1
00:00:00,017 --> 00:00:02,727
Well, that's a nail in the coffin for Jensen Huang. Game over.

2
00:00:02,727 --> 00:00:08,666
The CEO of NVIDIA was caught having another affair with an executive that directly reports to him.

3
00:00:08,991 --> 00:00:13,565
And this time he was caught on camera on video by someone filming nearby.

4
00:00:13,879 --> 00:00:17,707
The allegations now are he had an affair with a Chinese spy,

5
00:00:17,707 --> 00:00:22,987
allegedly, and now with an executive on the brink of being ousted now by his board.

6
00:00:22,987 --> 00:00:28,531
It's going to be decided over the next 24 hours. This is easily the worst week,

7
00:00:28,693 --> 00:00:32,025
I would say, of Nvidia's and Jensen Huang's life.

8
00:00:32,339 --> 00:00:37,507
And it's shedded now around $2 trillion, maybe just over $2 trillion worth of

9
00:00:37,507 --> 00:00:41,207
Nvidia's market cap. Now, I wish that was the only additional news when it comes

10
00:00:41,207 --> 00:00:42,979
to the Jensen spy scandal.

11
00:00:43,252 --> 00:00:47,547
But there's more, unfortunately. One, about a year ago, when the investigation

12
00:00:47,547 --> 00:00:50,659
started into his alleged affair with the Chinese spy.

13
00:00:51,124 --> 00:00:55,077
One of his legal counsel reached out to the governing authority that was leading

14
00:00:55,077 --> 00:00:57,837
the investigation to push back and say, hey, you need to stop this,

15
00:00:57,837 --> 00:01:00,874
effectively having interference in the investigation.

16
00:01:01,223 --> 00:01:06,127
And third and final bit of damning evidence is that the initial damage was expected

17
00:01:06,127 --> 00:01:09,773
to be around $2 billion sold to China worth of chips.

18
00:01:10,122 --> 00:01:13,247
It ended up being a lot higher, almost three times, or just over three times,

19
00:01:13,247 --> 00:01:19,477
at $7.5 billion, which happened to be more chips than he had sold to Elon Musk,

20
00:01:19,477 --> 00:01:21,207
one of the biggest advocates in America.

21
00:01:21,207 --> 00:01:25,220
So all the statements and claims that Jensen Huang had made at the time where

22
00:01:25,458 --> 00:01:30,007
he wanted America to stay number one and stay ahead of China ended up actually

23
00:01:30,007 --> 00:01:33,027
not being put into practicality when it came to selling chips.

24
00:01:33,027 --> 00:01:38,537
He was selling more to China. So all of this to say, this puts NVIDIA one of the, if not the.

25
00:01:38,977 --> 00:01:42,887
Biggest company in the AI economy, which effectively is the rest of the economy

26
00:01:42,887 --> 00:01:45,797
because the majority of the stock market relies on all the AI investment that

27
00:01:45,797 --> 00:01:48,318
is happening right now in jeopardy.

28
00:01:48,527 --> 00:01:51,987
And it looks like Jensen Huang will be ousted at the time of recording.

29
00:01:51,987 --> 00:01:54,896
And this is literally breaking news. The news broke about an hour ago.

30
00:01:55,122 --> 00:01:56,827
The Wall Street Journal is reporting that

31
00:01:57,234 --> 00:02:00,247
The board is currently deciding to oust him as early as tomorrow morning.

32
00:02:00,247 --> 00:02:02,354
So by the time you're watching it, it might have already happened.

33
00:02:02,592 --> 00:02:06,367
In fact, I already tried to record this episode about three hours ago,

34
00:02:06,367 --> 00:02:11,746
and then the news of this broke, and I had to add it in. And I won't lie, this is a sad day for AI.

35
00:02:12,071 --> 00:02:16,147
Jensen Huang has, for all intents and purposes, lived the American dream when

36
00:02:16,147 --> 00:02:19,717
he birthed the idea of NVIDIA over three decades ago in a Denny's,

37
00:02:19,717 --> 00:02:23,507
which is a fast, casual restaurant that serves pancakes. He was a dishwasher

38
00:02:23,507 --> 00:02:26,119
there. If you literally looked at his LinkedIn, that was the case.

39
00:02:26,497 --> 00:02:32,107
And he worked up on this idea to create NVIDIA into the most valuable company in the world.

40
00:02:32,107 --> 00:02:36,587
And the stock is down over 30 to 40%. So today I wanna unpack all the recent

41
00:02:36,587 --> 00:02:39,610
evidence that is pretty damning for Jensen's position,

42
00:02:39,929 --> 00:02:43,457
what this means for NVIDIA, who's likely to succeed him and what this means

43
00:02:43,457 --> 00:02:46,797
for the rest of the AI economy and how it's gonna play out over the next six,

44
00:02:46,797 --> 00:02:50,677
12 months because a lot of CapEx, a lot of investments, a lot of Anthropics

45
00:02:50,677 --> 00:02:53,107
revenues, upcoming IPOs, OpenAI, the same,

46
00:02:53,345 --> 00:02:57,047
relies on this one singular company, on this one singular individual.

47
00:02:57,047 --> 00:03:00,307
What is NVIDIA worth without Jensen Huang? We're going to get into all of this

48
00:03:00,307 --> 00:03:02,573
and more on today's roundup brought to you by Qualcomm.

