1
00:00:00,280 --> 00:00:03,800
In recent weeks, there's been a 
lot of market anxiety about the 

2
00:00:03,800 --> 00:00:08,160
sustainability of the AI boom. 
This was partly driven by the 

3
00:00:08,160 --> 00:00:12,880
outrage around Sarah Fryer, Open
AI finance chief, floating the 

4
00:00:12,880 --> 00:00:16,960
idea that a government backstop 
for its $1.4 trillion data 

5
00:00:16,960 --> 00:00:19,520
center build out might be a good
idea. 

6
00:00:19,960 --> 00:00:23,680
Fryer quickly walked back her 
suggestion in the LinkedIn post 

7
00:00:23,680 --> 00:00:27,760
later that day, saying that she 
had meant that the government 

8
00:00:27,760 --> 00:00:31,360
needed to play their part in 
combination with the private 

9
00:00:31,360 --> 00:00:35,280
sector to contribute to 
America's AI growth and that 

10
00:00:35,280 --> 00:00:38,880
Open AI was not seeking a 
government backstop for their 

11
00:00:38,880 --> 00:00:42,880
infrastructure commitments. 
Her statement, while attempting 

12
00:00:42,880 --> 00:00:46,960
to calm the outrage, only 
confused matters even further 

13
00:00:47,000 --> 00:00:51,240
about how the not yet profitable
start up plans to pay for its 

14
00:00:51,240 --> 00:00:54,560
massive AI data centre and chip 
commitments. 

15
00:00:54,960 --> 00:00:57,520
Sam Altman tweeted on the 
Everything app. 

16
00:00:57,800 --> 00:01:01,600
We do not have or want 
government guarantees for open 

17
00:01:01,720 --> 00:01:04,840
AI data centres. 
We believe that governments 

18
00:01:04,840 --> 00:01:08,480
should not pick winners or 
losers and the taxpayers should 

19
00:01:08,480 --> 00:01:12,280
not bail out companies that make
bad business decisions or 

20
00:01:12,280 --> 00:01:15,840
otherwise lose in the market. 
Then it turned into a Bill 

21
00:01:15,880 --> 00:01:19,960
Aikman tweet at that point where
he went on and on for around 20 

22
00:01:19,960 --> 00:01:22,440
pages. 
At first I was thinking, who 

23
00:01:22,440 --> 00:01:25,040
would write a tweet that long, 
and then I realized he he'd 

24
00:01:25,040 --> 00:01:28,920
probably just used ChatGPT. 
He knew that people would only 

25
00:01:28,920 --> 00:01:32,240
read the first few lines, but he
wanted to seem thoughtful. 

26
00:01:32,400 --> 00:01:34,520
So how to churn out an entire 
novel? 

27
00:01:34,760 --> 00:01:38,960
The core problem for Open AI is 
that they've signed more than 

28
00:01:38,960 --> 00:01:43,080
$1.4 trillion in infrastructure 
commitments over the last few 

29
00:01:43,080 --> 00:01:46,600
months with the goal of building
out the data centers that it 

30
00:01:46,600 --> 00:01:49,240
says are needed to meet soaring 
demand. 

31
00:01:49,480 --> 00:01:52,800
But they are nowhere near having
the money required to complete 

32
00:01:52,800 --> 00:01:55,760
those deals. 
FIRE gave the example of having 

33
00:01:55,760 --> 00:02:00,200
to hold back Sora 2 for months 
due to compute constraints. 

34
00:02:00,280 --> 00:02:02,720
I just want to be clear what it 
means when I say we're compute 

35
00:02:02,720 --> 00:02:05,720
constrained. 
It means that for example, we 

36
00:02:05,720 --> 00:02:09,080
cannot roll out our new models 
when they are ready. 

37
00:02:09,080 --> 00:02:13,280
So when Sora 2 was ready to when
Sora 2 actually launched, there 

38
00:02:13,280 --> 00:02:17,080
was probably a good 6-7 months 
actually gap there. 

