WEBVTT

1
00:00:00.000 --> 00:00:11.367
<v Sam>Thursday morning I opened the retirement app, and the biggest line in my default fund was one company. Then the alert landed: Nvidia's chief executive had resigned. Jensen Huang, gone, after thirty-three years.

2
00:00:11.367 --> 00:00:18.944
<v Alex>And my first thought was that the affair is the small story. The big one is what this company had quietly become.

3
00:00:18.944 --> 00:00:23.767
<v Sam>Which made the biggest line in that fund suddenly feel like a personal problem.

4
00:00:23.767 --> 00:00:32.722
<v Alex>The single load-bearing wall of the whole AI build-out had one person standing in front of it, and your retirement plan is leaning on that wall.

5
00:00:32.722 --> 00:00:49.256
<v Sam>Before we get into it, my week. My search lead forwarded me a Digiday piece about publishers. They're sitting around waiting for click data, and meanwhile Google keeps pushing its AI visibility features, which makes the decision to opt out a lot harder than it used to be.

6
00:00:49.256 --> 00:00:52.011
<v Alex>Harder how? Opting out sounds like a checkbox.

7
00:00:52.011 --> 00:01:05.789
<v Sam>It's a checkbox with a price tag. Leave, and you vanish from the answers people actually read. Stay, and you hand Google your pages and hope the clicks come back. Nobody can tell me yet what the clicks look like.

8
00:01:05.789 --> 00:01:07.167
<v Alex>So you're deciding blind.

9
00:01:07.167 --> 00:01:16.467
<v Sam>Blind, with my whole team watching me do it. Search traffic is a big slice of how my quarter gets judged, so you can imagine my Thursday.

10
00:01:16.467 --> 00:01:22.667
<v Alex>That is a week where somebody else's platform decision lands on your budget before you've finished your coffee.

11
00:01:22.667 --> 00:01:24.733
<v Sam>Which is, funnily enough, today's story.

12
00:01:24.733 --> 00:01:44.022
<v Alex>Welcome back to Dan's AI Intel. Sam has a retirement-app problem and I have a board problem, so I think we've earned this one. The point of the show is to help a curious person understand the AI revolution, not just keep pace with it, and to dig past the hype and the fear. I'm Alex.

13
00:01:44.022 --> 00:01:53.322
<v Sam>And I'm Sam. If you're enjoying the show, hit follow in whatever app you're in. It's one tap, it's free, and you won't miss the next one.

14
00:01:53.322 --> 00:02:12.611
<v Alex>Today a chief executive falls in two weeks, and everybody argues about the scandal. We want to ask something else. When one company holds up this much of the AI economy, what exactly are you betting on when you bet on it? The man, the chips, or something Nvidia has been doing with its own cash?

15
00:02:12.611 --> 00:02:24.667
<v Sam>We'll walk through the fourteen days, pull apart what's proven from what's only claimed, and follow the money from a divorce lawyer's desk to your pension. Then we go back to a teenager washing dishes.

16
00:02:24.667 --> 00:02:32.589
<v Alex>And somewhere in the middle there's a mechanism your retirement plan and my cloud contract both touch, and hardly anyone looks at it.

17
00:02:32.589 --> 00:02:35.000
<v Sam>And we do not agree about it.

18
00:02:35.000 --> 00:02:36.024
<v Alex>We really don't.

19
00:02:36.024 --> 00:02:46.610
<v Sam>So let's stake it out. The market has taken a big bite out of Nvidia, and the question on my mind is whether that's fear or sense. Where do you start?

20
00:02:46.610 --> 00:03:05.732
<v Alex>Fear. My starting position is that the price is a panic discount. The business more than doubled its revenue last quarter, my engineers can't leave CUDA without rewriting half of what they've built, and the stock trades at about thirteen times next year's earnings. That's the price of a company in decline, and this isn't one.

21
00:03:05.732 --> 00:03:22.122
<v Sam>Mine is that thirteen times is the market finally reading the footnotes. I don't think the man is the problem. I think the problem is what the money has been doing, and every time I look at it I end up asking the same thing about who pays.

22
00:03:22.122 --> 00:03:24.171
<v Alex>You haven't heard my case yet.

23
00:03:24.171 --> 00:03:28.610
<v Sam>I know your case. It's elegant. It's always elegant, Alex. That's the flaw.

24
00:03:28.610 --> 00:03:31.683
<v Alex>The flaw is my entire personality, I accept that.

25
00:03:31.683 --> 00:03:39.878
<v Sam>So that's the fight. A panic discount, or a fair price for a loop nobody has checked. We'll come back to who moved whom.

26
00:03:39.878 --> 00:03:50.805
<v Alex>Start on the twenty-ninth of October. The New York Times reports, citing US intelligence sources, that Huang has been the target of a Chinese espionage effort, through a woman close to him.

27
00:03:50.805 --> 00:03:53.537
<v Sam>Who is she? I only know a name.

28
00:03:53.537 --> 00:04:02.756
<v Alex>Alice Huang, no relation, which people keep needing to be told. An executive linked to Megaspeed, a Singapore data-centre company that had been an Nvidia cloud partner.

29
00:04:02.756 --> 00:04:10.610
<v Sam>Let me guess why Megaspeed matters. A partner's history becomes your story the moment something goes wrong. I've lived that with an agency.

30
00:04:10.610 --> 00:04:28.366
<v Alex>That's the shape of it. Megaspeed was already under investigation in the United States and Singapore over suspected diversion of restricted Nvidia chips to China. It began life as a company called 7Road International, tied to Chinese video games, and it grew sharply once the US restricted AI chip sales to China.

31
00:04:28.366 --> 00:04:36.220
<v Sam>And it became an Nvidia cloud partner within days of a Singaporean owner buying it. That detail made me put my coffee down.

32
00:04:36.220 --> 00:04:51.585
<v Alex>Same. Reporting traces the relationship between the two Huangs back well before this year. And the Times later reported that a classified memo, drafted under Secretary of State Marco Rubio, had warned officials about it before Huang joined President Trump's trip to China in May.

33
00:04:51.585 --> 00:04:52.951
<v Sam>So people knew something.

34
00:04:52.951 --> 00:04:56.707
<v Alex>Some people knew something. That's the careful way to say it.

35
00:04:56.707 --> 00:05:09.000
<v Sam>Because I've run a brand through a crisis, I read what Nvidia did next like a case study. They called the relationship not unusual. That is the sentence you write when you haven't understood what's coming.

36
00:05:09.000 --> 00:05:12.674
<v Alex>Trump did the same within days, calling Huang very above board.