49
00:03:06,521 --> 00:03:09,967
Okay, so the New York Times, who originally, I believe, were the ones that broke

50
00:03:09,967 --> 00:03:13,858
the story of him having his first affair with an alleged Chinese spy,

51
00:03:14,143 --> 00:03:18,507
Alice Huang, were also the ones that just broke their story about his second affair.

52
00:03:18,507 --> 00:03:22,707
And this time, it's with an NVIDIA executive that's been at the company for a while.

53
00:03:22,707 --> 00:03:26,737
At this point, I think just over a decade, used to be the executive or senior

54
00:03:27,097 --> 00:03:30,857
head of business development and then got promoted to report directly to Jensen

55
00:03:30,857 --> 00:03:34,987
to be the VP of healthcare. Kimberly Powell is meant to be the one that he's

56
00:03:34,987 --> 00:03:36,322
been in this affair with.

57
00:03:36,694 --> 00:03:40,577
And it's worse because it was caught on camera by a bystander that caught them

58
00:03:40,577 --> 00:03:43,334
being intimate at some restaurant in San Jose.

59
00:03:43,630 --> 00:03:45,842
And Jensen's security spotted this,

60
00:03:46,360 --> 00:03:50,302
approached the man or the individual and basically got them to sign an NDA.

61
00:03:50,722 --> 00:03:54,842
In exchange for what, I have no idea. But the truth is, it's just not looking great for Jensen Huang.

62
00:03:54,842 --> 00:04:00,182
And this is a shame because I have been and still am in many ways from a professional

63
00:04:00,182 --> 00:04:04,692
aspect, a huge fan of what Jensen has done for the entire industry and how he's built it.

64
00:04:04,692 --> 00:04:07,332
He has identified some of the most impressive opportunities.

65
00:04:07,332 --> 00:04:10,153
And you could argue actually that modern day AI, LLMs,

66
00:04:10,490 --> 00:04:15,046
Anthropic Open Air wouldn't actually be here today if Jensen Huang hadn't bet

67
00:04:15,360 --> 00:04:20,800
the house NVIDIA, GPUs, et cetera, on such a burgeoning industry and sector

68
00:04:21,066 --> 00:04:23,482
back in the day when no one really paid any attention?

69
00:04:23,482 --> 00:04:26,239
Now, if he's asked two questions arise in my mind. Number one,

70
00:04:26,470 --> 00:04:30,182
what happens to NVIDIA as a company? And two, who's going to replace him?

71
00:04:30,182 --> 00:04:33,652
Now, on that second question, I honestly don't have a good answer.

72
00:04:33,652 --> 00:04:37,006
I mean, if I ask you the question, is there anyone obvious in NVIDIA that would replace him?

73
00:04:37,350 --> 00:04:41,102
I don't know if someone naturally would come to mind. But there are some candidates

74
00:04:41,102 --> 00:04:45,580
being considered, one being the head of AWS Cloud Services, Matt Garman, who

75
00:04:45,819 --> 00:04:49,522
has had, or likely one of the very few individuals that has had experience spending

76
00:04:49,522 --> 00:04:52,862
hundreds of billions of dollars on CapEx, especially when it comes to GPUs,

77
00:04:52,862 --> 00:04:53,916
he would need to be poached.

78
00:04:54,206 --> 00:04:58,112
And then the other guy is Thomas Kurian, who does a very similar job at Google Cloud.

79
00:04:58,112 --> 00:05:01,942
These are two infrastructure scaling powers that could potentially,

80
00:05:01,942 --> 00:05:05,632
maybe the sum of them, fulfill or fit the shoes that Jensen had.

81
00:05:05,632 --> 00:05:09,942
But I can't help but think that Jensen has been with the company from the start

82
00:05:09,942 --> 00:05:12,527
and knows the intricacies of how that company is built. In fact,

83
00:05:12,747 --> 00:05:16,772
you could argue that the layout of authority at that company has been perfectly

84
00:05:16,772 --> 00:05:19,724
suited for Jensen Huang. So maybe you need someone completely different.

85
00:05:20,002 --> 00:05:22,446
Now, an alternative view to the replacement is...

86
00:05:22,801 --> 00:05:28,364
To remove Jensen as CEO, but keep him on the board or keep him as chairman so

87
00:05:28,364 --> 00:05:32,284
that he can have some form of influence on the strategy and ideation and goals

88
00:05:32,284 --> 00:05:35,534
and direction as where NVIDIA ultimately ends up headed.

89
00:05:35,534 --> 00:05:38,784
But there are also two other bits of evidence which are pretty damning for Jensen

90
00:05:38,784 --> 00:05:42,945
Huang. The first being the number of chips that he sold to China.

91
00:05:43,189 --> 00:05:48,024
So in the original instance, it was estimated to be around $2 billion to this

92
00:05:48,024 --> 00:05:53,034
company Megaspeed, which Alice Huang, the alleged Chinese spy, was the executive of.

93
00:05:53,318 --> 00:05:55,791
And so that was the number that was attached to it.

94
00:05:56,087 --> 00:06:00,424
Turns out that number is actually over three times larger at $7.5 billion.

