1
00:00:00,018 --> 00:00:03,928
Ejaaz:
Earlier this week, the most powerful people in finance stood around a table

2
00:00:03,928 --> 00:00:09,300
Ejaaz:
next to Jensen Huang and announced they'd raised $500 billion to buy NVIDIA GPUs.

3
00:00:09,526 --> 00:00:13,308
Ejaaz:
Now, if you're listening to this and you're thinking this is just an AI bubble

4
00:00:13,308 --> 00:00:16,608
Ejaaz:
circular economy type thing, you might not actually be wrong.

5
00:00:16,608 --> 00:00:21,848
Ejaaz:
Larry Fink, the head of BlackRock, actually likened this deal to mortgage-backed

6
00:00:21,858 --> 00:00:23,938
Ejaaz:
debt securities of the early 2000s.

7
00:00:23,938 --> 00:00:27,388
Ejaaz:
And if that sounds familiar, to which he created. Yes, to which he created.

8
00:00:27,388 --> 00:00:31,640
Ejaaz:
And if you're likening that to a kind of like a PTSD flashback,

9
00:00:31,913 --> 00:00:36,888
Ejaaz:
that's because that's exactly what happened in the 2008 financial crisis itself.

10
00:00:37,016 --> 00:00:40,828
Ejaaz:
But if you look at the news in general, if you look at the way that this deal

11
00:00:40,828 --> 00:00:45,897
Ejaaz:
is structured, it might actually hint at something completely different. In fact, the opposite.

12
00:00:46,158 --> 00:00:50,640
Ejaaz:
GPU prices for renting has gone sky high. It's up 40% on the year.

13
00:00:50,855 --> 00:00:54,688
Ejaaz:
And there's not enough GPUs to back a lot of the deals that Microsoft,

14
00:00:54,688 --> 00:00:58,518
Ejaaz:
Google, Anthropic and OpenAI are signing with NVIDIA. So the question that we're

15
00:00:58,518 --> 00:01:01,898
Ejaaz:
going to unpack on the show is, is this very much a bubble back deal?

16
00:01:02,218 --> 00:01:04,688
Ejaaz:
Or is this something completely different that we're missing?

17
00:01:04,967 --> 00:01:11,348
Josh:
A new paradigm of investing, a new paradigm of financial manufacturing and construction.

18
00:01:11,348 --> 00:01:14,188
Josh:
This is a new investable asset class. Yeah, this is an entirely new thing.

19
00:01:14,188 --> 00:01:15,207
Josh:
GPUs. Who would have thought?

20
00:01:15,445 --> 00:01:18,018
Josh:
Michael Burry, the guy who's responsible for the big short, he was like,

21
00:01:18,018 --> 00:01:22,097
Josh:
no, the price of these things are going down only. It turns out he could not have been more wrong.

22
00:01:22,353 --> 00:01:28,438
Josh:
And now Jensen has assembled Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR.

23
00:01:28,438 --> 00:01:31,895
Ejaaz:
It's the Avengers of finance. It's the Avengers of finance.

24
00:01:32,063 --> 00:01:34,786
Josh:
It's like Apollo alone has a trillion dollars.

25
00:01:35,221 --> 00:01:38,553
Josh:
Of assets managed. Blackstone has over 1.3 trillion.

26
00:01:38,872 --> 00:01:42,094
Josh:
Brookfield is over a trillion. And combined, that's 3.4 trillion.

27
00:01:42,373 --> 00:01:44,867
Josh:
BlackRock is bigger than all three of those combined.

28
00:01:45,307 --> 00:01:48,947
Josh:
And they're all doing this together. And together, they've signed this thing

29
00:01:48,947 --> 00:01:51,847
Josh:
called a memorandum of understanding. Now, I had to actually look up what this

30
00:01:51,847 --> 00:01:52,903
Josh:
means because I had no idea.

31
00:01:53,135 --> 00:01:57,588
Josh:
A memorandum of understanding, or an MOU, is a formal, usually non-binding document

32
00:01:57,872 --> 00:01:58,981
Josh:
signed by two or more groups.

33
00:01:59,184 --> 00:02:01,707
Josh:
It shows that the groups share a common goal and plan to work together.

34
00:02:01,707 --> 00:02:04,547
Josh:
So this is not a contractual obligation. We have to start with that.

35
00:02:04,547 --> 00:02:09,459
Josh:
This is not a guarantee that $500 billion is going to flow into this new economy.

36
00:02:09,685 --> 00:02:12,967
Josh:
But it is an intention that all of these people are going to be aligned to work

37
00:02:12,967 --> 00:02:17,227
Josh:
together towards funding this next build out of AI. And what I found most interesting

38
00:02:17,227 --> 00:02:20,552
Josh:
is that this was actually orchestrated entirely by Jensen.

39
00:02:20,953 --> 00:02:24,028
Josh:
Jensen reached out to all of these banks himself personally.

40
00:02:24,255 --> 00:02:27,667
Josh:
And he said, hey, I'd like to work together on this thing. And not a single

41
00:02:27,667 --> 00:02:29,264
Josh:
bank that he reached out to said no.

42
00:02:29,467 --> 00:02:32,887
Josh:
So here we are now with a moment on CNBC in which they're all sitting around

43
00:02:32,887 --> 00:02:37,827
Josh:
a table talking about how they are committing $500 billion to this new asset class.

44
00:02:37,827 --> 00:02:41,847
Josh:
And this is unbelievable. This feels like a, for better or worse,

45
00:02:41,847 --> 00:02:47,557
Josh:
a brand new paradigm for the AI era in which now the collective force of the

46
00:02:47,557 --> 00:02:52,067
Josh:
United States banking system is like starting to get behind this.

47
00:02:52,067 --> 00:02:57,157
Josh:
And I should say this is not for AI as a whole. This is purely for NVIDIA as a company.

48
00:02:57,395 --> 00:03:01,857
Ejaaz:
Yeah, and I want to take a moment to actually explain what's happening here,

49
00:03:01,857 --> 00:03:03,297
Ejaaz:
because I think there's a lot of confusion.

50
00:03:03,297 --> 00:03:05,327
Ejaaz:
There's a lot of headlines that people are getting worried over.

51
00:03:05,327 --> 00:03:10,297
Ejaaz:
NVIDIA stock tanks 4% on the news, but I think that's a little too early to

52
00:03:10,297 --> 00:03:14,617
Ejaaz:
judge. So firstly, what does this structure sort of look like?

53
00:03:14,617 --> 00:03:17,684
Ejaaz:
Well, it's what you're seeing on the screen right now, which is essentially.

54
00:03:18,468 --> 00:03:22,588
Ejaaz:
There's a problem in AI right now, which is all these hyperscalers,

55
00:03:22,588 --> 00:03:26,441
Ejaaz:
all these AI labs, Anthropic, OpenAI, Google, Meta, you name it,

56
00:03:26,906 --> 00:03:30,618
Ejaaz:
have spent a lot of money to buy GPUs.

57
00:03:30,628 --> 00:03:33,888
Ejaaz:
The reason why they're doing this is to train and inference brand new AI models,

58
00:03:33,888 --> 00:03:36,031
Ejaaz:
which they have a lot of paying customers for.

59
00:03:36,513 --> 00:03:40,088
Ejaaz:
But the issue they're facing is the money they've invested, which is now to

60
00:03:40,088 --> 00:03:44,188
Ejaaz:
the tune of $2.6 trillion converted over the next couple of years,

61
00:03:44,188 --> 00:03:46,851
Ejaaz:
I believe, is not enough for them.

62
00:03:47,217 --> 00:03:50,718
Ejaaz:
So much so that they're going into negative cash flow. So what happens when

63
00:03:50,718 --> 00:03:53,295
Ejaaz:
you've spent all the money that you have in your company, in your balance sheet?