37
00:05:12.674 --> 00:05:27.037
<v Sam>And then it unravels in public. Shares fall more than nine percent the day three House committee chairs ask Huang for a closed-door briefing. In my last crisis, the statement said calm and the traffic dashboard said otherwise, and I trusted the dashboard.

38
00:05:27.037 --> 00:05:37.391
<v Alex>The next day the Times reports the memo described Alice Huang as a spy for China's Ministry of State Security, and the Financial Times says US intelligence believed she'd been recruited.

39
00:05:37.391 --> 00:05:40.398
<v Sam>And on Sunday he writes to his own staff.

40
00:05:40.398 --> 00:05:47.412
<v Alex>I was unfaithful to my wife. My actions hurt her and our children. There is no excuse for what I did.

41
00:05:47.412 --> 00:05:53.758
<v Sam>It's a short letter. I'd have expected lawyers all over it, and there's nothing in it to take apart.

42
00:05:53.758 --> 00:06:02.777
<v Alex>The board opened an independent inquiry that same day, and two days later named Wachtell Lipton and a former acting US attorney, Joon Kim, to run it.

43
00:06:02.777 --> 00:06:09.791
<v Sam>Hold on. That same Sunday, the board said it had no reason to believe national security or company secrets were compromised.

44
00:06:09.791 --> 00:06:19.477
<v Alex>It did, and it said it had confidence in him. Keep that in your head, because that board went from defending him to accepting his resignation in four days.

45
00:06:19.477 --> 00:06:39.852
<v Alex>What turned it was not the spy claim, which hadn't been proved. The trigger was the second relationship. On the eleventh the Times reported an affair with a senior executive who reported directly to Huang. Neither Nvidia nor the executive has confirmed it. The Wall Street Journal said the board was preparing to remove him, and he resigned the next day.

46
00:06:39.852 --> 00:06:53.213
<v Sam>In my world that is the line. A private failing is a private failing. A relationship with someone who reports to you is a governance problem, and I have sat in the HR meeting where that distinction is explained slowly.

47
00:06:53.213 --> 00:06:59.893
<v Alex>And no board can investigate around it. You can't have the person under review at the head of the table.

48
00:06:59.893 --> 00:07:10.582
<v Sam>I manage a big team. If I were involved with a director who reports to me, I wouldn't get an inquiry. I'd get a meeting with HR and a very quiet afternoon.

49
00:07:10.582 --> 00:07:13.254
<v Alex>You'd pack the little plant from your desk.

50
00:07:13.254 --> 00:07:16.928
<v Sam>I'd pack the plant with dignity. What did he lose, exactly?

51
00:07:16.928 --> 00:07:31.959
<v Alex>Every role. President, chief executive, director. Colette Kress, the finance chief, is interim chief executive and keeps her finance job. Harvey Jones, a former Synopsys chief executive, becomes the company's first-ever board chair, and there's a search for a permanent chief executive, inside and out.

52
00:07:31.959 --> 00:07:35.967
<v Sam>And Reuters adds another relationship, with a further employee, that same day.

53
00:07:35.967 --> 00:07:52.000
<v Alex>Reported, not confirmed. And Washington swung as fast as the board did. Trump said the FBI was investigating and called it horrible and sad, then after the resignation suggested Apple's Tim Cook should replace Huang. The Times later reported he'd privately urged Huang to fight for his job.

54
00:07:52.000 --> 00:07:55.684
<v Sam>So the White House owns a piece of this now.

55
00:07:55.684 --> 00:08:01.947
<v Alex>A piece of it, yes. And Huang has agreed to testify to House committee leaders before Thanksgiving.

56
00:08:01.947 --> 00:08:11.526
<v Sam>This is the part I find hard. There's the affair, there's the spy claim, and there's the chips. They sound like one story and they aren't.

57
00:08:11.526 --> 00:08:19.632
<v Alex>Three claims, and I would sort them the way we sort incidents: what we know, what we think, what we are guessing.

58
00:08:19.632 --> 00:08:28.474
<v Sam>When my team has three claims tangled together, I make them label each one before I let anyone speak. Admitted, reported, alleged. Try that.

59
00:08:28.474 --> 00:08:35.474
<v Alex>Fine. Admitted is the affair with Alice Huang, in his own letter, and the resignation. Both on the record.

60
00:08:35.474 --> 00:08:36.211
<v Sam>Then reported.

61
00:08:36.211 --> 00:08:44.316
<v Alex>The relationship with a direct report is one newspaper's reporting, and Reuters has since reported another. Nobody at Nvidia has confirmed either.

62
00:08:44.316 --> 00:08:49.105
<v Sam>And third is the spying, which is the one the whole internet grabbed.

63
00:08:49.105 --> 00:08:59.421
<v Alex>The allegation is that Alice Huang passed information about Jensen to Beijing, and that she and her family's companies helped move Nvidia chips into China despite export controls.

64
00:08:59.421 --> 00:09:18.579
<v Sam>And I want to be careful with the word allegation, because on a feed it hardens into fact within the hour. I've watched a rumour about our brand travel faster than the correction, and the correction never catches up. So I'm going to keep saying alleged until somebody shows me the page.

65
00:09:18.579 --> 00:09:21.895
<v Sam>Why would officials take that seriously, if it's unproven?

66
00:09:21.895 --> 00:09:38.474
<v Alex>Concrete reasons. Megaspeed reportedly bought billions of dollars of Nvidia hardware after the export restrictions. A company founded by her family became an Nvidia-preferred cloud partner. And in 2025 Nvidia's general counsel contacted a senior Commerce Department official to ask about her detention in Singapore.

67
00:09:38.474 --> 00:09:41.053
<v Sam>Her detention. That one I hadn't heard.

68
00:09:41.053 --> 00:09:54.316
<v Alex>Singapore has since barred her from entering, saying she gave inaccurate information to its police. Megaspeed denies diverting chips and says Singapore closed its own investigation without action. China's foreign ministry called the reporting pure fabrication.

69
00:09:54.316 --> 00:09:58.737
<v Sam>So I've got smoke and a lot of people denying a fire.

70
00:09:58.737 --> 00:10:13.474
<v Alex>And inside the US government it's genuinely contested. Attorney General Todd Blanche said the Justice Department had reached no conclusion. Representative Jim Himes doubted the characterisation after a classified briefing. And CNN reported some intelligence officials thought it was over-torqued.

71
00:10:13.474 --> 00:10:16.789
<v Sam>Over-torqued. That's what I say about my agency's decks.