95
00:06:01,084 --> 00:06:04,905
And the reason why this is the case is because it wasn't just chips that were sold to Megaspeed.

96
00:06:05,178 --> 00:06:09,244
It was chips that were sold to other Southeast Asian satellite companies in

97
00:06:09,244 --> 00:06:13,424
the likes of Malaysia and Singapore, which were all fronts for basically the

98
00:06:13,424 --> 00:06:17,141
People's Republic of China. And that was discovered more recently.

99
00:06:17,373 --> 00:06:21,454
So the idea here is Jensen Huang was selling chips to China.

100
00:06:21,454 --> 00:06:23,387
Now the big question is, did he know it?

101
00:06:23,689 --> 00:06:26,054
And that brings me to the third and final bit of damning evidence,

102
00:06:26,054 --> 00:06:28,174
which is he might actually have.

103
00:06:28,174 --> 00:06:33,214
Because rewind just about a year ago, the investigations were already being

104
00:06:33,214 --> 00:06:37,234
underway for his relationship with Alice Huang, and it was being hosted by the

105
00:06:37,234 --> 00:06:39,287
Commerce Department of the US.

106
00:06:39,652 --> 00:06:44,344
But at the time, NVIDIA's head of legal counsel reached out when he found out

107
00:06:44,344 --> 00:06:46,915
about the investigation and said, you need to shut this down.

108
00:06:47,263 --> 00:06:51,054
And so the Commerce Department shut it down. But what they didn't tell NVIDIA

109
00:06:51,054 --> 00:06:54,984
and Jensen Huang is that they shifted the case to the Justice Department,

110
00:06:54,984 --> 00:06:58,544
which continued on the investigation and doubled down, especially because there

111
00:06:58,544 --> 00:07:00,092
was a sign of interference.

112
00:07:00,272 --> 00:07:04,742
I'm sitting back and I'm thinking, who directed the head of NVIDIA's legal counsel to reach out?

113
00:07:05,299 --> 00:07:09,004
Everyone reports directly to Jensen Huang. He has over 50 direct reports from

114
00:07:09,004 --> 00:07:12,321
execs. It's a very flat kind of structure. He has these conversations and open.

115
00:07:12,715 --> 00:07:16,544
So I have to say that it's probably a top-down decision and it probably came

116
00:07:16,544 --> 00:07:21,024
from Jensen Huang. So he was probably aware that things were going on and yet

117
00:07:21,024 --> 00:07:25,028
he knowingly still continued selling chips to some of these satellite companies.

118
00:07:25,405 --> 00:07:28,664
I don't want to say things for sure. The investigation is still underway and

119
00:07:28,664 --> 00:07:31,904
I'm rooting and hoping that there's still some good news that comes out of it,

120
00:07:31,904 --> 00:07:33,924
but it doesn't look great for Jensen.

121
00:07:33,924 --> 00:07:38,234
So I now think back to all the countless times over the last year that Jensen

122
00:07:38,234 --> 00:07:42,031
Huang has advocated so strongly for selling chips to China.

123
00:07:42,373 --> 00:07:45,994
And I question what his incentives and motives were, the most prominent one

124
00:07:45,994 --> 00:07:50,117
being his argument with Duakesh on his episode about six months ago,

125
00:07:50,372 --> 00:07:52,613
where Duakesh is pushing him and saying like, hey, it's like.

126
00:07:53,182 --> 00:07:57,547
Imagine you're in a nuclear war and you're selling like enriched uranium to your competitors.

127
00:07:57,547 --> 00:08:01,107
Just because the U.S. is in the lead, why would you sell them enriched uranium?

128
00:08:01,107 --> 00:08:04,407
And Chen Zetong would just hit back saying, it's not the same thing.

129
00:08:04,407 --> 00:08:07,677
Like if we don't sell chips to China, China will still make chips.

130
00:08:07,677 --> 00:08:11,627
And there was kind of like a circular weird end to that argument.

131
00:08:11,627 --> 00:08:15,414
And there wasn't really any real conclusion. And different people took different sides.

132
00:08:15,704 --> 00:08:19,267
And now in hindsight, it looks like he was just kind of defending his placement,

133
00:08:19,267 --> 00:08:24,007
either because he was very naively influenced by a spy or because he knew exactly

134
00:08:24,007 --> 00:08:26,896
what he was doing, but he wanted to double down and make money from it anyway.

135
00:08:27,111 --> 00:08:31,687
But personally, I look at Jensen Huang and I say, he is a fantastic builder.

136
00:08:31,687 --> 00:08:35,070
He, for all intents and purposes, built NVIDIA from the ground up.

137
00:08:35,447 --> 00:08:38,237
And he is the main reason why NVIDIA has grown into such a big company.

138
00:08:38,237 --> 00:08:39,267
I mentioned this earlier.

139
00:08:39,267 --> 00:08:43,835
When you think about the entire AI CapEx bubble, when you think about the entire AI bubble at all,

140
00:08:44,091 --> 00:08:48,073
Anthropics revenue, the demand, the capacity to be able to train new models,

141
00:08:48,328 --> 00:08:52,137
the revenue projections that Anthropics IPO has to inference these models and

142
00:08:52,137 --> 00:08:56,227
to continue selling and making money all depends on this one company,

143
00:08:56,227 --> 00:08:58,357
on this one individual who is making bets.