64
00:03:53,672 --> 00:03:59,218
Ejaaz:
You need to go to Wall Street. That's exactly what NVIDIA, specifically Jensen, has brokered.

65
00:03:59,218 --> 00:04:03,198
Ejaaz:
He's gone to Wall Street and he said, listen, we need more money to build more

66
00:04:03,198 --> 00:04:08,158
Ejaaz:
GPUs to sell to these different customers so that they can produce their products

67
00:04:08,158 --> 00:04:09,514
Ejaaz:
and services, their new models.

68
00:04:09,804 --> 00:04:12,868
Ejaaz:
And Wall Street has gone back and said, I have an issue with this,

69
00:04:12,868 --> 00:04:17,598
Ejaaz:
Jensen, which is GPUs aren't a versatile asset. like they can only be used for

70
00:04:17,598 --> 00:04:20,378
Ejaaz:
one thing specifically, which is either training a model or inferencing a model,

71
00:04:20,378 --> 00:04:21,785
Ejaaz:
and it's only customer-specific.

72
00:04:22,214 --> 00:04:25,448
Ejaaz:
And Jensen responded to them and said, that's not true at all.

73
00:04:25,715 --> 00:04:30,208
Ejaaz:
GPUs, specifically NVIDIA GPUs, are the most versatile asset out there.

74
00:04:30,353 --> 00:04:32,778
Ejaaz:
You can use it for anything. You can use it for training, you can use it for

75
00:04:32,778 --> 00:04:35,358
Ejaaz:
inference, and you can use it for any model, whether it's Claude,

76
00:04:35,358 --> 00:04:39,200
Ejaaz:
whether it's GPT, whether it's Gemini, whatever. You can use it for it,

77
00:04:39,414 --> 00:04:42,207
Ejaaz:
which means that it's a versatile customer base.

78
00:04:42,799 --> 00:04:46,253
Ejaaz:
Earns a lot of money. And then Wall Street shot back at him and said,

79
00:04:46,537 --> 00:04:49,497
Ejaaz:
well, hang on a second, these GPUs die after a couple of years.

80
00:04:49,497 --> 00:04:54,177
Ejaaz:
And Jensen goes, that's not actually true. In fact, we have 10-year-old GPUs

81
00:04:54,177 --> 00:04:57,915
Ejaaz:
that are being re-signed for another 10 years right now today at a higher price

82
00:04:58,188 --> 00:04:59,627
Ejaaz:
than they sold earlier on.

83
00:04:59,947 --> 00:05:05,212
Ejaaz:
So basically what he's pitched them is this is a new asset class and it can earn a ton of money.

84
00:05:05,421 --> 00:05:08,647
Ejaaz:
And so Wall Street looked at this, some of the biggest financial powerhouses

85
00:05:08,647 --> 00:05:10,349
Ejaaz:
in the world and thought, you know what?

86
00:05:10,587 --> 00:05:14,457
Ejaaz:
He might be right. This is an asset class that can be likened to property or

87
00:05:14,457 --> 00:05:15,783
Ejaaz:
railroads back in the day.

88
00:05:16,061 --> 00:05:20,780
Ejaaz:
And that's why Larry Fink is comparing it to the 2008 mortgage-backed securities.

89
00:05:20,994 --> 00:05:24,802
Ejaaz:
Now, if you're wondering, okay, well, this is like a financial crisis type thing,

90
00:05:25,145 --> 00:05:28,447
Ejaaz:
you might be right, except there was like a few different things going on there,

91
00:05:28,447 --> 00:05:30,085
Ejaaz:
which we'll unpack later in the episode.

92
00:05:30,451 --> 00:05:34,357
Josh:
Yeah, it feels very much like AI compute is equivalent to revenue.

93
00:05:34,584 --> 00:05:40,227
Josh:
And these are very now like durable, appreciating assets that can yield value over time.

94
00:05:40,227 --> 00:05:44,830
Josh:
So when you think of like a bond, per se, the value of a bond is implied to

95
00:05:45,079 --> 00:05:49,241
Josh:
go down over time, the underlying asset like the US dollar due to inflation,

96
00:05:49,532 --> 00:05:53,839
Josh:
but the yield it will come up with is going to outpace that and then some hopefully

97
00:05:54,170 --> 00:05:59,197
Josh:
the construction of a GPU is that not only do you get a yield in terms of the

98
00:05:59,412 --> 00:06:01,629
Josh:
value creation off the back of token generation.

99
00:06:01,850 --> 00:06:05,827
Josh:
But you now also have an asset that is likely going to appreciate.

100
00:06:05,827 --> 00:06:10,377
Josh:
And even in the face that it doesn't, Jensen is giving plunge protection.

101
00:06:10,377 --> 00:06:17,267
Josh:
So 25%. Yes. Depreciation insurance of up to 25% to help the banks get these

102
00:06:17,267 --> 00:06:19,536
Josh:
marginal deals over time. So banks initially were concerned.

103
00:06:19,798 --> 00:06:21,377
Josh:
They don't want to fund this because they don't want

104
00:06:21,858 --> 00:06:24,699
Josh:
nvidia to come out with the new gpu that's a thousand times better than this

105
00:06:24,699 --> 00:06:27,245
Josh:
one and it's going to knock all the margin out from underneath them

106
00:06:27,629 --> 00:06:31,369
Josh:
who knows the roadmap about nvidia gpus better than anyone else it's jensen

107
00:06:31,369 --> 00:06:34,333
Josh:
the guy who's building it so baked into this contract is the idea that

108
00:06:34,595 --> 00:06:38,239
Josh:
jensen will ensure you he will make sure that hey your gpus are not going to

109
00:06:38,239 --> 00:06:41,630
Josh:
fall drastically over time everything is going to be smooth and predictable

110
00:06:41,764 --> 00:06:43,824
Josh:
and we'll work together to fund these companies that don't have

111
00:06:44,184 --> 00:06:48,019
Josh:
the ability to do so so this comes in the form of like this long-term

112
00:06:48,799 --> 00:06:50,802
Josh:
Debt and asset backed structures.

113
00:06:51,087 --> 00:06:54,179
Josh:
And you think of it like if you're not a hyperscaler, if you're not Google who

114
00:06:54,179 --> 00:06:58,599
Josh:
has a couple hundred billion dollars to spend off your balance sheet this year,

115
00:06:58,599 --> 00:07:02,447
Josh:
but you still want to compete in the world of AI, you still need GPUs, these are who you go to.

116
00:07:02,755 --> 00:07:07,323
Josh:
And they will offer you GPUs in exchange for interest on these GPUs.

117
00:07:07,584 --> 00:07:11,359
Josh:
And in the worst case that it doesn't work out, they can just claw back the GPUs.

118
00:07:11,359 --> 00:07:14,037
Josh:
NVIDIA can claw back those GPUs and turn it into their own NeoCloud,

119
00:07:14,269 --> 00:07:19,535
Josh:
give it to another neocloud but the idea is that these assets are valuable they're increasing in money

120
00:07:19,825 --> 00:07:24,219
Josh:
they are transferable in an easy way that you can just take the gpu and plug

121
00:07:24,219 --> 00:07:28,445
Josh:
it in somewhere else or give the actual data center control over to someone else

122
00:07:28,823 --> 00:07:32,909
Josh:
and it's a really lucrative kind of bizarre thing it's like okay if you're a

123
00:07:32,909 --> 00:07:34,604
Josh:
bank 500 billion dollars

124
00:07:35,051 --> 00:07:38,708
Josh:
you have insurance you are now able to allocate this

125
00:07:39,213 --> 00:07:43,946
Josh:
incredibly valuable capital resource to anybody who you want and collect a pretty

126
00:07:43,946 --> 00:07:45,611
Josh:
high rate of return on top of that

127
00:07:45,999 --> 00:07:49,806
Josh:
and nvidia wants in on this too so initially it wasn't for nvidia nvidia now

128
00:07:49,806 --> 00:07:55,146
Josh:
is given the option to backstop up to 25 of each opportunity so that is 125

129
00:07:55,146 --> 00:07:56,402
Josh:
billion dollars at the ceiling

130
00:07:56,826 --> 00:08:01,666
Josh:
and basically now nvidia and co the avengers get to roam around choose who they

131
00:08:01,666 --> 00:08:05,406
Josh:
would like to give these gpus to and wrap it up in a really interesting financial

132
00:08:05,406 --> 00:08:08,227
Josh:
product that they can go off and, I guess, monetize.