39
00:02:21,960 --> 00:02:25,760
OK, the agreements that they've 
signed have raised lots of 

40
00:02:25,760 --> 00:02:29,080
questions around how a cash 
burning company with tiny 

41
00:02:29,080 --> 00:02:33,240
revenues relative to their plant
spending can possibly make such 

42
00:02:33,240 --> 00:02:36,720
huge commitment. 
This was not the first time Open

43
00:02:36,720 --> 00:02:39,040
AI has looked to Washington for 
help, either. 

44
00:02:39,280 --> 00:02:42,840
Just a month ago, the company 
sent a detailed letter to the 

45
00:02:42,840 --> 00:02:46,360
White House urging the federal 
government to double down ON 

46
00:02:46,360 --> 00:02:51,000
Semiconductor subsidies, asking 
the tax credits be expanded to 

47
00:02:51,000 --> 00:02:55,280
cover the entire AI supply 
chain, from chip fabrication to 

48
00:02:55,280 --> 00:02:59,520
data centres and grid hardware. 
The company argued that 

49
00:02:59,520 --> 00:03:03,400
broadening eligibility for 
taxpayer funded subsidies would 

50
00:03:03,400 --> 00:03:08,080
lower the effective cost of 
capital de risk early investment

51
00:03:08,080 --> 00:03:12,040
and unlock private capital. 
Open AI and it's data centre 

52
00:03:12,040 --> 00:03:15,680
partners are of course amongst 
the largest buyers of 

53
00:03:15,680 --> 00:03:19,400
semiconductors in the world, so 
any subsidy would directly 

54
00:03:19,400 --> 00:03:22,200
benefit them. 
AI is being pitched to 

55
00:03:22,200 --> 00:03:25,960
governments around the world as 
being a matter of grave national

56
00:03:25,960 --> 00:03:29,520
security and economic 
importance, akin to past 

57
00:03:29,520 --> 00:03:33,000
industrial mobilizations like 
the Manhattan Project and the 

58
00:03:33,000 --> 00:03:36,440
Space Race. 
If AI companies can put it on 

59
00:03:36,440 --> 00:03:40,040
that level and pitch it as being
too important to fail, a 

60
00:03:40,040 --> 00:03:43,000
government funded backstop might
make sense. 

61
00:03:43,680 --> 00:03:47,760
The irony though is that while 
lobbying for taxpayer support in

62
00:03:47,760 --> 00:03:51,200
the name of geopolitical 
survival, the same businesses 

63
00:03:51,360 --> 00:03:54,600
are pumping billions into 
building models that generate 

64
00:03:54,600 --> 00:03:58,440
weird anime girlfriends, 
SpongeBob deepfakes, Sam 

65
00:03:58,440 --> 00:04:03,360
Altman's Studio Ghibli style 
profile photo and in Elon Musk's

66
00:04:03,360 --> 00:04:07,160
case, a chatbot that appears to 
have been hard coded this week 

67
00:04:07,320 --> 00:04:11,000
to constantly flatter him in the
cringiest manner possible. 

68
00:04:11,200 --> 00:04:14,440
Which caused all sorts of 
hilarity on the everything app 

69
00:04:14,440 --> 00:04:16,760
formerly known as Twitter this 
week. 

70
00:04:17,000 --> 00:04:19,360
We'll come back to that in just 
a minute though. 

71
00:04:19,720 --> 00:04:23,480
While all of this was going on, 
NVIDIA warned for the first time

72
00:04:23,480 --> 00:04:27,560
in a regulatory filing that it's
customers ability to secure 

73
00:04:27,560 --> 00:04:32,080
capital and energy for AI data 
centres could potentially slow 

74
00:04:32,080 --> 00:04:35,360
it's growth. 
On top of that, Amazon launched 

75
00:04:35,360 --> 00:04:38,800
A complaint with the Public 
Utility Commission of Oregon 

76
00:04:39,000 --> 00:04:42,520
that the electric utility was 
failing to provide sufficient 

77
00:04:42,520 --> 00:04:45,240
power for the four new data 
centres that had built, 

78
00:04:45,480 --> 00:04:49,280
highlighting the strain that 
rapid data centre expansion is 

79
00:04:49,280 --> 00:04:52,920
putting on electric grids. 
I guess the utility agreed to 

80
00:04:52,920 --> 00:04:56,040
hook them up to the grid, but 
not necessarily to provide them 

81
00:04:56,040 --> 00:05:00,000
with the power that they wanted.
The question hanging over 

82
00:05:00,000 --> 00:05:04,000
Silicon Valley is not so much 
whether AI will change the 

83
00:05:04,000 --> 00:05:07,080
world, but whether the world can
afford to build it. 