144
00:08:58,357 --> 00:09:02,517
Remember, NVIDIA and Jensen Huang are the reason that a lot of the financing

145
00:09:02,517 --> 00:09:06,797
has been able to happen because traditional credit financing couldn't fund a

146
00:09:06,797 --> 00:09:12,778
lot of the loans that these neoclouds or hyperscalers required in order to fund the CapEx purchases.

147
00:09:13,010 --> 00:09:16,837
But if he's replaced, if he's ousted and replaced by someone else that is less

148
00:09:16,837 --> 00:09:19,802
risk prone or has a different tactics or strategy your view of the world,

149
00:09:19,976 --> 00:09:22,408
that means that the financial.

150
00:09:23,080 --> 00:09:26,888
Freedom that Jensen Wang and NVIDIA has offered may be pulled back.

151
00:09:27,132 --> 00:09:30,273
It may be restrained. And that might result in the bubble popping.

152
00:09:30,273 --> 00:09:32,635
It might result in the collapse of this entire thing.

153
00:09:33,053 --> 00:09:37,113
And I guess that's my main takeaway from this whole thing. I don't really quite

154
00:09:37,113 --> 00:09:40,077
know where all of this stands. A lot of companies rely on NVIDIA.

155
00:09:40,234 --> 00:09:43,483
If I had to guess, I think the company is going to be completely fine,

156
00:09:43,483 --> 00:09:46,028
not worried too much about the company over the long term.

157
00:09:46,318 --> 00:09:49,713
But over the near term, there's going to be insane volatility because there's

158
00:09:49,713 --> 00:09:53,723
so many deals, There's so much reliance on NVIDIA's GPUs being effectively delivered.

159
00:09:53,723 --> 00:09:55,495
All of this is going to distract away from that.

160
00:09:55,728 --> 00:09:58,526
It's going to cause people to churn and quit and lose trust.

161
00:09:59,019 --> 00:10:02,773
I can just see this impairing the ability for the entire AI movement.

162
00:10:02,773 --> 00:10:06,753
You see, the story isn't just about NVIDIA or Jensen Huang or his affairs.

163
00:10:06,753 --> 00:10:08,050
It's not about the gossip column.

164
00:10:08,230 --> 00:10:12,648
It's about how this affects the entire AI economy. From a competitive aspect,

165
00:10:12,979 --> 00:10:19,843
I can see reputationally some AI labs potentially wanting to move away from NVIDIA chips.

166
00:10:19,843 --> 00:10:24,723
And this is like a hot take from me to maybe try and use other types of competitor products.

167
00:10:24,723 --> 00:10:28,903
Now, we're already seeing a major shift to Google. I'm sure Google is smacking

168
00:10:28,903 --> 00:10:32,343
their lips at all of this drama because they just released two brand new TPU

169
00:10:32,343 --> 00:10:34,986
products, TPU being the main competitors to GPUs.

170
00:10:35,282 --> 00:10:37,483
So they're probably looking at this and thinking, yeah, okay,

171
00:10:37,483 --> 00:10:42,271
cool. I will ramp up TPU production and eat into NVIDIA's 90% monopoly.

172
00:10:42,579 --> 00:10:45,943
AMD is probably doing the same thing. Anthropic signed a major deal with AMD

173
00:10:45,943 --> 00:10:48,262
buying loads of their GPUs.

174
00:10:48,505 --> 00:10:51,543
OpenAire is doing the same with AMD and with Google as well.

175
00:10:51,543 --> 00:10:53,233
So I can see NVIDIA losing their

176
00:10:53,233 --> 00:10:56,683
market share on the GPU market just from this one instance and event.

177
00:10:56,683 --> 00:11:00,423
And of course, we'll have to see how things play out. I think in the worst case

178
00:11:00,423 --> 00:11:03,693
scenario, if he does get replaced, there will be some churn,

179
00:11:03,693 --> 00:11:05,973
there'll be some movement, there'll need to be some rebuilding of trust with

180
00:11:05,973 --> 00:11:08,653
whoever the new person is, but they'll end up figuring it out.

181
00:11:08,653 --> 00:11:12,688
Again, I'm an optimist for AI in general. And the reason why I'm so rocked about this isn't by.

182
00:11:13,251 --> 00:11:16,293
Necessarily the gossip and the affairs, which are bad in itself,

183
00:11:16,293 --> 00:11:18,023
but how it affects the whole AI economy.

184
00:11:18,353 --> 00:11:23,893
Now, speaking of the AI economy, there could be a worse timing for one of the

185
00:11:23,893 --> 00:11:30,021
biggest, if not the biggest AI IPO coming up rumored to be next week, which is Anthropics.

186
00:11:30,213 --> 00:11:34,178
And Anthropics IPO has, thankfully, a bunch of good news around it.

187
00:11:34,445 --> 00:11:41,190
Anthropics IPO is around the corner. And the valuation, as of two days ago,

188
00:11:41,515 --> 00:11:45,693
was estimated to be around $1 to $1.3 trillion.

189
00:11:46,353 --> 00:11:50,519
And the new story here was that they started their IPO roadshow officially.