72
00:10:16.789 --> 00:10:30.421
<v Alex>The pattern officials describe is a classic one. A well-connected businesswoman, introduced through her husband, a Chinese businessman named Kang Wei, builds a close personal tie to the most important man in a strategic industry while her…

73
00:10:30.421 --> 00:10:40.000
<v Sam>While her companies buy that industry's restricted products. Yes, I got there. It sounds like a spy novel, and that's the part that makes me nervous.

74
00:10:40.000 --> 00:10:54.155
<v Alex>It sounds like a pattern, which is different. In a post-mortem I'd call it a contributing factor, never a cause. Those facts make the concern reasonable. They don't prove she reported to Beijing, and the strongest evidence, if it exists, is classified.

75
00:10:54.155 --> 00:11:11.681
<v Sam>This makes me impatient. A claim that rests on evidence I can't see is exactly what a vendor hands me when the numbers are thin. I'm not saying that's what this is. I'm saying how I'd treat it in a budget meeting: not as a fact until somebody shows me the page.

76
00:11:11.681 --> 00:11:33.925
<v Alex>That's the standard I use in an incident review. You don't need the root cause to decide whether the person on call can stay on call. The board didn't have to settle the spy case. It had to settle whether it could keep a chief executive whose conduct it could no longer investigate. Those are different questions, and the second had an answer by the weekend.

77
00:11:33.925 --> 00:11:45.721
<v Sam>So here's what I'd let the team say in public today. Admitted, yes. Reported, with the paper named. Alleged, and only ever as alleged. If my comms lead wrote anything stronger, I'd send it back.

78
00:11:45.721 --> 00:11:54.483
<v Alex>And no public evidence shows any Nvidia secret or US national security information leaked. If I could put one line on a wall, it's that one.

79
00:11:54.483 --> 00:11:59.539
<v Sam>The testimony before Thanksgiving and the board's inquiry. Those are the places something could surface.

80
00:11:59.539 --> 00:12:06.953
<v Alex>And I've been sure before. My failed start-up died of me being sure, so I'm not saying more than anyone can prove.

81
00:12:06.953 --> 00:12:12.346
<v Sam>Here's what fills my feed. Is his wife about to walk off with half of Nvidia?

82
00:12:12.346 --> 00:12:27.849
<v Alex>Huang hasn't filed for divorce, and Lori Huang, his wife since the mid-eighties, has said nothing in public. But the question matters, because he's still one of the biggest individual shareholders. The latest proxy has him at about three and a half percent of the company.

83
00:12:27.849 --> 00:12:32.568
<v Sam>Which, with the company worth under four trillion, is a staggering amount of money.

84
00:12:32.568 --> 00:12:43.016
<v Alex>Around a hundred and forty billion dollars at mid-November prices. The bulk of it sits in a living trust where Lori is co-trustee, and a chunk more in the family foundation.

85
00:12:43.016 --> 00:12:48.745
<v Sam>So she's co-trustee on the bulk of it. That's a lot of leverage at a kitchen table.

86
00:12:48.745 --> 00:13:02.563
<v Alex>Leverage over the money, not over the company. Nvidia has one class of shares, one vote each. Unlike Meta or Alphabet there's no super-voting stock. His power came from being the founder and chief executive, and he has given up both.

87
00:13:02.563 --> 00:13:12.000
<v Sam>So a divorce changes nothing about who runs Nvidia. That's what I'd say to whoever in my feed has already picked a side. Money isn't a vote here.

88
00:13:12.000 --> 00:13:23.395
<v Alex>Nothing. His stake would at most become two stakes of about one point eight percent each, and neither is big enough to steer the company against the index funds and institutions that own far more.

89
00:13:23.395 --> 00:13:26.000
<v Sam>And you wanted to say something about Amazon.

90
00:13:26.000 --> 00:13:35.767
<v Alex>When MacKenzie Scott got about four percent of Amazon in her 2019 divorce, Jeff Bezos kept voting control over her shares. At Nvidia there's no control left to bargain over.

91
00:13:35.767 --> 00:13:42.279
<v Sam>What about the money, though? I run a payroll, I know what a big shareholder leaving does to a mood.

92
00:13:42.279 --> 00:13:55.953
<v Alex>California law sets the bulk of it. The stake was built during the marriage, so it's presumed community property, normally split equally. California is no-fault, so the affair itself wouldn't raise her share. A prenup could change that, and none is known.

93
00:13:55.953 --> 00:13:58.558
<v Sam>So the headline in my feed was clickbait.

94
00:13:58.558 --> 00:14:16.465
<v Alex>The practical consequence is quieter. Either of them could eventually sell part of a very large stake, a slow supply of shares the market has to absorb. And Huang was already selling steadily under a pre-arranged plan announced in 2025, and as a former officer he's no longer bound by the company's own trading windows.

95
00:14:16.465 --> 00:14:25.256
<v Sam>And the kids? Because if a divorce touches the company through two people who work there, that is the part I would worry about as a manager.

96
00:14:25.256 --> 00:14:37.953
<v Alex>Both his children work at Nvidia, in product roles, so the family is woven into the company in a way that outlasts any settlement. The governance question was settled the day he resigned. The money is a family matter.

97
00:14:37.953 --> 00:14:43.163
<v Sam>I want the human bit now. Because everything we've said treats him like a weather system.

98
00:14:43.163 --> 00:14:54.558
<v Alex>He was born Jen-Hsun Huang in Taipei and grew up in Tainan, the second son of a chemical engineer at an oil refinery and a schoolteacher. The family moved to Thailand when he was small.

99
00:14:54.558 --> 00:14:57.163
<v Sam>And then his parents send him to America.

100
00:14:57.163 --> 00:15:06.930
<v Alex>When he was nine. To an uncle in Tacoma, who enrolled him and his older brother at the Oneida Baptist Institute in Kentucky, believing the place a prestigious boarding school.

101
00:15:06.930 --> 00:15:12.140
<v Sam>And it was a reform school. I keep trying to picture a nine-year-old reading that brochure.

102
00:15:12.140 --> 00:15:18.000
<v Alex>A religious academy for troubled youth. His parents had sold nearly everything they owned to pay for it.

103
00:15:18.000 --> 00:15:24.186
<v Sam>They sold nearly everything for a school that turned out to be a reform school. That keeps bothering me.

104
00:15:24.186 --> 00:15:34.930
<v Alex>Soon after, the family settled in Beaverton, Oregon. He skipped grades and graduated early. And as a teenager he washed dishes, bussed tables and waited at a Denny's, sometimes on the graveyard shift.

105
00:15:34.930 --> 00:16:00.000
<v Sam>That's a kid being thrown across the world by parents who believed in a brochure. And this week my daughter's school sent home a parent survey about AI, and Maya, being Maya, filled out the student column and the parent column and made me read both before dinner. It said most parents want students to use AI, with strong safeguards. The part that stung was that four in ten teachers say they've had no training at all.