84
00:05:07,640 --> 00:05:11,000
This brings us to Nvidia's 
earnings report on Wednesday 

85
00:05:11,000 --> 00:05:13,920
night. 
The tech rally that has defined 

86
00:05:13,920 --> 00:05:17,840
much of 2025, especially after 
deliberation day sell off in 

87
00:05:17,840 --> 00:05:21,160
April, began to lose momentum in
early autumn. 

88
00:05:21,560 --> 00:05:25,600
Some analysts traced the 
inflection point to when Open AI

89
00:05:25,600 --> 00:05:30,280
announced a $300 billion cloud 
deal with Oracle and NVIDIA 

90
00:05:30,280 --> 00:05:34,200
pledged up to 100 billion in 
reciprocal investments. 

91
00:05:34,480 --> 00:05:37,560
Those headlines, which were 
meant to signal confidence, 

92
00:05:37,720 --> 00:05:42,320
instead raised questions about 
circular financing and the sheer

93
00:05:42,320 --> 00:05:46,440
scale of spending commitments. 
Private credit blow UPS added to

94
00:05:46,440 --> 00:05:50,880
the unease in markets, reviving 
concerns about lending standards

95
00:05:50,880 --> 00:05:54,120
and fraud in a market already 
stretched by aggressive 

96
00:05:54,120 --> 00:05:57,320
leverage. 
Valuations were lofty, and the 

97
00:05:57,320 --> 00:06:00,040
spaghetti diagrams of 
interlocking deals with 

98
00:06:00,040 --> 00:06:04,200
hyperscalers funding AI labs 
that fund chip makers that fund 

99
00:06:04,200 --> 00:06:08,080
hyperscalers started looking 
increasingly fragile. 

100
00:06:08,280 --> 00:06:11,760
No surprise, then, that bubble 
talk intensified. 

101
00:06:12,280 --> 00:06:16,200
Nvidia's earnings report on 
Wednesday temporarily eased 

102
00:06:16,200 --> 00:06:18,720
those fears. 
The world's most valuable 

103
00:06:18,720 --> 00:06:23,760
company and the beating heart of
the AI trade posted a 62% jump 

104
00:06:23,760 --> 00:06:27,800
in revenue for the three months 
to October, far ahead of 

105
00:06:27,800 --> 00:06:32,080
expectations. 
Data center sales hit $51.2 

106
00:06:32,080 --> 00:06:36,120
billion, and the company raised 
its revenue forecast for the 

107
00:06:36,120 --> 00:06:41,520
current quarter to $65 billion. 
For now, the numbers seem to 

108
00:06:41,520 --> 00:06:44,760
just by the hype. 
As Robert Armstrong put it on 

109
00:06:44,760 --> 00:06:48,640
the Unhedged podcast, the worry 
is not invidious price to 

110
00:06:48,640 --> 00:06:51,960
earnings ratio. 
The worry is that the revenue 

111
00:06:51,960 --> 00:06:54,960
it's earning, and the growth 
rate of that revenue is 

112
00:06:54,960 --> 00:06:58,680
ultimately unsustainable. 
At today's pace, Nvidia's 

113
00:06:58,680 --> 00:07:02,040
valuation makes sense. 
The question is whether the 

114
00:07:02,040 --> 00:07:04,880
growth curve can defy gravity 
indefinitely. 

115
00:07:05,000 --> 00:07:09,080
Open AI's finances look even 
more precarious than most people

116
00:07:09,080 --> 00:07:11,880
realize. 
Microsoft's September earning 

117
00:07:12,080 --> 00:07:16,640
filing revealed that Open AI 
lost roughly 11 1/2 billion 

118
00:07:16,640 --> 00:07:20,320
dollars in a single quarter, 
it's worst on record. 