190
00:11:50,873 --> 00:11:56,260
A few headline stats is they're going to issue around, I think it's 825 million shares,

191
00:11:56,579 --> 00:11:59,833
with the bulk of that coming from employees, mainly the founders,

192
00:11:59,833 --> 00:12:05,193
who are all going to sell, or each of them going to sell around 3.3 million

193
00:12:05,193 --> 00:12:06,585
shares, which is roughly around

194
00:12:06,956 --> 00:12:13,789
$360 million at the top end of the price range if they were to sell each share around $78.

195
00:12:14,178 --> 00:12:19,353
That is a hell of a payday for each founder. But they're also giving their underwriters,

196
00:12:19,353 --> 00:12:23,373
so the banks, JP Morgan and Chase, and I believe it is Goldman Sachs,

197
00:12:23,750 --> 00:12:26,513
the ability to buy another $10 billion.

198
00:12:26,513 --> 00:12:31,133
So the initial raise or what they're planning to raise is around $64 billion.

199
00:12:31,709 --> 00:12:34,173
And then they're giving banks an additional $10 billion.

200
00:12:34,826 --> 00:12:40,643
But there's also a secret side deal, $26 billion allocated to seven companies,

201
00:12:40,643 --> 00:12:42,825
which includes the likes of NVIDIA,

202
00:12:43,349 --> 00:12:46,717
AMD, Micron, Google. Now, these

203
00:12:46,717 --> 00:12:51,592
are all companies that have extreme stakeholder interest in Anthropic.

204
00:12:51,929 --> 00:12:55,998
And ironically, are the companies that Anthropic are already buying from.

205
00:12:56,248 --> 00:12:59,417
So, Anthropic has a huge CapEx commitment, right?

206
00:12:59,417 --> 00:13:03,987
It's around $900 billion, almost a trillion dollars worth of CapEx spend that

207
00:13:03,987 --> 00:13:07,767
they need to spend over the next couple of years, with around,

208
00:13:07,767 --> 00:13:12,367
I think, $70 billion of that being committed to over the next 12 months.

209
00:13:12,367 --> 00:13:16,367
Now, Anthropic, if you look at their cash balance sheet, is looking pretty reasonable.

210
00:13:16,367 --> 00:13:18,117
Right now, it's around $130 billion.

211
00:13:18,517 --> 00:13:21,957
And after this raise, it would put them at around $200 billion.

212
00:13:22,572 --> 00:13:24,981
So you could argue that they have enough money to pay for these things.

213
00:13:25,202 --> 00:13:30,197
But it is sort of weird that the companies who they're paying money to are now

214
00:13:30,197 --> 00:13:32,847
reinvesting money back into Anthropic.

215
00:13:32,847 --> 00:13:36,447
Now, actually, when I sit back and look at this, I think, okay, well, that makes sense.

216
00:13:36,447 --> 00:13:40,807
If these companies are incredibly bullish in Anthropic, and they are one of

217
00:13:40,807 --> 00:13:47,357
the leading frontier labs, certainly by revenue, they recently cracked around 80 billion worth of ARR

218
00:13:47,677 --> 00:13:52,067
over the last month, they jumped another $10 billion, just insane rapid progression,

219
00:13:52,433 --> 00:13:53,917
you'd probably want to double down.

220
00:13:53,917 --> 00:13:56,737
And listen, Google already is a major stakeholder in Anthropic,

221
00:13:56,737 --> 00:14:02,117
I think they own around 17% of the company, they sell almost all of their TPUs

222
00:14:02,117 --> 00:14:04,390
to Anthropic because there's an insatiable demand there.

223
00:14:04,634 --> 00:14:08,247
So it makes sense to just increase their stake. And it makes sense if you are

224
00:14:08,247 --> 00:14:10,437
a memory manufacturer like Micron, you want to do the same thing.

225
00:14:10,437 --> 00:14:15,677
If you are a GPU provider like NVIDIA or AMD, you also want to do the same thing.

226
00:14:15,677 --> 00:14:19,537
And so Anthropic has offered the ability for another 26 billion.

227
00:14:19,537 --> 00:14:22,787
Now that would put the rough raise estimate at anywhere between,

228
00:14:22,787 --> 00:14:27,757
I think, 70 to 80 billion dollars roughly, which is around the same ballpark

229
00:14:27,757 --> 00:14:29,618
that SpaceX actually raised.

230
00:14:30,129 --> 00:14:35,301
Except that new news suggests that the round is massively oversubscribed.

231
00:14:35,557 --> 00:14:40,693
Despite the market downturn, despite the crazy valuations, it's obvious that

232
00:14:40,693 --> 00:14:46,913
investors want to buy the Anthropic IPO to the point where this might get pushed to $100 billion.

233
00:14:47,453 --> 00:14:52,867
Worth of capital raise on IPO day, which is the largest amount that we've seen from any company.

234
00:14:53,047 --> 00:14:56,723
And it beats SpaceX's previous record that was only set a few months ago.

235
00:14:56,723 --> 00:15:03,573
Now, this would mean Anthropic would have around $212 to $220 plus billion of

236
00:15:03,573 --> 00:15:05,533
just cash on their balance sheet.