106
00:16:00.000 --> 00:16:06.207
<v Alex>That's an EdSurge survey from this month, yes. Parents want it taught, and the teachers haven't been taught.

107
00:16:06.207 --> 00:16:17.931
<v Sam>And what gets me is that all of it traces back to people like Huang placing a bet on a chip. Now my girl's homework runs on it, and her teachers are playing catch-up.

108
00:16:17.931 --> 00:16:28.276
<v Alex>I'd want the teacher in that classroom to have played with the tool before the kids did. It's the thing I keep nagging the engineering managers I work with about.

109
00:16:28.276 --> 00:16:40.000
<v Sam>Which is why I want you to listen for one thing in his story. A man who survives by never stopping. Hold that, since I think it comes back at us in the loop.

110
00:16:40.000 --> 00:16:42.759
<v Alex>Noted. I'll try to leave you the dessert.

111
00:16:42.759 --> 00:16:58.621
<v Alex>He studied electrical engineering at Oregon State, chosen for its low tuition, and met Lori Mills in the lab. First job, designing microprocessors at AMD. Then LSI Logic, where he worked on a graphics accelerator for Sun Microsystems with two engineers, Chris Malachowsky and Curtis Priem.

112
00:16:58.621 --> 00:17:01.034
<v Sam>That partnership became Nvidia, in a Denny's.

113
00:17:01.034 --> 00:17:15.172
<v Alex>Of course in a Denny's. April 1993, East San Jose, two hundred dollars each. Their reasoning was that personal computers would need dedicated chips to draw 3D graphics, and that video games were the market hungry enough to pay for them.

114
00:17:15.172 --> 00:17:17.586
<v Sam>And he's chief executive on day one.

115
00:17:17.586 --> 00:17:21.379
<v Alex>At thirty. Priem later said the founders basically deferred to Jensen.

116
00:17:21.379 --> 00:17:23.793
<v Sam>And the early years nearly killed it.

117
00:17:23.793 --> 00:17:37.931
<v Alex>The first chip flopped, and an investment from Sega kept the company alive. Before the RIVA 128 launched, Nvidia had roughly one month of payroll left. Huang's unofficial motto was that the company was thirty days from going out of business.

118
00:17:37.931 --> 00:17:44.828
<v Sam>Thirty days of payroll. I know exactly what that looks like. It looks like the last week of every quarter.

119
00:17:44.828 --> 00:17:51.034
<v Alex>The RIVA 128 sold, and a couple of years later came the GeForce 256 and the public listing.

120
00:17:51.034 --> 00:18:04.483
<v Sam>What I notice is that everyone says endurance is the point, that the man outlasts everyone. But a person who built a company by refusing to stop is also a person who kept the strategy in his own hands.

121
00:18:04.483 --> 00:18:18.621
<v Alex>I've been that founder. Both my start-ups ran on me refusing to stop and keeping every decision in my hands. One sold, one failed spectacularly. That habit is the strength his successors inherit, and the dependence they now have to unwind.

122
00:18:18.621 --> 00:18:40.000
<v Alex>And this is the bit that still makes me grin. A gaming company became the AI company by noticing what game chips actually were. A graphics processor does thousands of small calculations at once, which is also what scientific simulation and, later, neural networks need. With the GeForce 8800, Nvidia launched CUDA, software that let programmers use graphics chips for general computing.

123
00:18:40.000 --> 00:18:46.242
<v Sam>Was that a hit? I'd imagine a big new software layer got funded fast and cheered loudly.

124
00:18:46.242 --> 00:18:53.586
<v Alex>Shareholders saw it as a cost. It added expense to every chip for years while the market value stayed modest.

125
00:18:53.586 --> 00:19:00.930
<v Sam>So the thing that made them is the thing investors hated, which is the most marketing-department story I've ever heard.

126
00:19:00.930 --> 00:19:11.211
<v Alex>Then in 2012 researchers in Toronto trained AlexNet, the image-recognition network that started the deep-learning boom, on two GeForce gaming cards that cost about five hundred dollars each.

127
00:19:11.211 --> 00:19:19.289
<v Sam>Gaming cards. For that. Imagine telling your CFO the future of the industry came out of a box at a games shop.

128
00:19:19.289 --> 00:19:33.977
<v Alex>Huang recognised what had happened and bet the company on it. A few years on, Nvidia built the DGX-1, an AI computer priced at a hundred and twenty-nine thousand dollars, and he personally delivered one of the first to OpenAI.

129
00:19:33.977 --> 00:19:39.852
<v Sam>Personally. That's a founder, not a vendor. Nobody in my vendor stack has ever hand-delivered anything.

130
00:19:39.852 --> 00:19:46.461
<v Alex>And later it bought the networking company Mellanox so it could sell whole data-centre systems rather than chips.

131
00:19:46.461 --> 00:19:53.438
<v Sam>Then what's the moat? Everyone says moat, and I've learned people say it when they can't explain the lock-in.

132
00:19:53.438 --> 00:20:20.977
<v Alex>Software and systems, not silicon alone. Think of the point-of-sale system in my stores. Replacing it isn't about the till, it's the years of workflows, reports and staff habits built around it. Hundreds of CUDA libraries and two decades of developer habit mean most AI code assumes Nvidia hardware in just that way. NVLink and the networking let thousands of chips act as one machine. And the annual roadmap keeps rivals chasing the last generation.

133
00:20:20.977 --> 00:20:24.648
<v Sam>But Google has TPUs, Amazon has Trainium, AMD has chips.

134
00:20:24.648 --> 00:20:33.828
<v Alex>Real competitors. But switching means rewriting software and retraining staff, which is how Nvidia can charge prices that produce a gross margin around seventy-five percent.

135
00:20:33.828 --> 00:20:42.641
<v Sam>I'd end the call. When I see a vendor margin like that, my first move is asking my team what we could build ourselves.

136
00:20:42.641 --> 00:21:01.000
<v Alex>And it keeps buying whatever might breach the wall. A licensing deal for Groq's inference chips, and the proposed Hugging Face purchase. The one wall it couldn't hold is political. US export rules shut it out of China's data-centre market, and its guidance assumes no data-centre sales there at all.

137
00:21:01.000 --> 00:21:08.344
<v Sam>So I'm stuck on one thing. If the moat is that deep, why does everyone still tell me to worry?

138
00:21:08.344 --> 00:21:21.195
<v Alex>Because the moat protects the product. When my engineers say they can't leave CUDA, that's a product fact. Nobody on my team can tell me what protects the financing, or the person at the top.