119
00:07:20,560 --> 00:07:25,760
That pushes year to date losses 
north of $25 billion against 

120
00:07:25,760 --> 00:07:29,800
projected annual revenue of 
about $20 billion. 

121
00:07:30,080 --> 00:07:34,960
The company has raised nearly 
$58 billion in equity so far and

122
00:07:34,960 --> 00:07:38,280
was valued at $500 billion last 
month. 

123
00:07:38,760 --> 00:07:43,560
The company is talking about an 
IPO at a $1 trillion valuation 

124
00:07:43,560 --> 00:07:47,400
next year, which would float the
shares on an exchange and 

125
00:07:47,400 --> 00:07:50,800
possibly bring in about $60 
billion in cash. 

126
00:07:51,040 --> 00:07:55,920
But that's just over 4% of it's 
$1.4 trillion infrastructure 

127
00:07:55,920 --> 00:07:59,480
commitment. 
To bridge the gap, Open AI has 

128
00:07:59,480 --> 00:08:01,920
leaned on creative deal 
structures. 

129
00:08:02,520 --> 00:08:06,400
NVIDIA has pledged up to $100 
billion in reciprocal 

130
00:08:06,400 --> 00:08:11,800
investments, while AMD granted 
Open AI warrants to buy 10% of 

131
00:08:11,800 --> 00:08:15,720
its stock for a penny per share 
if deployment milestones are 

132
00:08:15,720 --> 00:08:18,960
mad. 
Sarah Fryer, Open AICFO, 

133
00:08:19,120 --> 00:08:21,920
explained the company's 
financing at the Wall Street 

134
00:08:21,920 --> 00:08:24,400
Journal event. 
While she's from Northern 

135
00:08:24,400 --> 00:08:27,520
Ireland, she must have been in 
Silicon Valley long enough to 

136
00:08:27,520 --> 00:08:31,400
know that the first step in 
raising capital is to use the 

137
00:08:31,400 --> 00:08:33,880
magic word. 
The innovation on the finance 

138
00:08:33,880 --> 00:08:37,559
side to pay for it is massive. 
She then went on to say. 

139
00:08:37,760 --> 00:08:41,039
We've raised equity as a private
company, very kind of typical 

140
00:08:41,039 --> 00:08:44,920
path, but we've raised a lot. 
We're building a really healthy 

141
00:08:44,920 --> 00:08:47,000
business. 
So free cash flow, CFO's 

142
00:08:47,240 --> 00:08:51,520
favorite way to fund anything 
that is absolutely climbing 

143
00:08:51,520 --> 00:08:54,000
quickly. 
But I think the third area we've

144
00:08:54,000 --> 00:08:57,600
gotten into is really working 
with our ecosystem to do some 

145
00:08:57,600 --> 00:08:59,720
really interesting financing 
deals. 

146
00:08:59,720 --> 00:09:02,800
I'm particularly proud of the 
AMD warrant structure that we 

147
00:09:02,800 --> 00:09:06,000
put in place just a few weeks 
back because it's very strong 

148
00:09:06,000 --> 00:09:09,600
alignment of incentives. 
This was a really bizarre claim 

149
00:09:09,600 --> 00:09:13,040
as Open AI can't fund anything 
with free cash flow. 

150
00:09:13,040 --> 00:09:16,920
When that cash flow is negative,
she then digs into explaining 

151
00:09:17,120 --> 00:09:20,440
the AMD warrant. 
What we've seen is when someone 

152
00:09:20,440 --> 00:09:23,640
comes out and says we're going 
to work with Open AI, they admit

153
00:09:23,640 --> 00:09:26,800
immediately are often seeing 
kind of impact on their stock 

154
00:09:26,800 --> 00:09:29,000
price. 
And so to the extent that that's

155
00:09:29,000 --> 00:09:31,520
going to happen, we would like 
to have some alignment on that. 