133
00:08:08,529 --> 00:08:12,906
Ejaaz:
And I want to stress that this is only for NVIDIA GPUs specifically.

134
00:08:12,906 --> 00:08:16,428
Ejaaz:
Jensen brokered this deal for his company only.

135
00:08:16,747 --> 00:08:21,646
Ejaaz:
And he has a reason to do that because GPUs for the longest time has been very

136
00:08:21,646 --> 00:08:25,476
Ejaaz:
broad based. If you look at some of the GPUs that Google makes or that Meta

137
00:08:25,476 --> 00:08:29,890
Ejaaz:
is making or that even OpenAIR and Anthropic are reportedly working on their own specialized chips.

138
00:08:30,226 --> 00:08:33,466
Ejaaz:
These are exactly what I just said. They're specialized. They can't be used

139
00:08:33,466 --> 00:08:35,846
Ejaaz:
for many other models. It's only used specifically for their things.

140
00:08:35,846 --> 00:08:41,140
Ejaaz:
So it's a much more niche case to create a type of loan or credit-backed security for.

141
00:08:41,558 --> 00:08:44,948
Ejaaz:
Jensen has the opposite issue, which is like, it's too broad.

142
00:08:45,383 --> 00:08:51,502
Ejaaz:
But that makes it an amazing financial asset. So in effect, NVIDIA is sort of becoming a bank.

143
00:08:51,734 --> 00:08:55,101
Ejaaz:
And I wouldn't be surprised if Jensen starts to make a lot of money from this.

144
00:08:55,101 --> 00:08:59,891
Ejaaz:
Over the last couple of weeks, something that he's also started doing is backstopping

145
00:08:59,891 --> 00:09:04,131
Ejaaz:
specific Frontier AI labs and saying, hey, don't worry, I got you.

146
00:09:04,131 --> 00:09:07,465
Ejaaz:
I'll front up the money that you need to purchase my GPUs.

147
00:09:07,698 --> 00:09:12,346
Ejaaz:
And in return, whatever money you make on the products that you're building,

148
00:09:12,602 --> 00:09:14,231
Ejaaz:
you can give me a revenue split from that.

149
00:09:14,231 --> 00:09:19,701
Ejaaz:
I think he signed like a reportedly 10% revenue split from Safe Super Intelligence,

150
00:09:19,701 --> 00:09:23,411
Ejaaz:
which is Ilya Sutskiver's new lab for their breakthrough that they're launching pretty soon.

151
00:09:23,649 --> 00:09:26,681
Ejaaz:
And I think he's gonna do the same for a lot of neoclouts like CoreWeave,

152
00:09:26,681 --> 00:09:29,921
Ejaaz:
Nebius and such like that, that are reporting crazy earnings.

153
00:09:29,921 --> 00:09:34,951
Ejaaz:
I think this morning, CoreWeave reported 464% increase in revenue year upon

154
00:09:34,951 --> 00:09:38,823
Ejaaz:
year, which is just a precursor to like the insane demand that they're seeing right there.

155
00:09:38,979 --> 00:09:43,601
Ejaaz:
But then a question that comes into mind is, where on earth is this money coming from?

156
00:09:43,601 --> 00:09:49,841
Ejaaz:
And on the screen here, it's like the main claimants are pension funds, right?

157
00:09:49,841 --> 00:09:53,971
Ejaaz:
So pension funds who have amassed a large amount of wealth and typically don't

158
00:09:53,971 --> 00:09:56,971
Ejaaz:
invest in high volatile type assets.

159
00:09:56,971 --> 00:09:59,860
Ejaaz:
They kind of stick to real estate, very low interest types of things,

160
00:10:00,168 --> 00:10:04,741
Ejaaz:
are the ones that are going to be backing a lot of this new GPU asset class.

161
00:10:04,741 --> 00:10:05,881
Ejaaz:
And you have the biggest, most

162
00:10:05,881 --> 00:10:09,414
Ejaaz:
powerful financial people in the world that are kind of pushing this on.

163
00:10:09,640 --> 00:10:12,125
Ejaaz:
And so I'm thinking, is this reckless behavior?

164
00:10:12,717 --> 00:10:15,997
Ejaaz:
Well, if you take the word of Goldman Sachs CEO David Solomon,

165
00:10:16,217 --> 00:10:21,682
Ejaaz:
he goes, $500 billion sounds like a lot, but there are $9 trillion in US money

166
00:10:21,682 --> 00:10:26,689
Ejaaz:
market funds and more than $100 trillion in US equities.

167
00:10:26,922 --> 00:10:30,323
Ejaaz:
He has a deep belief in this opportunity and Goldman brings its extraordinary

168
00:10:30,608 --> 00:10:33,766
Ejaaz:
distribution network. Larry Fink, CEO of BlackRock, also says,

169
00:10:34,097 --> 00:10:37,870
Ejaaz:
he said it's a very attractive opportunity with long-dated, long-term returns.

170
00:10:38,050 --> 00:10:41,062
Ejaaz:
They will be talking to pension funds. So it seems like the two most powerful

171
00:10:41,362 --> 00:10:44,662
Ejaaz:
financial connoisseurs in the world are convinced that this new asset class

172
00:10:44,662 --> 00:10:48,042
Ejaaz:
is a very real thing, which means that they've probably looked at the balance sheets.

173
00:10:48,042 --> 00:10:50,772
Ejaaz:
They've probably looked at the revenue demand that a lot of these frontier labs

174
00:10:50,772 --> 00:10:53,462
Ejaaz:
that are meant to be purchasing these things are going to do.

175
00:10:53,462 --> 00:10:56,652
Ejaaz:
And they're looking at it and they're saying, this is an obvious no-brainer.

176
00:10:56,652 --> 00:10:59,882
Ejaaz:
Now, if you're listening to me and you're thinking, dude, this happened with

177
00:10:59,882 --> 00:11:01,752
Ejaaz:
railroads and it didn't work out.

178
00:11:01,903 --> 00:11:05,332
Ejaaz:
This happened with the housing environment, mortgage-backed debt securities

179
00:11:05,332 --> 00:11:06,928
Ejaaz:
in 2008, that didn't work out.

180
00:11:07,224 --> 00:11:10,586
Ejaaz:
I have to say, it's a very different story on our end.

181
00:11:10,736 --> 00:11:13,262
Josh:
Well, that's what I was going to ask you. I was going to say like,

182
00:11:13,262 --> 00:11:15,492
Josh:
hey, obviously they're going to come out and say these things.

183
00:11:15,492 --> 00:11:18,332
Josh:
I mean, we've seen them manipulate markets for a long time. We just saw what

184
00:11:18,332 --> 00:11:22,272
Josh:
Citadel did to Leopold. It's like everyone is very clearly out in their own

185
00:11:22,272 --> 00:11:25,882
Josh:
best interest. So if we look at this deal, okay, they're not going to say it's

186
00:11:25,882 --> 00:11:27,078
Josh:
anything less than exceptional.