139
00:21:21.195 --> 00:21:48.000
<v Alex>I'll confess something from the board meeting. The first slide I showed wasn't about the scandal at all. The slide was about weight. At its September peak, Nvidia was close to eight percent of the entire S&amp;P 500, a concentration nobody has seen in a single stock since at least the seventies. By one estimate it accounted for something like a fifth of the index's gains in recent quarters, roughly double its weight.

140
00:21:48.000 --> 00:21:53.142
<v Sam>And on Thursday I did something I'd never done before. I opened my retirement account.

141
00:21:53.142 --> 00:21:53.828
<v Alex>Oh no.

142
00:21:53.828 --> 00:22:11.310
<v Sam>Oh yes. I've got an S&amp;P 500 fund in my workplace plan, the default I picked years ago and never looked at again. One of the finance sites ran the numbers this month, and Nvidia is about seven percent of that kind of fund now, after being over eight in September.

143
00:22:11.310 --> 00:22:14.738
<v Alex>So you own it whether you meant to or not.

144
00:22:14.738 --> 00:22:20.908
<v Sam>I own a huge slab of it and I never chose it. It chose me by getting bigger.

145
00:22:20.908 --> 00:22:29.478
<v Alex>Mine works the same way, and I never looked either. Weights follow value, so as Nvidia rose, every index fund quietly bought more of it.

146
00:22:29.478 --> 00:22:42.504
<v Sam>Which is what stopped me. If one company's boardroom crisis can move my retirement number before I've picked a single stock, then the question isn't Jensen's marriage. It's how many of us are holding one bet without knowing.

147
00:22:42.504 --> 00:22:45.247
<v Alex>Anyone with a pension or an index fund.

148
00:22:45.247 --> 00:23:00.672
<v Sam>And that's what stays with me. My team has a rule that nobody signs a vendor contract without asking what happens if the vendor's biggest customer disappears. I have never once asked that question about my own retirement, and it turns out I should have.

149
00:23:00.672 --> 00:23:05.814
<v Sam>And it isn't only the stock. You told me its chips sit underneath everyone's spending.

150
00:23:05.814 --> 00:23:20.897
<v Alex>Microsoft, Amazon, Google, Meta, and the labs, OpenAI and Anthropic among them. Analysts expect the five largest cloud companies to spend about one and a third trillion dollars on capital projects next year, and the bulk of the computing in those budgets is Nvidia's.

151
00:23:20.897 --> 00:23:33.238
<v Sam>So the stock is also the supplier. That changes it for me. A stock I don't hold on purpose I can ignore, but a supplier that all my vendors are all standing on isn't an option.

152
00:23:33.238 --> 00:23:51.406
<v Alex>And that's why a chip company's boss ends up on my risk register. When I mapped our workloads for the board, nearly every model we pay for traced back to one supplier. My board wanted to know how exposed our AI plans are to one chip supplier, now that the inquiry is open.

153
00:23:51.406 --> 00:23:58.262
<v Sam>What did you tell them? Because I'd want an answer that doesn't just start and end with the word CUDA.

154
00:23:58.262 --> 00:24:18.144
<v Alex>That the franchise outlasts any one person. CUDA, the developer base, the integrated platform, all of that stays in place under a new chief executive. Analysts are saying customers with multiyear commitments, the hyperscalers, the clouds, the AI developers, will look harder at single-supplier exposure. And Oracle has been adding rival accelerators, so there's at least a door.

155
00:24:18.144 --> 00:24:25.000
<v Sam>At least a door. That sounds confident, but a door nobody has used is a floor plan, not an exit.

156
00:24:25.000 --> 00:24:28.392
<v Alex>It's a door. I'm not claiming anyone has walked through it.

157
00:24:28.392 --> 00:24:35.484
<v Sam>I'd have been sweating in that meeting, and I'd have asked what it costs us if the answer turns out to be wrong.

158
00:24:35.484 --> 00:24:43.193
<v Alex>I was. But it's the right question, because the answer isn't about Jensen. It's about how many of our dependencies lead back to one place.

159
00:24:43.193 --> 00:24:49.360
<v Sam>Has anything been this big before? The word unprecedented gets thrown around a lot, and I distrust it on principle.

160
00:24:49.360 --> 00:25:00.152
<v Alex>Close. IBM was the largest company in the index through much of the eighties, just before personal computers wore down its grip. Before that, AT&amp;T and General Motors were a big slice of it together.

161
00:25:00.152 --> 00:25:05.394
<v Sam>And Cisco, which you have strong opinions about and have been waiting all episode to bring up.

162
00:25:05.394 --> 00:25:10.019
<v Alex>Back in oh-one I was a young engineer watching Cisco's stock like a football team.

163
00:25:10.019 --> 00:25:15.878
<v Sam>Every time you say that, somewhere a dial-up modem screams, and then you tell me about your first router.

164
00:25:15.878 --> 00:25:26.362
<v Alex>In March of 2000 Cisco briefly became the most valuable company in the world. It made the routers the internet boom was built on, and it traded at about two hundred times its earnings.

165
00:25:26.362 --> 00:25:33.762
<v Sam>Two hundred times earnings. And Nvidia's at thirteen, which sounds like a different planet to me, though you're telling me it's the same genre.

166
00:25:33.762 --> 00:25:44.555
<v Alex>Which is part of why I say panic discount. By the end of 2002 Cisco's shares had fallen about eighty-five percent, and they only got back to their 2000 price in December of last year.

167
00:25:44.555 --> 00:25:47.638
<v Sam>That's not a comfort. That's a quarter of a century.

168
00:25:47.638 --> 00:26:06.139
<v Alex>It is, and that is the other half of the lesson. Each of those giants was the toolmaker for a new infrastructure, and each time its weight reflected a bet on how long the build-out would last. I have bought into that kind of bet with a platform contract, so I know the feeling of asking how long it lasts.

169
00:26:06.139 --> 00:26:08.298
<v Sam>There's a leadership one too, isn't there?

170
00:26:08.298 --> 00:26:20.940
<v Alex>The closer precedent is Intel in 2018. Its chief executive Brian Krzanich resigned over a past consensual relationship with an employee, which broke Intel's rules on relationships with subordinates, and the board made its finance chief, Robert Swan, interim chief executive.

171
00:26:20.940 --> 00:26:23.715
<v Sam>Finance chief steadies the ship. Exactly what Kress is.

172
00:26:23.715 --> 00:26:38.516
<v Alex>Almost identical. I lived a version of it: a platform vendor lost its head of engineering, a roadmap slipped by a full release cycle, and my team spent a year planning around the gap. A leadership vacuum at a chip company is measured in product cycles, not quarters.