237
00:15:05,724 --> 00:15:10,473
In fact, one stat that surprised me from the initial S1 filing was that per

238
00:15:10,473 --> 00:15:15,796
dollar compute spend, they generated around $2 of revenue.

239
00:15:16,080 --> 00:15:20,523
So that margin, I'm sure we'll get eaten into depending on how much they need

240
00:15:20,523 --> 00:15:25,583
to expend on CapEx going forwards, But that indicates that there is a positive

241
00:15:25,583 --> 00:15:28,793
ROI somewhere soon in Anthropik's future post-IPO.

242
00:15:28,793 --> 00:15:32,203
Now, some other important takeaways from Anthropik's roadshow that is currently

243
00:15:32,203 --> 00:15:35,591
underway. Apparently, a few statements they're making are quite bold.

244
00:15:35,956 --> 00:15:40,903
One of the main ones being our reliance on SpaceX, buying compute from SpaceX.

245
00:15:40,903 --> 00:15:48,333
I think they're spending around $1 to $3 billion per month to access or acquire SpaceX's compute.

246
00:15:48,484 --> 00:15:52,093
SpaceX being one of the number one Neocard providers because Elon Musk acquired

247
00:15:52,093 --> 00:15:57,413
all the GPUs, they're looking to sever that relationship, that contract by at

248
00:15:57,413 --> 00:16:00,347
least the middle of 2027.

249
00:16:00,637 --> 00:16:05,303
So they pretty much are saying they have a confident path to getting enough

250
00:16:05,303 --> 00:16:08,743
compute at a very fair rate, whether it's their own chips that they're spinning

251
00:16:08,743 --> 00:16:12,740
up or access to NVIDIA GPUs or access to AMD GPUs

252
00:16:13,002 --> 00:16:16,972
such that they can replace the expensive premium that they're paying Elon Musk.

253
00:16:17,315 --> 00:16:21,683
Now, the immediate thought that pops to my mind is, would Elon Musk still be

254
00:16:21,683 --> 00:16:25,169
able to charge the premium that he is, that is earning SpaceX's.

255
00:16:25,662 --> 00:16:31,238
A load of revenue right now? And does that put into question SpaceX's revenue projections?

256
00:16:31,238 --> 00:16:36,488
Like he famously said on his most recent quarterly earnings report that he expects

257
00:16:36,488 --> 00:16:42,183
around anywhere between $500 billion to $1 trillion of ARR by the end of next year.

258
00:16:42,404 --> 00:16:46,048
Anthropic is one of their major cloud customers. So if Anthropic severs that

259
00:16:46,048 --> 00:16:49,468
tie by mid next year, that could threaten his estimate.

260
00:16:49,468 --> 00:16:53,928
The second is Dario seems to have taken a shot at open source saying that open

261
00:16:53,928 --> 00:16:59,308
source lacks the talent that Anthropic has and that Anthropic has better stability.

262
00:16:59,308 --> 00:17:03,168
And what he's inferring here is that, listen, we haven't been releasing models

263
00:17:03,168 --> 00:17:07,538
as frequently as we could have, and we would have expected by now for open source

264
00:17:07,538 --> 00:17:10,302
to have then caught up, but they haven't.

265
00:17:10,545 --> 00:17:14,158
And that would imply that the talent, the resources that are required to build

266
00:17:14,158 --> 00:17:17,668
bleeding frontier level models isn't available to open source.

267
00:17:17,668 --> 00:17:21,328
And so he's kind of putting the death knell on them. They're saying,

268
00:17:21,328 --> 00:17:23,490
yeah, see, I told you they can't make it.

269
00:17:23,694 --> 00:17:26,480
You have to bet on Anthropic because we're the only ones that are going to make it.

270
00:17:26,846 --> 00:17:31,240
And the third bold statement that he makes is that the AI market is a winner-takes-all market.

271
00:17:31,519 --> 00:17:34,898
And as long as we execute properly, we're going to take it all,

272
00:17:34,898 --> 00:17:36,478
we being Anthropic in this case.

273
00:17:36,478 --> 00:17:39,818
So there's some very bold statements that you would expect ahead of an IPO.

274
00:17:39,818 --> 00:17:43,528
He's talking his book. They're talking their game. Now, on the point of AI companies

275
00:17:43,528 --> 00:17:47,928
crushing their revenue targets, I want to point out a very obvious narrative

276
00:17:47,928 --> 00:17:50,947
violation, which is being pebbled over the last couple of weeks, which is

277
00:17:51,289 --> 00:17:55,188
NeoClouds, especially the smaller ones like CoreWeave are going to implode because

278
00:17:55,188 --> 00:17:55,951
they're over leveraged.

279
00:17:56,195 --> 00:18:00,658
And number two, that data centers aren't generating enough revenue to trickle

280
00:18:00,658 --> 00:18:02,568
down into the communities that they're being built.

281
00:18:02,934 --> 00:18:06,278
Both of those have just been disproved. And I wanna get into it on today's segment

282
00:18:06,278 --> 00:18:07,306
brought to you by Qualcomm.

283
00:18:11,229 --> 00:18:14,874
Now, off the bat, Corweave crushed their earnings. They beat it on revenue.

284
00:18:15,101 --> 00:18:18,677
Their backlog is growing around 30% from the previous quarter,

285
00:18:18,677 --> 00:18:23,131
which is absolutely insane. I think it's around $130 billion or so.

