange you have the free cash flow of
  125. 4:57the semiconductor companies. Do you
  126. 4:59notice anything?
  127. 5:02Their capex is almost dollar fordoll
  128. 5:04free cash flow to the infrastructure
  129. 5:06companies.
  130. 5:08Okay. Where were we at the start of the
  131. 5:10year? This is a really important
  132. 5:14framework to get your head around. You
  133. 5:16may remember a certain podcast I did
  134. 5:19with Sam Alman and Satcha in October of
  135. 5:21last year. And I asked Sam a very basic
  136. 5:24question that was on everybody's mind.
  137. 5:26How can you commit to a trillion dollars
  138. 5:28in capex when you have $13 billion of
  139. 5:31gap revenue?
  140. 5:33I thought this would be a way to clear
  141. 5:35up some confusion in the market.
  142. 5:37Instead, he told me to sell my shares.
  143. 5:42But then in the beginning of December we
  144. 5:44have opus 4.5 and claude code and then
  145. 5:48in January anthropics revenue was two
  146. 5:51billion and then in February it's 4
  147. 5:53billion and in March it was 11 billion.
  148. 5:56Do you notice anything that happened
  149. 5:57there in March? People started figuring
  150. 5:59out what anthropics revenue was. We had
  151. 6:02a an exclamation point answer to the
  152. 6:05question will the AI revenue showed up?
  153. 6:07It showed up in a massive way and that's
  154. 6:09why we got this historic run in April
  155. 6:11and May. We ripped off the bottom
  156. 6:14because the fuse was lit by Anthropic's
  157. 6:16monthly revenue. And then in in June and
  158. 6:20July, we had some consolidation. Why?
  159. 6:22Because Anthropic came out and said,
  160. 6:24"Oh, our annual run rate revenue is 65."
  161. 6:26People thought it was 75. So they had to
  162. 6:29revise down their estimates a little
  163. 6:30bit. And we had some concerns about open
  164. 6:33source. Is open source catching up? you
  165. 6:35know, is Anthropic going to continue to
  166. 6:36be able to generate those revenues? And
  167. 6:39so, we've kind of moved sideways since
  168. 6:40then.
  169. 6:44It's hard to get your head around the
  170. 6:46fact that these revenues have never
  171. 6:49happened before in the history of
  172. 6:51capitalism,
  173. 6:53right? We are on parabolic double
  174. 6:55exponential curves around these
  175. 6:57revenues. So the collective run rate,
  176. 7:00let's just call it of the top three
  177. 7:01labs, Anthropic, Open AI, and SpaceX is
  178. 7:05about a hundred billion based upon all
  179. 7:08the rumors that are out there coming out
  180. 7:10of July.
  181. 7:12I think they need to collectively get to
  182. 7:15at least $180 billion by the end of the
  183. 7:17year, add another $80 billion across
  184. 7:20those three labs just to keep the AI
  185. 7:23trade intact.
  186. 7:27So this is the single most important
  187. 7:29data point in the market today, right?
  188. 7:32Is Anthropics monthly revenue is
  189. 7:34OpenAI's monthly revenue going to be 4
  190. 7:36billion or 8 billion? It's almost hard
  191. 7:38to get your head around. Most of us
  192. 7:40who've been in venture capital for a
  193. 7:42long time, if you added a billion, if
  194. 7:44you got to a billion dollars in software
  195. 7:46revenue over like four or five years,
  196. 7:49you were in the top 5% of software
  197. 7:51companies. But this is what the world is
  198. 7:54now pricing in. So why is this so
  199. 7:56important? It's this slide. If you're
  200. 7:59going to build a a trillion half dollars
  201. 8:02a year in capex, somebody has to pay for
  202. 8:05it, right? Microsoft's not paying for
  203. 8:08it. They're building it to rent it.
  204. 8:10Google's not paying for it. They're
  205. 8:12building it to rent it. Amazon's
  206. 8:14building it to rent it. Well, who is the
  207. 8:16person renting it? We have to have the
  208. 8:18offtake revenues in order to pay that
  209. 8:20rent. So if we exit this year around,
  210. 8:23you know, let's call it 200 billion of
  211. 8:25run rate revenue. That's below the the
  212. 8:27years there. I think you have to go from
  213. 8:29200 to 450 to 800 or a trillion dollars
  214. 8:34just to keep up. The blue bar is the
  215. 8:37expected capex just from the MAG 5 and
  216. 8:40the orange bar is the offtake revenue,
  217. 8:42the gap offtake revenue that we need to
  218. 8:45see in order to keep this trade intact
  219. 8:48over the course of the next few years.