173
00:26:38.516 --> 00:26:49.000
<v Sam>So the real fear isn't November. It's a few product cycles from now, and nobody can see that far. I make plans a year at a time and I'd still call that stretching it.

174
00:26:49.000 --> 00:26:58.119
<v Alex>That's the part that keeps my engineers awake. A chip roadmap runs on a yearly rhythm, and if the next generation slips, our plans slip with it.

175
00:26:58.119 --> 00:27:04.536
<v Sam>All right, you've been saving this. The loop. Explain it like I'm the one who has to approve it.

176
00:27:04.536 --> 00:27:17.371
<v Alex>Nvidia's demand is mostly real. Hyperscalers, frontier labs and governments are spending hundreds of billions on data centres. But a growing share of the buying has been made easier by Nvidia itself, and the mechanism forms a circle.

177
00:27:17.371 --> 00:27:28.517
<v Sam>Let me try it in my world. It's as if my pet-food supplier lent me the money to buy more of its boxes, so my orders stay high and its quarter looks great.

178
00:27:28.517 --> 00:27:45.742
<v Alex>That's close, and it's the right shape. The supplier's cash funds stakes in, and guarantees for, the labs and clouds that buy its chips. Those customers buy more chips and sell AI services. Only if those services earn enough can they keep paying Nvidia, which refills the cash that started it.

179
00:27:45.742 --> 00:27:49.457
<v Sam>And the bit where my boxes go off in the cupboard?

180
00:27:49.457 --> 00:27:56.887
<v Alex>Your customers eat the food. Nvidia's customers have to turn chips into something other people pay for, and that can take years.

181
00:27:56.887 --> 00:28:01.278
<v Sam>So which part worries you? Because I can see five worrying things already.

182
00:28:01.278 --> 00:28:16.477
<v Alex>Two, really. The stakes first. Over a year ago Nvidia announced it would invest up to a hundred billion dollars in OpenAI as OpenAI deployed its systems. Huang later said that figure was never a commitment, and the cheque that followed was about thirty billion.

183
00:28:16.477 --> 00:28:21.543
<v Sam>That gap is the story. A hundred was the headline and thirty was the cheque.

184
00:28:21.543 --> 00:28:39.106
<v Alex>That is the part I'd defend. A supplier backing a customer it believes in is a bet, and bets get sized down. The guarantee is what I can't wave off. Credit support of up to a hundred and five billion dollars behind the leases on an OpenAI affiliate's planned site in Ohio.

185
00:28:39.106 --> 00:28:50.589
<v Sam>One building's rent, guaranteed by the company that sold the building its chips. When my agency tells me it will cover a vendor's invoice if the vendor flakes, I ask who covers the agency.

186
00:28:50.589 --> 00:29:05.450
<v Alex>Nvidia has also backstopped CoreWeave's unused capacity, so if CoreWeave can't fill its racks, Nvidia pays. And it organised platforms with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to raise more than five hundred billion dollars of outside capital for AI data centres.

187
00:29:05.450 --> 00:29:17.947
<v Sam>That I'd call a vendor getting its customers a bank. Fine on its own. Together it's a company that sells you the shovel, lends you the money for the shovel, and guarantees the lease on the field.

188
00:29:17.947 --> 00:29:22.000
<v Alex>When you say it like that I'm tempted to sell my stock.

189
00:29:22.000 --> 00:29:25.216
<v Sam>You don't own any, you own an expensive cloud contract.

190
00:29:25.216 --> 00:29:43.225
<v Alex>Let me argue the other side properly, because Huang has a point when he calls the word circular ridiculous. This is ordinary vendor finance. The money flows one way, into real data centres, and it's legal and disclosed. I've signed deals where a vendor's own financing arm closed the sale, and nobody called it a scandal.

191
00:29:43.225 --> 00:29:54.480
<v Sam>And I'd have been fine with that, until the vendor's financing arm was also its biggest customer's landlord. Where does it stop being finance and start being the thing that makes the numbers look good?

192
00:29:54.480 --> 00:30:02.841
<v Alex>The weak link is a link Nvidia can't control. A financed customer can keep buying chips only if its own customers pay it enough for AI.

193
00:30:02.841 --> 00:30:08.630
<v Sam>Which brings me back to what I'd say in every budget review. Okay, but who pays for that?

194
00:30:08.630 --> 00:30:10.881
<v Alex>That's the whole episode in five words.

195
00:30:10.881 --> 00:30:22.780
<v Sam>If the labs' revenue doesn't catch up with their spending, the guarantees stop being a sales tool and become a bill. And a bill like that doesn't land on a rounding error. It lands on whoever signed.

196
00:30:22.780 --> 00:30:33.070
<v Alex>We argued exactly this in episode forty-nine, the AI bubble one, back in late August. The exposure sits with the pure-model labs whose revenues have to catch up with what they've spent.

197
00:30:33.070 --> 00:30:35.322
<v Sam>And in episode seventy-eight, the Anthropic filing.

198
00:30:35.322 --> 00:30:42.718
<v Alex>That one showed the other side of the same bet. Compute commitments of nearly five hundred billion dollars that a lab can't cancel.

199
00:30:42.718 --> 00:31:02.656
<v Alex>I'll say the uncomfortable thing first. Nobody has credibly alleged any of this is illegal. Round-tripping, where companies inflate each other's revenue in secret, is illegal, and Nvidia discloses these commitments in its filings. I have signed contracts where the vendor's financing arm closed the deal, so I know where my own line sits: disclosed and priced is fine, disguised is not.

200
00:31:02.656 --> 00:31:04.907
<v Sam>Disguise how, if it's in the filing?

201
00:31:04.907 --> 00:31:12.304
<v Alex>A chip sale backed by Nvidia's own guarantee looks the same in the revenue line as one paid for by a self-funding customer.

202
00:31:12.304 --> 00:31:20.665
<v Sam>It's like me counting an autoship box as a sale after I paid the customer's first month. The line says sale. The cash says something else.

203
00:31:20.665 --> 00:31:22.916
<v Alex>Which is why investors have to dig.

204
00:31:22.916 --> 00:31:48.000
<v Sam>I read the footnotes, because that is what I do to anyone who sends me a deck with a number I like. The latest quarterly filing lists about a hundred and eight billion dollars of maximum guarantee exposure, and more than half of it is a single guarantee for a single OpenAI site. If one of my agencies had half its risk riding on one client filling one building, I would have a very different conversation with them.

205
00:31:48.000 --> 00:31:53.250
<v Alex>Hm, and more than half of that is one guarantee, for one OpenAI site.