187
00:11:27,287 --> 00:11:32,112
Josh:
So how do we kind of vet this? How do we fit this into a specific piece of context

188
00:11:32,112 --> 00:11:32,962
Josh:
that I guess we could reference?

189
00:11:32,962 --> 00:11:37,392
Josh:
And there's an interesting example of like aircraft finance versus mortgage

190
00:11:37,392 --> 00:11:40,638
Josh:
finance, because this is something that has happened in the past where

191
00:11:41,102 --> 00:11:45,369
Josh:
when you have an expensive standardized asset that's transferable between operators

192
00:11:45,526 --> 00:11:50,872
Josh:
and has a lot of demand for these secondary markets, it creates this interesting marketplace that

193
00:11:51,377 --> 00:11:54,432
Josh:
I think is much more comparable to aircrafts than mortgages.

194
00:11:54,432 --> 00:11:59,192
Josh:
And I'll explain. So like a Boeing 747 or 737 or whatever, that's been built

195
00:11:59,192 --> 00:12:02,542
Josh:
like 20 years ago. And trust me, you've flown on these. The airlines kind of

196
00:12:02,542 --> 00:12:03,730
Josh:
suck. You're flying in some old planes.

197
00:12:04,123 --> 00:12:06,050
Josh:
That is just as valuable today

198
00:12:06,741 --> 00:12:11,185
Josh:
As it was 20 years ago, because they're able to derive so much value from it.

199
00:12:11,185 --> 00:12:13,022
Josh:
It does the same exact job.

200
00:12:13,411 --> 00:12:16,875
Josh:
The same plane that was built today is doing the same job that was built 20

201
00:12:16,875 --> 00:12:19,982
Josh:
years ago. And sure, perhaps you would prefer to fly on the newer plane.

202
00:12:20,249 --> 00:12:25,619
Josh:
But the reality is, is that tickets are sold out on the 2005 plane and the 2025 plane.

203
00:12:25,938 --> 00:12:30,695
Josh:
And when you think of GPUs, they exhibit a lot of the same traits and characteristics

204
00:12:30,695 --> 00:12:33,705
Josh:
as an airplane, where it's expensive, standardized, it's transferable.

205
00:12:33,873 --> 00:12:35,702
Josh:
It has a lot of liquidity in secondary markets.

206
00:12:35,992 --> 00:12:39,365
Josh:
And I think this is an interesting way of looking at it relative to mortgage

207
00:12:39,365 --> 00:12:43,504
Josh:
finances, which is where we got in trouble. And this isn't the first time this has happened before.

208
00:12:43,730 --> 00:12:46,366
Josh:
There is something that has been done similar to this with Broadcom,

209
00:12:46,668 --> 00:12:50,105
Josh:
where like that Google Anthropic structure actually runs through this thing

210
00:12:50,105 --> 00:12:54,235
Josh:
called an SPV, a special purpose vehicle that buys TPUs and leases them with

211
00:12:54,235 --> 00:12:57,146
Josh:
Broadcom providing the residual value guarantees,

212
00:12:57,476 --> 00:13:00,845
Josh:
and then Apollo and Blackstone supplying the private credit to fund all of this.

213
00:13:00,845 --> 00:13:04,662
Josh:
So people have experimented with these structures, before. It has worked.

214
00:13:05,033 --> 00:13:09,235
Josh:
We haven't seen it at this scale. I mean, the alarm bells are partially ringing.

215
00:13:09,235 --> 00:13:12,455
Josh:
I'm like, just out of instinct, like out of an e-joke reaction,

216
00:13:12,455 --> 00:13:15,627
Josh:
like, oh, wow, this is a lot of money. This is a lot of powerful people who

217
00:13:16,051 --> 00:13:18,425
Josh:
can very much control and sway the way the market moves.

218
00:13:19,197 --> 00:13:23,511
Josh:
So far, it seems like a pretty reasonable thing. It's like, hey, we need GPUs.

219
00:13:23,818 --> 00:13:28,044
Josh:
GPUs are transferable. They're kind of like they're fungible,

220
00:13:28,044 --> 00:13:31,162
Josh:
I guess. I'm like thinking of the word. I'm like, well, this feels kind of crypto adjacent.

221
00:13:31,429 --> 00:13:34,104
Josh:
There's like these fungible assets that can be transferred that are valuable.

222
00:13:34,104 --> 00:13:37,344
Josh:
So I don't know. There's a chance this goes over. OK.

223
00:13:37,564 --> 00:13:41,944
Ejaaz:
And it's important to not extrapolate too far into the future,

224
00:13:41,944 --> 00:13:46,005
Ejaaz:
like what we can feasibly attain from the data, which, by the way,

225
00:13:46,260 --> 00:13:49,441
Ejaaz:
is publicly available. If you're listening to this and you don't believe anything that we're saying.

226
00:13:50,024 --> 00:13:52,614
Ejaaz:
Maybe we should actually link to a bunch of sources. Maybe we'll link this artifact

227
00:13:52,614 --> 00:13:54,091
Ejaaz:
that you're seeing on the screen right now.

228
00:13:54,329 --> 00:13:57,974
Ejaaz:
The data is all available through quarterly earnings of every single company

229
00:13:57,974 --> 00:14:02,164
Ejaaz:
that is leading at every layer of the AI stack. So you can see the data,

230
00:14:02,164 --> 00:14:04,545
Ejaaz:
digest it yourself and figure it out for yourself.

231
00:14:04,702 --> 00:14:08,573
Ejaaz:
But what I will say is when you look at the 2008 financial crisis,

232
00:14:08,695 --> 00:14:12,654
Ejaaz:
when you look at the railroad crisis, when you look at the telecom crisis back

233
00:14:12,654 --> 00:14:16,300
Ejaaz:
then there was a huge amount of oversupply.

234
00:14:16,886 --> 00:14:22,534
Ejaaz:
Which didn't have the back demand that it stated it had so 2008 people assumed

235
00:14:22,534 --> 00:14:26,081
Ejaaz:
that property prices would just keep going up and at some point

236
00:14:26,319 --> 00:14:29,774
Ejaaz:
that got two head over heels if you look at the railroad they built too much

237
00:14:29,774 --> 00:14:33,674
Ejaaz:
if you look at the telecom they built too many cables right in this case

238
00:14:34,052 --> 00:14:38,334
Ejaaz:
we're constrained by a few things number one physically it takes so many different

239
00:14:38,334 --> 00:14:40,501
Ejaaz:
substrate layers to build a gpu.

240
00:14:41,070 --> 00:14:45,831
Ejaaz:
Every single layer right now in the world of physical atoms is incredibly constrained.

241
00:14:45,831 --> 00:14:51,327
Ejaaz:
There's not enough. The GPU demand is overweight, the actual supply that is available.

242
00:14:51,542 --> 00:14:56,931
Ejaaz:
Number two, the AI demand, which is driving GPU demand, is accelerating much

243
00:14:56,931 --> 00:14:59,391
Ejaaz:
faster than the supply itself can.

244
00:14:59,391 --> 00:15:02,921
Ejaaz:
So if you look at memory as a basis for this, they can increase capacity.

245
00:15:02,921 --> 00:15:08,044
Ejaaz:
This is the top docs, top memory manufacturers can increase capacity around 20% per year.

246
00:15:08,369 --> 00:15:12,294
Ejaaz:
But demand is compounding at 45 to 60% per year.