  220. 8:49Otherwise, we can't build this much
  221. 8:51capex.
  222. 8:53So, the labs are increasingly,
  223. 8:56you know, like the conversation I had
  224. 8:58with Sam last October, they're
  225. 9:00aggressively expanding compute. Why? All
  226. 9:02the things you heard yesterday, we're
  227. 9:04heading into recursive cycles. They're
  228. 9:06seeing, you know, incredible demand for
  229. 9:08the product. And so, they're building
  230. 9:10out compute. This year the total amount
  231. 9:12of uh compute added so 2026 about 19
  232. 9:16gawatt and about seven of those
  233. 9:18gigawatts went to the two leading labs.
  234. 9:21Next year this is semi analysis so Dylan
  235. 9:23Patel forecast that the work that we're
  236. 9:26going to add 43 gawatt and that about 14
  237. 9:29gawatt are going to go to the leading
  238. 9:32labs and we'll come back to this
  239. 9:33question of can we really stand up 43
  240. 9:37gawatts of compute next year. Notice
  241. 9:39that the amount that we're adding next
  242. 9:41year is as much as the total compute we
  243. 9:43have in the United States this year, the
  244. 9:46cumulative compute we have in the United
  245. 9:49States this year. So that's what the
  246. 9:51market is anticipating.
  247. 9:53What happens if that doesn't happen? And
  248. 9:56by the way, if you added those bars up,
  249. 9:58by 2028, to David Sax's point yesterday,
  250. 10:01over half of the total compute in the
  251. 10:04country is controlled by two labs.
  252. 10:07So it does the TAM exist? Does the TAM
  253. 10:11exist? Right? So if we look at the total
  254. 10:13TAM of knowledge work, right? That this
  255. 10:16is a massive massive category. You got
  256. 10:18consumer plus ads plus coding and all
  257. 10:21these white collar workflows, millions
  258. 10:24of enterprises. I would argue it's the
  259. 10:25largest TAM in the history of the world.
  260. 10:28You only have to get to about 4% of that
  261. 10:30TAM or 1.2 trillion in order to pay for
  262. 10:33the capex. So I don't think it's a TAM
  263. 10:35issue. Um, obviously Jensen was on our
  264. 10:39was on the pod and he he he talked two
  265. 10:41years ago that inference was going to 1
  266. 10:43billionx and remember all the people
  267. 10:45saying he's full of there's no way this
  268. 10:47can 1 billionx. That's exactly what
  269. 10:49we've done in the a age of agents,
  270. 10:52right? We've had this exponential token
  271. 10:54growth this year. 47 quadrillion tokens
  272. 10:58that are going to be produced. So it's
  273. 11:00not a question again of demand. Codeex
  274. 11:02users have grown 40x in the last eight
  275. 11:05months and knowledge work at enterprises
  276. 11:08if you look at the median amount that
  277. 11:10enterprises are spending about 17x over
  278. 11:13the course of the last 18 months and
  279. 11:14I've talked to many people in the
  280. 11:16audience here small businesses medium
  281. 11:18businesses large businesses businesses
  282. 11:20like alimter we can't operate our
  283. 11:22business without buying AI so this is
  284. 11:26not overly surprising to me and when we
  285. 11:29think about the productivity dividend to
  286. 11:30the economy me. You know, [snorts] if
  287. 11:32you look at 2015 to 2025, EPS growth was
  288. 11:36about 10%. That was 6% revenue plus
  289. 11:40about 38 bips of margin expansion every
  290. 11:42year out of the NASDAQ. So, here's the
  291. 11:44question. Can we turn the 38 bips to 100
  292. 11:47bips of margin expansion because of AI?