206
00:31:53.250 --> 00:31:58.125
<v Sam>So it isn't diversified. It's a bet on one customer filling one building.

207
00:31:58.125 --> 00:32:12.750
<v Alex>It's… okay, I'm going to say it. Lucent, in the dot-com years. Vendor-financing commitments about a quarter of revenue, much of it to start-up carriers. One of them, Winstar, took two billion dollars from Lucent and later went bankrupt.

208
00:32:12.750 --> 00:32:15.750
<v Sam>You have waited this entire episode for that.

209
00:32:15.750 --> 00:32:25.500
<v Alex>Since the first minute. Nvidia's guarantees alone are a bit over a quarter of its annual run rate, so the shape is similar. Similar, not identical.

210
00:32:25.500 --> 00:32:42.750
<v Sam>And that's why I return to the unpaid bills. They're the number my CFO would circle. Over six months, Nvidia's unpaid customer bills rose from the high thirty billions to over sixty billion, faster than its sales, and five direct customers owe the bulk of them.

211
00:32:42.750 --> 00:32:48.000
<v Alex>I'll give you that. It's the part of the bull case I like least.

212
00:32:48.000 --> 00:32:54.750
<v Sam>Customers taking longer to pay, and the people owing the money are five names. Write that down somewhere.

213
00:32:54.750 --> 00:33:00.375
<v Alex>I'm writing it. My panic-discount line is a little less comfortable than an hour ago.

214
00:33:00.375 --> 00:33:06.000
<v Sam>There's another grey area, and it's the one that makes my skin crawl. The pressure.

215
00:33:06.000 --> 00:33:14.625
<v Alex>The Justice Department has been investigating Nvidia's sales practices since 2024, after rivals complained it threatened to punish customers who bought competitors' chips.

216
00:33:14.625 --> 00:33:20.250
<v Sam>I know that move. Vendors do it to my agencies in the nicest possible language.

217
00:33:20.250 --> 00:33:31.125
<v Alex>Investigators have reportedly asked whether customers that buy Nvidia's networking equipment, complete systems or Groq-based processors get bigger chip allocations. They've also questioned its proposed purchase of Hugging Face.

218
00:33:31.125 --> 00:33:33.000
<v Sam>And what does Nvidia say?

219
00:33:33.000 --> 00:33:42.750
<v Alex>That it allocates scarce chips to customers who can deploy them fastest, and that its dominance reflects product quality. None of it has produced a charge.

220
00:33:42.750 --> 00:33:51.750
<v Sam>But you can feel the pressure in the yearly product cadence, can't you? Skip an upgrade and you fall behind the rival who didn't.

221
00:33:51.750 --> 00:34:05.250
<v Alex>My hardware refresh planning feels it. Skip a generation and my team is renting yesterday's speed while a competitor ships on today's. That's why the investor Michael Burry framed Huang's exit as Nvidia losing its enforcer.

222
00:34:05.250 --> 00:34:07.125
<v Sam>Enforcer. That's a mafia word.

223
00:34:07.125 --> 00:34:13.875
<v Alex>It's Burry's word, and he has a view to sell. I'd say it explains why people are asking.

224
00:34:13.875 --> 00:34:34.500
<v Alex>Which brings us to the number you came here to fight about. What the market is saying today is that investors believe Nvidia's current profits and doubt they'll last. Revenue in the quarter to late July was about ninety-six billion dollars, up a hundred and six percent on the year, almost all from data centres.

225
00:34:34.500 --> 00:34:39.000
<v Sam>And the stock? Where does it actually sit against its own history?

226
00:34:39.000 --> 00:34:47.843
<v Alex>Around thirty percent below its high of the past year, with the company worth under four trillion. That's about thirteen times the earnings analysts expect next year.

227
00:34:47.843 --> 00:34:52.428
<v Sam>And you call that cheap, which is a word I'd normally keep for sofas.

228
00:34:52.428 --> 00:35:07.493
<v Alex>Cheaper than AMD, and cheaper than Broadcom, and nothing like Cisco in 2000. A multiple that low is the market pricing in a decline, not a boom. It's discounting a slowdown in AI spending, the financing exposure, the succession and the political risk, all at once.

229
00:35:07.493 --> 00:35:14.044
<v Sam>Which is what I'm saying. A price that low is the market telling you something, not the market being silly.

230
00:35:14.044 --> 00:35:22.231
<v Alex>Which is where we disagree. You hear a market that read the footnotes. I hear a market that threw out the baby with the bathwater.

231
00:35:22.231 --> 00:35:28.127
<v Sam>Analysts are split, aren't they? They always seem to be when a stock is down and everyone's embarrassed.

232
00:35:28.127 --> 00:35:34.022
<v Alex>Evercore cut its price target by more than half. The average target is still well above today's price.

233
00:35:34.022 --> 00:35:40.572
<v Sam>So even the people paid to guess are miles apart, which is oddly comforting. At least it isn't only me.

234
00:35:40.572 --> 00:35:49.087
<v Alex>Which is why I distrust my certainty today. I called a few things wrong back in my start-up days, and I'm still paying for the confidence.

235
00:35:49.087 --> 00:35:51.380
<v Sam>So what happens next? Give me dates.

236
00:35:51.380 --> 00:35:58.585
<v Alex>Two dates and one test. On the seventeenth, Nvidia reports results. A strong quarter with steady guidance would show customers didn't flinch.

237
00:35:58.585 --> 00:36:04.153
<v Sam>That's days away, which means this recording will be out of date almost the moment it's published.

238
00:36:04.153 --> 00:36:12.341
<v Alex>Days. Then the choice of a permanent chief executive. Kress or an outsider will tell us whether the board wants a steward or a strategist.

239
00:36:12.341 --> 00:36:19.218
<v Sam>And the test? Because a date I can put in a diary, but a test sounds like something that takes years.

240
00:36:19.218 --> 00:36:31.664
<v Alex>The loop. If AI applications earn enough for the companies Nvidia has financed to keep paying, today's price looks like a panic discount. If they don't, the guarantees turn into losses, and the Cisco comparison stops being rhetorical.

241
00:36:31.664 --> 00:36:40.507
<v Sam>Give me the bull case in one breath, and then I'll want the bear case just as fast, because I trust nobody who only gives me one.

242
00:36:40.507 --> 00:36:54.262
<v Alex>Management has pointed to revenue growth of about seventy percent in the fiscal year that starts in February. And each gigawatt of data centre built on its next platform, Vera Rubin, is reported to carry far more Nvidia content than Blackwell did.

243
00:36:54.262 --> 00:36:59.175
<v Sam>And now the bear case, the one my CFO would read out in the meeting.