247
00:15:12,630 --> 00:15:16,841
Ejaaz:
So if you do the math, if that continues, you're going to be in a constrained

248
00:15:16,841 --> 00:15:21,711
Ejaaz:
supply state for like at least until 2028 or 2029 until some of these other

249
00:15:21,711 --> 00:15:25,581
Ejaaz:
chip fabs can get increased. So we're still kind of in a holding period, right?

250
00:15:25,871 --> 00:15:29,131
Ejaaz:
Now, if you look at the backlog for some of these companies.

251
00:15:29,631 --> 00:15:33,241
Ejaaz:
You might be like, well, customers don't want AI as much as these guys are making

252
00:15:33,241 --> 00:15:35,014
Ejaaz:
it out to be. They're just kind of like pushing their bags.

253
00:15:35,426 --> 00:15:39,351
Ejaaz:
Well, look at Google's backlog. It doubled this year in a matter of months.

254
00:15:39,351 --> 00:15:43,478
Ejaaz:
It is now at $460 billion, and we're at the halfway mark of this year.

255
00:15:43,710 --> 00:15:46,631
Ejaaz:
It's probably going to increase even more. This is the case across Microsoft

256
00:15:46,631 --> 00:15:48,662
Ejaaz:
and a bunch of other hyperscalers as well.

257
00:15:48,964 --> 00:15:51,401
Ejaaz:
Then if you look at the GPU rental prices, my favorite thing,

258
00:15:51,401 --> 00:15:54,740
Ejaaz:
which kind of came out this morning, Josh, or maybe yesterday,

259
00:15:55,030 --> 00:16:00,688
Ejaaz:
CallWeave had their earnings, and they said, we recently signed an A100 contract that extends into 2029.

260
00:16:00,989 --> 00:16:05,314
Ejaaz:
For those of you who don't know, and A100 is an NVIDIA GPU that was created in 2020.

261
00:16:05,790 --> 00:16:09,430
Ejaaz:
And its lifecycle prediction back then was three and a half years.

262
00:16:09,842 --> 00:16:13,181
Ejaaz:
Now they're predicting that it's actually going to be good to go until 2029.

263
00:16:13,181 --> 00:16:15,484
Ejaaz:
That's because it's not just being used for bleeding edge training.

264
00:16:15,786 --> 00:16:17,992
Ejaaz:
It's being used for inference and a bunch of other stuff. So the point is,

265
00:16:18,253 --> 00:16:22,665
Ejaaz:
these GPUs are very versatile, NVIDIA specifically, and that's why they raised a crap ton of money.

266
00:16:22,938 --> 00:16:27,111
Josh:
Yeah, and more valuable over time. It's this really bizarre thing in which the

267
00:16:27,531 --> 00:16:33,294
Josh:
useful lifecycle of a hardware object is increasing instead of depreciating for the first time.

268
00:16:33,584 --> 00:16:36,301
Josh:
And we've never really seen this phenomenon at scale before.

269
00:16:36,580 --> 00:16:39,451
Josh:
But like you mentioned, there's not many reasons in which it's going to slow

270
00:16:39,451 --> 00:16:43,122
Josh:
down in the near term future. When I look at this, I'm kind of looking at it

271
00:16:43,697 --> 00:16:47,380
Josh:
like around the corner. And then you can't really see around the next corner,

272
00:16:47,380 --> 00:16:51,603
Josh:
but we have an idea of the first corner. And that first corner is sold out supply

273
00:16:51,934 --> 00:16:55,427
Josh:
for at least 2027, likely 2028.

274
00:16:55,723 --> 00:17:01,130
Josh:
And then by the end of the decade, 2029, 2020, 2030, we start to run into larger

275
00:17:01,130 --> 00:17:04,442
Josh:
constraints, mostly around energy and power.

276
00:17:04,791 --> 00:17:08,480
Josh:
And we start to like run into resource constraints that we don't quite have now.

277
00:17:08,480 --> 00:17:11,580
Ejaaz:
So are you basically saying there's like multiple corners, Josh?

278
00:17:11,580 --> 00:17:15,920
Ejaaz:
Like I'm curious, like, for the GPU specifically, do you think it's like six

279
00:17:15,920 --> 00:17:19,190
Ejaaz:
months? Do you think it's also like 12 months? Like, what's your guess if you had to...

280
00:17:19,240 --> 00:17:23,190
Josh:
Well, there's a somewhat clear trajectory for the next 24 months,

281
00:17:23,490 --> 00:17:25,451
Josh:
like 18 to 24 months in terms of...

282
00:17:26,031 --> 00:17:30,867
Josh:
It seems fairly predictable where we know how many, like the lithography machines,

283
00:17:30,867 --> 00:17:32,730
Josh:
we know how many chips they can create.

284
00:17:32,945 --> 00:17:36,654
Josh:
Then we know how many of those chips can be packaged into usable chips.

285
00:17:36,898 --> 00:17:40,287
Josh:
And then we know roughly how many data centers can be built that can actually

286
00:17:40,287 --> 00:17:41,641
Josh:
turn those chips on and power them.

287
00:17:41,861 --> 00:17:45,737
Josh:
And you can somewhat project that out up to 24 months loosely,

288
00:17:45,737 --> 00:17:50,367
Josh:
very loosely, because there is only so much throughput for these machines.

289
00:17:50,367 --> 00:17:53,557
Josh:
So if you assume everyone's operating at full capacity, you can kind of work

290
00:17:53,557 --> 00:17:56,867
Josh:
those numbers backwards and understand like, OK, they're sold out and they're

291
00:17:56,867 --> 00:18:00,877
Josh:
still not going to be enough to satisfy the demand, assuming the demand continues,

292
00:18:00,877 --> 00:18:02,647
Josh:
which there is no signs of slowing down.

293
00:18:02,647 --> 00:18:07,007
Josh:
All these use cases for AI, particularly around agentic AI, require a tremendous

294
00:18:07,007 --> 00:18:10,607
Josh:
amount of inference and even including efficiency upgrades to the software,

295
00:18:10,607 --> 00:18:13,149
Josh:
something similar to what we imagine SSI is working on.

296
00:18:13,387 --> 00:18:16,907
Josh:
There's still a huge amount of demand that will fulfill that this is jevin's

297
00:18:16,907 --> 00:18:20,707
Josh:
paradox which we really should name jensen's paradox because that seems to be

298
00:18:20,707 --> 00:18:22,617
Josh:
a little more accurate in terms of how this is working

299
00:18:22,937 --> 00:18:26,797
Josh:
but we can kind of project out till then and we know all right gpus fully sold

300
00:18:26,797 --> 00:18:29,737
Josh:
out fully constrained after that things get a little

301
00:18:30,197 --> 00:18:33,877
Josh:
more hairy right it's because you have to assume by that time we'll have something

302
00:18:33,877 --> 00:18:38,552
Josh:
similar to agi asi self-recursive improvements we should be getting a lot of

303
00:18:38,906 --> 00:18:42,537
Josh:
innovation breakthroughs around efficiency and software and we don't really know what

304
00:18:42,917 --> 00:18:45,849
Josh:
the power market's going to look like. We're not sure if we're able to make enough

305
00:18:46,464 --> 00:18:50,121
Josh:
Energy to satisfy the demand of the gpu centers that are being projected out

306
00:18:50,580 --> 00:18:56,033
Josh:
into 20 to 30 so it seems like this is a very long duration thing that's going

307
00:18:56,033 --> 00:19:00,981
Josh:
to need to play out but in the short term in the intermediary term it seems like i mean

308
00:19:01,474 --> 00:19:04,553
Josh:
i'd like to find the steel man against this because this seems important and

309
00:19:04,553 --> 00:19:07,610
Josh:
we should talk about like what are the possible ways in which this breaks

310
00:19:07,918 --> 00:19:12,074
Josh:
but just looking at demand of inference and our capability of serving inference

311
00:19:12,353 --> 00:19:16,273
Josh:
and there is a huge mismatch in the case of inference demand that's just not

312
00:19:16,273 --> 00:19:17,363
Josh:
going to be met for a really long time.