  293. 11:50The answer is obviously yes. Every
  294. 11:52company I talk to, Uber says we're going
  295. 11:53to grow 20%, we're not going to grow
  296. 11:55headcount. Snowflake says we're going to
  297. 11:57grow 30%, we're not going to grow
  298. 11:58headcount. That's what's happening. That
  299. 12:00is margin expansion. The single largest
  300. 12:03cost input to every one of these
  301. 12:04companies are humans and engineers. It's
  302. 12:06not that they're going to fire
  303. 12:07everybody. They're just not going to
  304. 12:08hire them at the rate that they hired
  305. 12:10him before. And then of course, we're
  306. 12:11going to have consumer agents in
  307. 12:13everybody's pocket. You may remember
  308. 12:14this bet I had with Bill. You know, when
  309. 12:16are we going to be able to book a hotel
  310. 12:19using your consumer agent in your
  311. 12:20pocket? I think we've just gotten that
  312. 12:22with Muse and with Instinct. This could
  313. 12:24be another trillion dollar category, but
  314. 12:26it's definitely going to consume massive
  315. 12:29tokens. Here are the three risks and
  316. 12:31challenges. And sorry I'm going so
  317. 12:33quick, but I want to keep it keep it
  318. 12:35moving. Regulation. We heard a lot about
  319. 12:37this yesterday. Power. And then what's
  320. 12:41going on with interest rates? So, you
  321. 12:42know, this is the regulation tugof-war,
  322. 12:45right? Um, and I I've heard from a lot
  323. 12:48of people, they're like, you're on both
  324. 12:49sides of the issue. Here's the fact,
  325. 12:52okay, the answer is not going to be on
  326. 12:55one end or the other. We're going to
  327. 12:58have to have common sense, pragmatic
  328. 12:59solutions that get my mom, my sister,
  329. 13:02and the br brother off the cliff, right?
  330. 13:05We need to give comfort and confidence
  331. 13:07to the people who elect our
  332. 13:09representatives that it's safe. And so,
  333. 13:11you heard Elon yesterday give a great
  334. 13:13suggestion around peer review. So, I'm
  335. 13:15confident that we're going to get there,
  336. 13:17but it's kind of messy the sausage
  337. 13:18making along the way. But we have a
  338. 13:21prior history of excess regulation. When
  339. 13:24people get scared, when activists start
  340. 13:27pushing an agenda, we shut down 67
  341. 13:30fision reactors in this country. We
  342. 13:32unilaterally disarmed against China.
  343. 13:35It's been a disaster for the country.
  344. 13:37Right? All the clean energy we could
  345. 13:39have gotten. Instead, we've gotten
  346. 13:41non-clean emissions because we had a
  347. 13:43group of activists who were hellbent on
  348. 13:45shutting down nuclear. can't allow this
  349. 13:47to occur to AI.
  350. 13:51[applause]
  351. 13:53So regulation's a threat. It's a risk.
  352. 13:57Atoms and energy are hard. So getting
  353. 14:00back to can we stand up 43 gawatt of
  354. 14:02compute. Our total compute in the
  355. 14:04country is less than 40 gawatt. Doing
  356. 14:06this in one year, we've got to overcome
  357. 14:08permitting and local opposition. You
  358. 14:10guys see all of that. grid
  359. 14:12interconnection delays, skilled labor
  360. 14:14shortages, power equipment is sold out.
  361. 14:16It's the largest buildout in the history
  362. 14:18of the country. I would suggest Dylan's
  363. 14:20forecast to 43 gawatt next year is too
  364. 14:23aggressive. I don't think we're going to
  365. 14:25get there. I think the total amount
  366. 14:27we're actually going to stand up is
  367. 14:28somewhere closer to 25 gawatt. And I
  368. 14:31think of that 25 gawatt, half of it will
  369. 14:34be for anthropic and open AI.
  370. 14:38Um I think that's enough to generate the
  371. 14:41revenue. Remember Anthropics revenue
  372. 14:43reportedly this year 100 billion 110
  373. 14:46billion. If they do that they're doing
  374. 14:48it with a gigawatt and a half of
  375. 14:49compute. So if they add another four or
  376. 14:52five gawatts of compute that's certainly
  377. 14:54enough to add another 100 billion in
  378. 14:56revenue. So I don't think we need more
  379. 14:58gigawatts to get to the revenue targets
  380. 15:01for next year. But this is my hunch that
  381. 15:03we're not going to get to the 43.
  382. 15:05And then of course we're going to hear
  383. 15:07more tomorrow. I think rate hikes are
  384. 15:08coming. I think it's now over 90% chance
  385. 15:10that we're going to have rate hikes
  386. 15:12tomorrow. Why does this matter? Because
  387. 15:13all of this now borrowed money, right?