286
00:18:23,409 --> 00:18:28,117
And they have enough power to set all of this up. So the financial loans as

287
00:18:28,117 --> 00:18:30,887
well that they're taking are also at a reasonable rate.

288
00:18:30,887 --> 00:18:33,510
Apparently, the next loan that they take is going to be between,

289
00:18:33,777 --> 00:18:36,837
I think, 7% to 8%, which is considered investment grade.

290
00:18:36,837 --> 00:18:41,017
So what I'm saying here is all the accusations from the bears around NeoCloud's

291
00:18:41,017 --> 00:18:45,247
imploding, which came from the likes of Chamath and Elon Musk saying CoreWeave

292
00:18:45,247 --> 00:18:48,081
might be questionable at this point, was actually false.

293
00:18:48,336 --> 00:18:51,977
But the other thing I want to point out is the demand for CoreWeave's GPUs,

294
00:18:51,977 --> 00:18:54,634
which is a good barometer for the rest of the NeoCloud industry.

295
00:18:55,163 --> 00:19:01,928
Is overwhelming to the point where they have 10x qualified demand for the available

296
00:19:01,928 --> 00:19:03,266
capacity that they have.

297
00:19:03,510 --> 00:19:07,545
So what that means is 2027's capacity completely sold out.

298
00:19:07,783 --> 00:19:11,298
And any capacity that is coming online right now, which they're trying to figure

299
00:19:11,298 --> 00:19:15,393
out for 2028, is being sold immediately for a premium.

300
00:19:15,637 --> 00:19:18,406
But it's not just the brand new GPUs that they're sending out.

301
00:19:18,679 --> 00:19:23,738
It's also the older GPUs. You see, they have old NVIDIA GPUs like Hoppers,

302
00:19:23,738 --> 00:19:26,578
for example, which are selling at a premium.

303
00:19:26,578 --> 00:19:30,608
I think it's around 30% to 50% above the average rental price.

304
00:19:30,608 --> 00:19:33,984
And that's increased over the last couple of months. So remember,

305
00:19:34,292 --> 00:19:38,228
GPUs don't necessarily need to be used for training. They can be used for inference,

306
00:19:38,228 --> 00:19:39,720
routing, whatever that might be.

307
00:19:39,981 --> 00:19:44,078
They can have a home to service this entire AI economy, which means that the

308
00:19:44,078 --> 00:19:50,848
depreciation cycle, the life cycle, the extent, lifespan of a GPU goes way beyond

309
00:19:50,848 --> 00:19:52,098
what we were originally expecting.

310
00:19:52,098 --> 00:19:56,148
And that's an important point to keep in mind. Okay, okay. But what about the data centers?

311
00:19:56,148 --> 00:20:00,458
We recently had elections here in the US, the midterms, where it was a sweep

312
00:20:00,458 --> 00:20:05,108
for the Dems. And the main policy effort around the Dems was apparently they.

313
00:20:05,668 --> 00:20:08,222
Forbid or are against AI data centers.

314
00:20:08,443 --> 00:20:12,611
And the idea here is they're going to slow down up to 45% of data center capacity

315
00:20:12,854 --> 00:20:14,625
that is coming online next year.

316
00:20:14,985 --> 00:20:19,818
If that ended up happening, this would quell the entire momentum of AI and it

317
00:20:19,818 --> 00:20:22,960
would slow a lot of progress down. So all the revenue that Anthropic is making,

318
00:20:23,128 --> 00:20:27,493
all the revenue that SpaceX is making is ultimately going to slow down because of this.

319
00:20:28,191 --> 00:20:32,248
This doesn't necessarily seem to be the case. On one hand, you have the Memphis

320
00:20:32,248 --> 00:20:38,198
mayor saying from a single data center hosted by Elon Musk, this must be Colossus

321
00:20:38,198 --> 00:20:39,696
1, 2, or whatever that might be,

322
00:20:40,068 --> 00:20:44,238
they have earned between $20 to $54 million.

323
00:20:45,118 --> 00:20:50,918
Worth of taxes from this data center, which makes this single data center site

324
00:20:50,918 --> 00:20:56,612
from the single individual, single company, the largest taxpaying entity for that state.

325
00:20:56,879 --> 00:21:01,117
So just to put that into context, AI data centers are generating more revenue

326
00:21:01,332 --> 00:21:06,713
for these individual township states that they are placed in than we've ever seen before.

327
00:21:06,951 --> 00:21:11,746
That money can be used to subsidize property taxes, subsidize electricity bills,

328
00:21:12,094 --> 00:21:15,705
pay teachers higher bonuses that we're seeing in Louisiana, for example.

329
00:21:15,954 --> 00:21:21,308
The idea here is that wealth from AI is actually trickling down into the regular

330
00:21:21,308 --> 00:21:24,040
locals and individuals of where these data centers are placed.

331
00:21:24,167 --> 00:21:28,508
And if you also look at SpaceX's Arizona data center, the leader there basically

332
00:21:28,508 --> 00:21:33,397
said we could actually reduce property taxes up to 20% in a specific county

333
00:21:33,641 --> 00:21:35,348
because of how much money we're making from them.