313
00:19:17,618 --> 00:19:21,072
Josh:
So those H100s from 2020, or the A100s, I should say, from 2020,

314
00:19:21,298 --> 00:19:24,985
Josh:
are still going to be useful in 2027, 2028.

315
00:19:25,613 --> 00:19:29,054
Josh:
And that is particularly valuable when you're investing in GPUs at this scale.

316
00:19:29,443 --> 00:19:32,753
Ejaaz:
Well, actually, now that you say it, a lot of that inference demand,

317
00:19:32,753 --> 00:19:37,313
Ejaaz:
at least in the next six months or so, is going to come from AI agents.

318
00:19:37,313 --> 00:19:39,259
Ejaaz:
I'm in no doubt about that.

319
00:19:39,584 --> 00:19:43,383
Josh:
You said the word agents. It's funny you should mention agents because we have

320
00:19:43,383 --> 00:19:46,927
Josh:
something to say about agents from our sponsor of this episode, Ledger.

321
00:19:47,473 --> 00:19:51,013
Josh:
If you're building with AI agents, you are probably worried about security and

322
00:19:51,013 --> 00:19:54,523
Josh:
rightfully so because as we've seen recently, these agents have been kind of

323
00:19:54,523 --> 00:19:55,493
Josh:
doing some funky things.

324
00:19:55,493 --> 00:19:58,463
Josh:
So Ledger has this three-step approach to solving this. The first is that the

325
00:19:58,463 --> 00:20:01,601
Josh:
agent proposes a change, then the human approves the change,

326
00:20:01,891 --> 00:20:04,703
Josh:
and then the Ledger signer actually enforces this change.

327
00:20:04,703 --> 00:20:08,393
Josh:
There's a three-step process to make sure your agent doesn't do anything you

328
00:20:08,393 --> 00:20:11,573
Josh:
do not want it to do they have this thing called the ledger agent stack which

329
00:20:11,573 --> 00:20:15,057
Josh:
is an open source software stack that gives you a series of tools to help you navigate

330
00:20:15,301 --> 00:20:20,224
Josh:
your journey with your agents it works with cloud code it works with codex cursor anywhere that you use

331
00:20:20,555 --> 00:20:23,883
Josh:
your ai models it's available now it is open source you can find the link in

332
00:20:23,883 --> 00:20:28,298
Josh:
the description down below and thank you so much to ledger for sponsoring this episode the agent

333
00:20:28,705 --> 00:20:32,191
Josh:
barrage? I guess we have to talk about the downside effects,

334
00:20:32,191 --> 00:20:36,832
Josh:
right? What could happen if things don't go as well as planned?

335
00:20:37,157 --> 00:20:41,021
Josh:
And where does that risk actually live? So, Ijaz, it seems like this has been

336
00:20:41,021 --> 00:20:43,791
Josh:
fully prepared, lovely by our clawed artifact right here. So,

337
00:20:44,171 --> 00:20:48,201
Josh:
what is actually the downside risk? What do we need to look out for when we're

338
00:20:48,201 --> 00:20:49,653
Josh:
evaluating how to invest around this?

339
00:20:49,844 --> 00:20:53,021
Ejaaz:
Okay. So, we are the Limitless Show. And as anyone who's listened to us for

340
00:20:53,021 --> 00:20:57,991
Ejaaz:
a while knows, we are grounded or we are trying to ground ourselves a lot more from the bullish case.

341
00:20:58,031 --> 00:20:59,399
Josh:
I mean, if you got to say it. Yeah.

342
00:20:59,794 --> 00:21:04,621
Ejaaz:
So there are a few ways where this can obviously go wrong.

343
00:21:04,621 --> 00:21:07,861
Ejaaz:
And I want to kind of like walk through some of these and get your take on this,

344
00:21:07,861 --> 00:21:13,680
Ejaaz:
Josh. So number one, the thing that's like blaring to me is this is all based

345
00:21:13,999 --> 00:21:17,772
Ejaaz:
on the fact that AI demand not only is sustained.

346
00:21:17,923 --> 00:21:21,487
Ejaaz:
So you have paying customers to buy cloud subscriptions, GPT subscriptions,

347
00:21:21,766 --> 00:21:25,847
Ejaaz:
companies paying tens to hundreds of millions of dollars a year for API access. But...

348
00:21:26,276 --> 00:21:29,527
Ejaaz:
That it increases. Right now, it's increasing at a crazy rate.

349
00:21:29,742 --> 00:21:33,985
Ejaaz:
We see all these quarterly earnings, revenues compounded between 100% to 500% year upon year.

350
00:21:34,160 --> 00:21:37,171
Ejaaz:
It is insane, but that's not sustainable. It's not going to keep doing that.

351
00:21:37,171 --> 00:21:39,634
Ejaaz:
It'll presumably eventually plateau.

352
00:21:39,895 --> 00:21:43,477
Ejaaz:
So if that does plateau, or in worst case, if that plummets,

353
00:21:43,831 --> 00:21:48,391
Ejaaz:
then these companies are going to need fewer GPUs, which means that Jensen's

354
00:21:48,391 --> 00:21:51,933
Ejaaz:
$500 billion, these debt-backed securities are going to be in less demand.

355
00:21:52,125 --> 00:21:54,284
Ejaaz:
And that's where you might see a default.

356
00:21:54,574 --> 00:22:01,121
Ejaaz:
The second major thing here is that the GPUs themselves depreciate a lot faster.

357
00:22:01,121 --> 00:22:06,475
Ejaaz:
And that has been the Michael Burry, the guy that did the famous big short back in 2008.

358
00:22:06,846 --> 00:22:10,651
Ejaaz:
That's been his view this entire time. He says that the upgrade cycle for a

359
00:22:10,721 --> 00:22:15,983
Ejaaz:
lot of these NVIDIA GPUs are actually a lot shorter than what Jensen NVIDIA claims.

360
00:22:16,163 --> 00:22:19,241
Ejaaz:
However, in practicality, this seems to not be the case.

361
00:22:19,241 --> 00:22:25,080
Ejaaz:
However, NVIDIA is now releasing a lot of GPUs at a much more higher frequency

362
00:22:25,207 --> 00:22:29,691
Ejaaz:
rate, which means that they're going to replace more of the GPUs in the prior

363
00:22:29,691 --> 00:22:32,609
Ejaaz:
market, in the prior cycle, and they'll start flooding the market.

364
00:22:32,795 --> 00:22:36,643
Ejaaz:
My countess of that is simply you can't make GPUs that quickly.

365
00:22:36,759 --> 00:22:40,881
Ejaaz:
It takes a lot of technical expertise, and it is limited by the likes of TSMC

366
00:22:40,881 --> 00:22:44,126
Ejaaz:
and wafer capacity and a bunch of other technical stuff, which I don't want to get into on this show.

367
00:22:44,339 --> 00:22:48,531
Ejaaz:
So I'm struggling to actually see how these two factors might actually be triggered.

368
00:22:48,531 --> 00:22:50,272
Ejaaz:
But I don't know if you have a different opinion, Josh.

369
00:22:50,528 --> 00:22:54,441
Josh:
Yeah, the thing that I'm looking out for most is the return on invested capital

370
00:22:54,441 --> 00:22:55,833
Josh:
from the large hyperscalers.

371
00:22:56,083 --> 00:22:58,301
Josh:
It feels like they just run the world. They're spending all the capex.

372
00:22:58,301 --> 00:23:00,709
Josh:
They are basically floating the entire economy right now.