  388. 15:16We're borrowing money in order to stand
  389. 15:17up these data centers. So the hurdle
  390. 15:19rate for that money is going up. And um
  391. 15:23so that's not only a challenge for data
  392. 15:25centers, but remember, you know, as
  393. 15:27Warren Buffett says that, you know,
  394. 15:29interest rates are to stocks what
  395. 15:32gravity is to matter,
  396. 15:34right? Right? If you can earn 5 1.5 or
  397. 15:366% on your money without taking equity
  398. 15:39risk, then you know it's going to be a
  399. 15:41challenge for stocks. So, here's where I
  400. 15:43think we are and we're going to end on
  401. 15:45this slide and and then then we'll bring
  402. 15:47the guys out and chop it up a little
  403. 15:48bit. But this is the flight path. This
  404. 15:50is how I think about managing the
  405. 15:51portfolio, right? We're up about the
  406. 15:55NASDAQ's up about 15%. But as I sit here
  407. 15:58and think about the risk, do I want to
  408. 15:59be small, medium, or large? This is how
  409. 16:02I think about the fan of potential
  410. 16:03outcomes. If the monthly AI lab revenues
  411. 16:06are closer to that$8 billion number, I
  412. 16:09think it's takeoff. Okay. I think we are
  413. 16:11going to see an IPO this year. Um, and
  414. 16:14so I'm paying very close attention to
  415. 16:17what in fact are those monthly revenue
  416. 16:19numbers. I think the trends are intact,
  417. 16:21but we will see. That's going to be the
  418. 16:22single most important thing as to
  419. 16:24whether or not uh between now and the
  420. 16:26end of the year uh we we have liftoff.
  421. 16:28Second one is rates, the election, oil
  422. 16:30prices. Obviously, rates are following
  423. 16:32oil prices to a certain extent. So, what
  424. 16:35happens there? You know, if rates were
  425. 16:37to go to five and a half on the 10-year,
  426. 16:38that's going to be a big burden uh you
  427. 16:41know, on on the equity market. Then
  428. 16:43finally, the regulation, the anthropic
  429. 16:45IPO, we saw a trade down yesterday
  430. 16:47because people are concerned that maybe
  431. 16:49we're going to have a halt or or
  432. 16:50postponement. That would obviously be a
  433. 16:52major issue. I don't think that's going
  434. 16:54to happen, but again, that would be a
  435. 16:56concern. So that's the fan of potential
  436. 16:58outcomes. From here we're up 15%.
  437. 17:01I think we could go higher uh you know
  438. 17:03through the balance of the year. Uh or
  439. 17:05we could go lower. I'll leave you with
  440. 17:07this.
  441. 17:08The period of 2023 to 2025, you only had
  442. 17:12to get one thing right. That AI was
  443. 17:15going to be the biggest super cycle in
  444. 17:16the history of technology. And you
  445. 17:18needed to shove your chips into the AI
  446. 17:19trade. That's it. Right? If you were in
  447. 17:22the AI trade, you made money. That is
  448. 17:24not where we are in 2026. Everybody
  449. 17:26knows about AI. It's all priced now.
  450. 17:29It's about facts and circumstances. Stay
  451. 17:32mentally flexible. Follow the facts.
  452. 17:35Don't yolo. Okay? Don't go forex like
  453. 17:38our friend up north who uh gave all his
  454. 17:40money to Citadel, right? Um Forex
  455. 17:44levered in this market, you know, very
  456. 17:47dangerous. So, in my estimation, we're
  457. 17:50medium position. We're mentally
  458. 17:52flexible. If we see those revenues come
  459. 17:54in big for these next few months and we
  460. 17:57see oil prices retreat, we're going to
  461. 17:59put more chips on the table. If not,
  462. 18:02we'll, you know, we'll reserve the right
  463. 18:03to go even smaller. With that, thank you
  464. 18:05all. Thanks for having me. [applause]

About this transcript

This page contains the full transcript of Brad Gerstner: No AI Bubble, Semis Eat the Nasdaq & AI's Take Off Problem by All-In Podcast, generated from the public captions YouTube serves with the video. The transcript has 3,118 words across 464 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.