156
00:09:31,760 --> 00:09:34,680
And I think Lisa and team did 
something incredibly creative 

157
00:09:34,680 --> 00:09:38,360
with that warrant structure. 
The warrant deal between Open AI

158
00:09:38,360 --> 00:09:42,440
and AMD is a strategic 
partnership where Open AI 

159
00:09:42,440 --> 00:09:46,640
commits to buying billions of 
dollars worth of AM Dai chips 

160
00:09:46,800 --> 00:09:52,040
and in return, AMD grants Open 
AI warrants to purchase up to 

161
00:09:52,040 --> 00:09:56,520
160 million of its shares, which
is about a 10% stake in the 

162
00:09:56,520 --> 00:10:00,680
company at a nominal price of 1 
cent per share. 

163
00:10:00,960 --> 00:10:06,600
When the deal was announced, AMD
stock went up at 24%, but the 

164
00:10:06,600 --> 00:10:12,400
deal only vests if Open AI buys 
6 gigawatts of AMD chips, hits 

165
00:10:12,440 --> 00:10:16,920
undisclosed milestones, and AM 
DS share price triples. 

166
00:10:17,480 --> 00:10:22,320
The AMD deal would bring in 
almost $100 billion worth of AMD

167
00:10:22,320 --> 00:10:26,760
stock if all of the targets were
hit, including the tripling of 

168
00:10:26,760 --> 00:10:31,200
AMD stock price. 
But it's tied to six gigawatts 

169
00:10:31,200 --> 00:10:34,240
of chip purchases, which she 
later explains. 

170
00:10:34,240 --> 00:10:38,400
So one GW data center build 
today is about a $50 billion 

171
00:10:38,400 --> 00:10:40,120
investment. 
That's for one gig. 

172
00:10:40,360 --> 00:10:44,560
How that really breaks down is 
about 15 billion is for the land

173
00:10:44,560 --> 00:10:47,400
PowerShell and about 35 billion 
is for the chips. 

174
00:10:47,840 --> 00:10:53,880
So to bring in $100 billion, 
they have to spend $300 billion.

175
00:10:54,240 --> 00:10:59,360
Nvidia's $100 billion pledge to 
invest in Open AI is also tied 

176
00:10:59,360 --> 00:11:03,240
to reciprocal commitments. 
If all of these deals worked 

177
00:11:03,240 --> 00:11:07,320
out, Open AI could bring in $200
billion. 

178
00:11:07,520 --> 00:11:11,480
But that still leaves them $1.2 
trillion short. 

179
00:11:11,760 --> 00:11:15,280
And they're burning 10s of 
billions of dollars per year 

180
00:11:15,360 --> 00:11:18,720
with no end. 
Insight The unit economics of 

181
00:11:18,720 --> 00:11:22,160
running the current generation 
of LLM is dire. 

182
00:11:22,640 --> 00:11:26,320
As Paul Kedrosky explained it on
the Odd Lots podcast. 

183
00:11:26,520 --> 00:11:30,680
The incentive seems to be for 
all players to just grow the top

184
00:11:30,680 --> 00:11:35,240
line as much as possible, even 
if adding more users just leads 

185
00:11:35,240 --> 00:11:39,560
to greater and greater losses. 
The models have negative unit 

186
00:11:39,560 --> 00:11:43,640
economics, which is a fancy way 
of saying we lose money on every

187
00:11:43,640 --> 00:11:46,120
sale and try to make it up on 
volume. 

188
00:11:46,480 --> 00:11:51,400
In AI, costs rise almost 
linearly with usage, which is 

189
00:11:51,400 --> 00:11:53,760
very different to traditional 
software. 

190
00:11:53,960 --> 00:11:56,960
There's no marginal cost magic 
going on. 

191
00:11:57,400 --> 00:12:02,120
According to Forbes, despite an 
invitation only roll out, Open 

192
00:12:02,200 --> 00:12:07,360
AI may be losing around $15 
million a day or $5 billion 

193
00:12:07,360 --> 00:12:11,800
annualized on Sora 2. 
It's AI video generating app. 