373
00:23:01,046 --> 00:23:04,819
Josh:
And if the returns on that investment start to go down, for example,

374
00:23:05,046 --> 00:23:08,481
Josh:
that seems like a very scary thing. So looking at Google's earnings reports,

375
00:23:08,481 --> 00:23:13,821
Josh:
we see like, okay, They have $514 billion dollars.

376
00:23:15,048 --> 00:23:18,955
Josh:
They're spending. Can they keep returning revenue on that on schedule?

377
00:23:19,181 --> 00:23:23,842
Josh:
If the answer is yes, if there's still revenue to be made on AI spend, that is amazing.

378
00:23:24,202 --> 00:23:27,385
Josh:
In the case that that turns and we start seeing earnings reports from companies

379
00:23:27,385 --> 00:23:31,055
Josh:
who are spending huge amounts of capex saying, our margins are actually shrinking.

380
00:23:31,055 --> 00:23:34,814
Josh:
Our revenue is not coming at the multiple that we expected.

381
00:23:35,046 --> 00:23:38,436
Josh:
That seems to be a red flag because that will slow down spending significantly across the board.

382
00:23:38,657 --> 00:23:40,520
Josh:
Basically, we want to make sure that all this stays profitable.

383
00:23:40,706 --> 00:23:43,694
Josh:
So we want to make sure that Inference demand is actually continuing.

384
00:23:43,886 --> 00:23:46,074
Josh:
Companies are actually able to meaningfully monetize this.

385
00:23:46,364 --> 00:23:49,885
Josh:
The enterprises that are spending billions, hundreds of billions of dollars

386
00:23:49,885 --> 00:23:52,925
Josh:
a year on AI spend, we need to make sure that they're actually getting value.

387
00:23:52,925 --> 00:23:55,908
Josh:
Otherwise, they're going to cut those contracts. That is probably the most important thing.

388
00:23:56,227 --> 00:23:58,810
Josh:
The second is just monitoring the rental rates. Like currently,

389
00:23:59,339 --> 00:24:02,295
Josh:
a lot of companies are terrified to re-sign their long-term GP rental deals

390
00:24:02,295 --> 00:24:05,215
Josh:
because the price that they're going to get them at this time around is going

391
00:24:05,215 --> 00:24:06,914
Josh:
to be double the price that they got originally.

392
00:24:07,338 --> 00:24:10,885
Josh:
That is a phenomenon that like no one was really expecting, but here we are.

393
00:24:11,245 --> 00:24:15,505
Josh:
And if that trend continues as well that's something i'm kind of looking out

394
00:24:15,505 --> 00:24:17,765
Josh:
for so i'm looking at what is the hour double

395
00:24:17,765 --> 00:24:19,625
Ejaaz:
Like three years from now right i mean it.

396
00:24:19,625 --> 00:24:23,125
Josh:
Might it might but i'm saying this this is just something that like you should

397
00:24:23,125 --> 00:24:24,605
Josh:
keep a close eye on you know how long these

398
00:24:24,605 --> 00:24:26,945
Ejaaz:
Contracts are for josh that they're signing i know.

399
00:24:26,945 --> 00:24:31,405
Josh:
They vary quite a bit like some are short they're like a year some are longer

400
00:24:31,405 --> 00:24:36,975
Josh:
out to like three years maybe um but i they're they're variable and i know that

401
00:24:36,975 --> 00:24:40,659
Josh:
when the time has is coming to kind of re-sign this contract

402
00:24:40,935 --> 00:24:44,760
Josh:
the price is higher not lower for the same supply so ensuring that this

403
00:24:45,376 --> 00:24:48,744
Josh:
continues this trend continues and even if it doesn't continue making sure it

404
00:24:48,744 --> 00:24:52,684
Josh:
doesn't flip negative because that might change things and granted nvidia has

405
00:24:52,684 --> 00:24:58,274
Josh:
your 25 plunge protection service you're available but you don't really want to put that 125

406
00:24:58,274 --> 00:25:01,704
Ejaaz:
Billion dollars by the way for those of you trying to do the math.

407
00:25:01,704 --> 00:25:04,863
Josh:
That's a lot of money man a lot of money um and then third is just like

408
00:25:05,293 --> 00:25:10,334
Josh:
what the yields actually is from these gpus like how much the real yield yeah

409
00:25:10,334 --> 00:25:11,429
Josh:
the actual real yield and

410
00:25:11,742 --> 00:25:15,364
Josh:
these are also like those mous this is an assigned deal and we've had something

411
00:25:15,364 --> 00:25:21,244
Josh:
similar to this before remember that crazy project back in the day called project stargate where elon and

412
00:25:21,564 --> 00:25:27,444
Josh:
masa or not sorry not elon sam altman and masayoshi son and even donald trump

413
00:25:27,444 --> 00:25:29,099
Josh:
all stood in an office together and they said

414
00:25:29,424 --> 00:25:32,229
Josh:
we're going to spend x billion dollars on this data center build out

415
00:25:32,663 --> 00:25:36,669
Josh:
it hasn't really happened as planned so this is not a contractual obligation

416
00:25:36,924 --> 00:25:39,506
Josh:
to spend 500 billion dollars this is a hey dude

417
00:25:40,150 --> 00:25:43,453
Josh:
We're all rich. We'll commit to like $500 billion and we'll see how it goes.

418
00:25:43,691 --> 00:25:47,754
Josh:
And that's kind of what they have. It's a handshake deal to build this new financial

419
00:25:47,754 --> 00:25:51,274
Josh:
economic instrument around the GPU, particularly as it relates to NVIDIA.

420
00:25:51,274 --> 00:25:56,414
Josh:
So huge win for NVIDIA, probably a large win for a lot of the companies that

421
00:25:56,414 --> 00:25:59,944
Josh:
are not able to afford this, and probably a huge win for the banks.

422
00:25:59,944 --> 00:26:04,119
Josh:
At the end of the day, they seem to always win. And that's kind of what the deal is here.

423
00:26:04,854 --> 00:26:10,290
Ejaaz:
I am really struggling to think about a world, an alternative scenario,

424
00:26:10,650 --> 00:26:19,061
Ejaaz:
where AI doesn't require GPUs, specifically the monopolistic GPUs from NVIDIA.

425
00:26:19,328 --> 00:26:23,160
Ejaaz:
They just have such a stronghold on the entire market.

426
00:26:23,525 --> 00:26:28,104
Ejaaz:
And even if you have some kind of novel LLM architecture that gets created in

427
00:26:28,104 --> 00:26:31,394
Ejaaz:
the future that completely disrupts the current paradigm, you're still going

428
00:26:31,394 --> 00:26:32,884
Ejaaz:
to need hardware to run these things.

429
00:26:32,884 --> 00:26:37,304
Ejaaz:
And that hardware is very much GPUs that are being designed and created by Jensen Huang.

430
00:26:37,304 --> 00:26:42,184
Ejaaz:
So however way I skin this cat, I still think that you're going to need these GPUs.

431
00:26:42,184 --> 00:26:46,104
Ejaaz:
You still need token generation. Gavin Baker has made this point across so many

432
00:26:46,104 --> 00:26:49,794
Ejaaz:
other podcast episodes in the last two weeks that it's ingrained in my head at this point, right?

433
00:26:49,794 --> 00:26:54,794
Ejaaz:
And then the other thing I think about is, okay, well, if NVIDIA becomes a bank

434
00:26:54,794 --> 00:26:58,702
Ejaaz:
themselves and they start taking revenue splits from all these frontier AI labs,

435
00:26:59,097 --> 00:27:01,734
Ejaaz:
that's a completely new revenue line for NVIDIA.

436
00:27:01,734 --> 00:27:04,850
Ejaaz:
So when I think about this with my investing hat on, I'm thinking.