334
00:21:35,591 --> 00:21:39,248
Okay, and then quickly, the last two stories, Gemini Carbon,

335
00:21:39,248 --> 00:21:44,828
which is Google's latest model, which up until Fable 5.2's release was theoretically

336
00:21:44,828 --> 00:21:47,668
the number one model, is officially publicly available.

337
00:21:47,668 --> 00:21:52,238
But also in other Google news, they released two brand new versions of their

338
00:21:52,238 --> 00:21:55,898
TPUs, TPUs being the main competitor that they have to NVIDIA's GPUs.

339
00:21:55,898 --> 00:22:01,609
And you could argue that Google's TPUs is the second most threat to NVIDIA's entire monopoly.

340
00:22:01,842 --> 00:22:04,582
Anthropic OpenAI are making huge problems.

341
00:22:05,248 --> 00:22:09,672
Billions of dollars worth of purchases of Google's TPUs to train and inference their models.

342
00:22:09,991 --> 00:22:13,241
Google came up with two new versions. And what I find interesting about these

343
00:22:13,241 --> 00:22:17,551
two versions is one is optimized for private confidential compute.

344
00:22:17,551 --> 00:22:20,415
So it's optimized for your data not leaking.

345
00:22:20,700 --> 00:22:23,921
Why I think this is important is if you're an enterprise or Fortune 500 company

346
00:22:23,921 --> 00:22:28,879
that is fearful of Anthropic or OpenAI stealing your data and training,

347
00:22:29,343 --> 00:22:33,051
I guess, an AI version of your own product and service and eventually replacing

348
00:22:33,051 --> 00:22:37,000
you, you don't want to give them your data, but you're ending up using their models.

349
00:22:37,227 --> 00:22:41,058
What if you could secure it at the infrastructure level? I think Google's TPUs

350
00:22:41,308 --> 00:22:46,481
will be a good instance to eat into that market share and sell potentially tens

351
00:22:46,481 --> 00:22:49,098
to hundreds of millions of dollars worth of compute that way.

352
00:22:49,562 --> 00:22:55,907
But the other version is specifically optimized for long agentic horizon coding, which means that.

353
00:22:56,517 --> 00:23:00,959
This narrative, this trend that I'm seeing of AI becoming really good at long,

354
00:23:00,959 --> 00:23:05,389
hard, complex tasks and it being able to act independently on its own and not

355
00:23:05,389 --> 00:23:10,899
come to you for help or approval all the time is ultimately the trend that we're going to see happening.

356
00:23:10,899 --> 00:23:14,709
And the final story is Meta is back on the scene with a new model.

357
00:23:14,709 --> 00:23:17,419
It's not their biggest model. It's not their most impressive or intelligent

358
00:23:17,419 --> 00:23:21,581
model, but it is their cheapest model. It's called Meta Ember,

359
00:23:21,877 --> 00:23:26,029
and it basically competes with the cheapest tier of all the frontier labs.

360
00:23:26,029 --> 00:23:29,209
So if it is ChatGPT, it's going to be Luna.

361
00:23:29,499 --> 00:23:33,098
And if it is Anthropic, it's going to be Claude Haiku.

362
00:23:33,406 --> 00:23:38,279
Now, it comes in at only a third of the cost for each of those individual competitor

363
00:23:38,279 --> 00:23:39,925
models. So it's incredibly cheap.

364
00:23:40,181 --> 00:23:43,861
And the idea here is it's not meant to be used for your most hardest work,

365
00:23:44,122 --> 00:23:48,011
but it's meant to be used for the bulk of the work that you already know what needs to get done.

366
00:23:48,273 --> 00:23:51,732
You need a model that's just intelligent enough, but doesn't need to be the most expensive.

367
00:23:51,970 --> 00:23:56,169
This can be your workhorse model, and it's going to come in cheaper than anything.

368
00:23:56,169 --> 00:24:01,919
And this honestly echoes Zuck and Meta's entire strategy around AI since the

369
00:24:01,919 --> 00:24:07,169
start, which has been, we will provide you the cheapest AI model possible and

370
00:24:07,169 --> 00:24:09,165
we will beat our competitors on cost.

371
00:24:09,380 --> 00:24:13,185
And that will result in getting AI into the hands of as many people as we can,

372
00:24:13,360 --> 00:24:15,490
which means that we'll end up owning the entire market.

373
00:24:15,856 --> 00:24:19,855
It's an interesting thesis. We'll see if it plays out. But that is today's episode.

374
00:24:20,186 --> 00:24:24,109
Lots going on this week. So much more happening next week if we have the Anthropic

375
00:24:24,109 --> 00:24:28,453
IPO and probably a new open AI model, hopefully GPT 6.2 Astra.

376
00:24:28,737 --> 00:24:31,689
So yeah, thank you so much for listening. And if you enjoyed this,

377
00:24:31,689 --> 00:24:34,449
please share it with a friend. If you're listening to this on YouTube,

378
00:24:34,449 --> 00:24:36,916
Spotify, Apple Music, wherever you are, please give us a rating,

379
00:24:37,212 --> 00:24:40,629
leave us a comment, turn on notifications. It helps us out massively.

380
00:24:40,629 --> 00:24:43,651
And yeah, thank you for listening to today's episode brought to you by Qualcomm.