244
00:36:59.175 --> 00:37:09.000
<v Alex>Gross margin expected to dip as memory costs rise, nearly fifty billion dollars invested in frontier labs whose valuations depend on the same boom, and customers taking longer to pay.

245
00:37:09.000 --> 00:37:23.227
<v Sam>The bear case is every bit as concrete as the bull one, and that's what I'd tell my CFO if she asked whether to pull a tool renewal forward. I'd say we can't read the price, so we don't bet our renewal on it.

246
00:37:23.227 --> 00:37:26.460
<v Alex>And I'd be a hypocrite if I only quoted one.

247
00:37:26.460 --> 00:37:48.771
<v Alex>Let me pull the thread out to the whole economy, and then back to us. For the economy, it's a wealth shock in every index fund, Sam's included, and a line on my risk register I now have to explain. For politics, the affair has become a national-security fight. Senators want Huang and Commerce Secretary Howard Lutnick to testify, and China policy and chip exports are back under scrutiny.

248
00:37:48.771 --> 00:37:51.358
<v Sam>And the White House owns part of it.

249
00:37:51.358 --> 00:38:16.901
<v Alex>Because of the shifting comments. And here is where it lands on my desk. OpenAI, Anthropic and the AI clouds have relied on a supplier willing to finance them, so a new chief executive under scrutiny, more careful with guarantees and stakes, means capital gets scarcer for every lab that cannot fund its own compute. That is the lever I would watch at my cloud contract renewal: the price I am quoted is built on those labs staying funded.

250
00:38:16.901 --> 00:38:35.009
<v Sam>That's what scares me. Not the headline. The quiet bit where a lab slows down because its supplier stopped being generous. For me it shows up as a price change on a tool renewal, months later, with nobody connecting it to Nvidia. I've started asking my vendors whose cloud they run on, and they hate it.

251
00:38:35.009 --> 00:38:48.912
<v Alex>And the cost lands on the young first. Diego's first loan payment came out this week, and the video interview he did on Tuesday was screened by software. He hasn't heard back, and I don't have a clever thing to say to him.

252
00:38:48.912 --> 00:38:50.852
<v Sam>I'm sorry. That's a hard week.

253
00:38:50.852 --> 00:38:54.086
<v Alex>He will be fine. I have decided to believe that.

254
00:38:54.086 --> 00:39:09.283
<v Sam>There's a trust question under this as well. A Pew survey last month found Americans are open to AI in healthcare but want humans making the final calls. About half say it will make the patient and doctor relationship worse, and only one in ten says better.

255
00:39:09.283 --> 00:39:24.480
<v Alex>Elena would nod at that. She works emergency, and she'll take a note-taking tool that saves her an hour, but she won't accept a machine making the final call. And it matters here, because the loop only closes if people trust AI enough to pay for it.

256
00:39:24.480 --> 00:39:27.390
<v Sam>So the demand for chips depends on strangers' comfort.

257
00:39:27.390 --> 00:39:40.000
<v Alex>Down the chain, yes, and it lands on my budget line. The AI spend the directors fund only pays back if our customers let a bot near the decisions they care about, a refund, a size exchange, a complaint.

258
00:39:40.000 --> 00:39:42.216
<v Sam>So who moved? We said we'd say.

259
00:39:42.216 --> 00:39:53.929
<v Alex>You moved me on the receivables. Bills rising faster than sales, with five customers owing the bulk of them, I can't wave away. It weakens my panic-discount line, because a panic discount assumes the footing is sound.

260
00:39:53.929 --> 00:40:16.088
<v Sam>And I will admit what moved me, which is the lock-in. Your point-of-sale story landed. I came in assuming a customer could walk, and if ripping out CUDA is like ripping out a store system, the revenue base is stickier than I assumed. I have not moved on the price. I still call it fair, not panic, because of the unpaid bills, but I will not call the business weak.

261
00:40:16.088 --> 00:40:27.801
<v Alex>So my split is that I think the stock is priced for more trouble than the numbers show. But I'm doing the supplier-exposure review for the directors anyway, because I'd rather own that decision than inherit it.

262
00:40:27.801 --> 00:40:38.247
<v Sam>And mine is that I'm going to look at that S&amp;P fund I picked years ago and decide whether I meant to hold that much of one company. Not to sell. To choose.

263
00:40:38.247 --> 00:40:43.312
<v Alex>Choosing is what I'd want everyone to do. That's the thing I'd put on the record.

264
00:40:43.312 --> 00:40:47.111
<v Sam>Careful. Last time you said that we had to read it back.

265
00:40:47.111 --> 00:40:49.010
<v Alex>Fine. Nobody's reading this one back.

266
00:40:49.010 --> 00:40:58.507
<v Sam>Let's pull it together. Jensen Huang resigned after admitting an affair, with further relationships reported inside his own company. The spy claim is unproven, and contested inside the US government.

267
00:40:58.507 --> 00:41:05.788
<v Alex>A divorce would split the money, not the control, because Nvidia has one class of shares and he has given up both seats.

268
00:41:05.788 --> 00:41:14.652
<v Sam>Nvidia got so big that its leadership became a macroeconomic variable, and its weight in your index fund means you hold it whether you chose it or not.

269
00:41:14.652 --> 00:41:23.832
<v Alex>And the deeper story is the loop. Nvidia's own cash backs the customers who buy its chips, and the weak link is whether those customers earn enough to pay.

270
00:41:23.832 --> 00:41:35.545
<v Sam>Thank you for listening. I hope you came away seeing a bit more clearly where all of this is heading, because it's a complicated, fast-moving picture, and the horizon on what we can know is very short.

271
00:41:35.545 --> 00:41:51.373
<v Alex>And one honest note on how this is made. It's AI-generated. Dan builds a custom stack of AI tools to research, analyse, verify and illustrate the questions worth understanding, mostly to learn them himself, and publishes it for anyone who'd like to follow along. AI-assisted, fact-checked, worth a second look.

272
00:41:51.373 --> 00:42:11.000
<v Sam>Before we go, one thing that genuinely helps. Follow the show. Whatever app you're listening in, there's a follow or a plus button. It's one tap, it's free, and you'll get each new episode as it lands. For a small independent show, a follow is the single biggest thing that helps it reach other people trying to make sense of all this.

273
00:42:11.000 --> 00:42:24.217
<v Alex>And one more. If there's something in here you'd push back on, or a thread you want us to pull harder next time, write to us at podcast@connectiveshift.com. What you send genuinely shapes what we dig into next.

274
00:42:24.217 --> 00:42:25.609
<v Sam>See you next time.

275
00:42:25.609 --> 00:42:27.000
<v Alex>See you next time.