437
00:27:05,471 --> 00:27:10,267
Ejaaz:
Okay, not only is NVIDIA supplying the foundational element that is required

438
00:27:10,267 --> 00:27:14,225
Ejaaz:
to run and inference these GPUs, train these GPUs, but they're also

439
00:27:14,660 --> 00:27:19,257
Ejaaz:
being the ones that are driving cost down per token, right? So like they're

440
00:27:19,257 --> 00:27:20,686
Ejaaz:
doing this with their CUDA software mode.

441
00:27:20,970 --> 00:27:24,927
Ejaaz:
And then I think about the financing side of things. So they're being the financiers

442
00:27:24,927 --> 00:27:26,047
Ejaaz:
of this entire thing as well.

443
00:27:26,047 --> 00:27:31,237
Ejaaz:
Now, that does sound like a house of cards if the demand wavers, if the demand plummets.

444
00:27:31,237 --> 00:27:34,557
Ejaaz:
And I can easily see the market being very volatile and reacting to any kind

445
00:27:34,557 --> 00:27:37,687
Ejaaz:
of headline like they did to this initial headline but i don't know it just

446
00:27:37,687 --> 00:27:41,267
Ejaaz:
seems very bullish to me on nvidia at least and yeah i don't really know how

447
00:27:41,267 --> 00:27:42,367
Ejaaz:
to think about it yeah yeah.

448
00:27:42,367 --> 00:27:46,377
Josh:
And i mean like in this case like i do kind of lean on the opinions of people

449
00:27:46,377 --> 00:27:50,607
Josh:
who are more in the know than me yeah someone like elon who is now exclusively

450
00:27:50,607 --> 00:27:54,423
Josh:
committed to purchasing only nvidia gpus for the new data center build out

451
00:27:54,824 --> 00:27:58,087
Josh:
and they are effectively the best data center builders in the world so i you

452
00:27:58,087 --> 00:28:00,397
Josh:
have to like have a little bit of trust in the opinions of the

453
00:28:00,717 --> 00:28:04,137
Josh:
true experts who are in the arena doing things when i look at that and i see

454
00:28:04,137 --> 00:28:07,757
Josh:
like they exclusively want nvidia and they are building the best fastest most

455
00:28:07,757 --> 00:28:11,297
Josh:
efficient data centers i'm like okay that's pretty good signal like micro hard

456
00:28:11,297 --> 00:28:13,969
Josh:
the new data center that the spacex ai team is working on

457
00:28:14,276 --> 00:28:17,107
Josh:
is i think like a third the footprint of macro

458
00:28:17,107 --> 00:28:18,557
Ejaaz:
Hard right no micro hard.

459
00:28:18,977 --> 00:28:20,337
Josh:
No no there is there's micro hard is

460
00:28:20,337 --> 00:28:21,039
Ejaaz:
A micro hot.

461
00:28:21,300 --> 00:28:24,057
Josh:
Yes and micro hard is about a third of the footprint i might be getting this

462
00:28:24,057 --> 00:28:27,867
Josh:
wrong half or a third of the footprint of macro hard easy but it contains the

463
00:28:27,867 --> 00:28:31,627
Josh:
same cluster of 200 000 gpus wow they just figured out how to do it much more

464
00:28:31,627 --> 00:28:33,757
Josh:
efficiently and much more dense so

465
00:28:34,197 --> 00:28:37,397
Josh:
there's a huge amount of innovation clearly they know things that the rest of

466
00:28:37,397 --> 00:28:40,057
Josh:
the industry does not and when they come out the biggest purchaser right

467
00:28:40,548 --> 00:28:46,257
Josh:
yeah and they're committed exclusively to the nvidia gpu wow and when vera rubin

468
00:28:46,257 --> 00:28:49,717
Josh:
comes out at scale man oh my god i keep saying this for like holy smokes

469
00:28:49,717 --> 00:28:52,617
Ejaaz:
Those models are gonna be insane those models are gonna be absolutely insane.

470
00:28:52,959 --> 00:28:57,537
Josh:
So buckle up good time to be good time to be nvidia good time to be a GPU.

471
00:28:58,517 --> 00:29:01,649
Josh:
Yeah, that's the update. So is this a house of cards?

472
00:29:02,091 --> 00:29:06,043
Josh:
Is it all going to come tumbling down? Is this financial innovation in a new

473
00:29:06,043 --> 00:29:08,865
Josh:
era of the United States of GPUs?

474
00:29:09,120 --> 00:29:11,583
Josh:
Let us know in the comments down below if you enjoyed this episode.

475
00:29:11,583 --> 00:29:14,508
Josh:
Don't forget to share it with a friend who might also enjoy this episode.

476
00:29:14,751 --> 00:29:17,712
Josh:
Ejaz and I, cool story. We were walking down the street last night after dinner

477
00:29:17,927 --> 00:29:19,073
Josh:
and two people walked up to us.

478
00:29:19,073 --> 00:29:21,883
Josh:
They were like, hey, you guys are those podcast guys. You host the show.

479
00:29:21,883 --> 00:29:24,313
Josh:
And we were like, yeah, we do. Like, cool. Thanks for watching.

480
00:29:24,313 --> 00:29:28,963
Josh:
So that's like so cool whenever that happens. And that is because of you sharing

481
00:29:28,963 --> 00:29:31,383
Josh:
with your friends, letting everyone else know that the show exists.

482
00:29:31,758 --> 00:29:34,583
Josh:
And if you enjoyed it, you know, don't forget. Thumbs up. You could subscribe.

483
00:29:34,583 --> 00:29:36,833
Josh:
You could give us a five-star review on your favorite podcast platform.

484
00:29:37,293 --> 00:29:38,259
Josh:
Any parting thoughts, EJs?

485
00:29:38,596 --> 00:29:41,829
Ejaaz:
Yeah. Homework for you guys. If you see us on the street, don't be shy.

486
00:29:41,974 --> 00:29:43,353
Ejaaz:
Come and say hi. We want to meet you guys.

487
00:29:43,353 --> 00:29:44,273
Josh:
Don't be shy. Say hi.

488
00:29:44,653 --> 00:29:48,993
Ejaaz:
Yeah. It's lovely to meet you guys. We've now met people. We've now met listeners,

489
00:29:49,873 --> 00:29:52,987
Ejaaz:
in New York and we've met them in San Francisco.

490
00:29:53,613 --> 00:29:56,903
Ejaaz:
I have people calling in from Europe to my family talking about these random

491
00:29:56,903 --> 00:30:02,123
Ejaaz:
guys that yap about AI. Turns out it's us. Like it's really cool to see the

492
00:30:02,123 --> 00:30:03,563
Ejaaz:
momentum that we're getting here.

493
00:30:03,749 --> 00:30:07,923
Ejaaz:
And it's all thanks to you guys. So if you're one of these people that care

494
00:30:07,923 --> 00:30:11,853
Ejaaz:
passionately about what we talk about and tune in every day, thank you so much.

495
00:30:12,126 --> 00:30:15,785
Ejaaz:
And if you're not, turn on notifications, please. Subscribe to us.

496
00:30:15,971 --> 00:30:19,623
Ejaaz:
We bring the best news, hot, fresh out the oven, every single day,

497
00:30:19,623 --> 00:30:20,876
Ejaaz:
or rather four times a week.

498
00:30:21,137 --> 00:30:24,243
Ejaaz:
And we would love to hear from you. Leave us a comment, DM us on X.

499
00:30:24,243 --> 00:30:26,664
Ejaaz:
And yeah, I think that's it. Thank you so much for listening.

500
00:30:27,070 --> 00:30:27,813
Josh:
See you next time.