194
00:12:12,440 --> 00:12:16,000
Tech firms have always been 
creative about financing, but 

195
00:12:16,000 --> 00:12:19,800
Open AI's approach borders on 
the surreal, where it's become 

196
00:12:19,800 --> 00:12:22,880
all about trying to find 
infinite money glitches. 

197
00:12:23,120 --> 00:12:27,040
MicroStrategy, or Strategy as 
it's now called, is trying a 

198
00:12:27,040 --> 00:12:30,360
similar trick with its Bitcoin 
investments, which I don't 

199
00:12:30,360 --> 00:12:34,280
expect to end very well. 
Behind the headlines is a 

200
00:12:34,280 --> 00:12:38,040
financing structure that looks 
increasingly baroque. 

201
00:12:38,240 --> 00:12:42,600
Hyperscalers and AI labs are 
using special purpose vehicles 

202
00:12:42,760 --> 00:12:46,080
so that they can borrow but keep
the debt off their balance 

203
00:12:46,080 --> 00:12:48,520
sheets. 
Tech firms have essentially been

204
00:12:48,520 --> 00:12:52,920
reinventing structured finance 
to build AI models so that they 

205
00:12:52,920 --> 00:12:57,280
can generate AI girlfriends. 
That is just the world that we 

206
00:12:57,280 --> 00:13:06,000
live in and I will always love 
you, Sarah Fryer explained at 

207
00:13:06,000 --> 00:13:10,040
the Wall Street Journal Offend 
that each GW of compute costs 

208
00:13:10,040 --> 00:13:14,800
around $50 billion, where 15 
billion is the land and 

209
00:13:14,800 --> 00:13:19,320
infrastructure and $35 billion 
is the GPU's. 

210
00:13:19,560 --> 00:13:22,280
From a financing perspective, 
people know how to finance data 

211
00:13:22,280 --> 00:13:25,320
centres. 
They typically all have 2025, 

212
00:13:25,320 --> 00:13:29,160
even 30 year lives. 
Those are easy things. 

213
00:13:29,160 --> 00:13:32,720
I would say today to finance 
chips have not been as easy to 

214
00:13:32,720 --> 00:13:36,000
finance because number one, I 
think we're all still getting 

215
00:13:36,000 --> 00:13:39,760
our arms around what is the life
of a frontier chip. 

216
00:13:40,240 --> 00:13:43,880
What this means is that the more
innovation that happens with 

217
00:13:43,880 --> 00:13:47,600
chips, the faster they can be 
expected to depreciate. 

218
00:13:47,840 --> 00:13:52,280
And so the $35 billion worth of 
chips in a $50 billion data 

219
00:13:52,280 --> 00:13:54,880
centre are very difficult to 
finance. 

220
00:13:55,240 --> 00:13:59,240
People don't want to own them if
they might collapse in value 

221
00:13:59,240 --> 00:14:02,560
when a new one comes out, and 
people really don't want to 

222
00:14:02,560 --> 00:14:04,720
accept them as collateral on 
alone. 

223
00:14:05,200 --> 00:14:07,560
That's when she put forth this 
idea. 

224
00:14:07,840 --> 00:14:11,280
And so this is where we're 
looking for an ecosystem of 

225
00:14:11,360 --> 00:14:18,440
banks, private equity, maybe 
even governmental, the ways 

226
00:14:18,440 --> 00:14:21,200
governments can come to bear. 
Meaning like a federal subsidy 

227
00:14:21,200 --> 00:14:22,720
or. 
Something meaning like just 

228
00:14:22,720 --> 00:14:26,760
first of all, the the backstop, 
the guarantee that allows the 

229
00:14:26,760 --> 00:14:30,560
financing to happen, that can 
really drop the the cost of the 

230
00:14:30,560 --> 00:14:34,120
financing, but also increase the
the loan to value. 

231
00:14:34,120 --> 00:14:37,720
So the amount of debt that you 
can take on top of an equity 

232
00:14:37,960 --> 00:14:40,120
portion? 
For some federal backstop for 

233
00:14:40,120 --> 00:14:41,480
CHIP investment. 
Exactly. 

234
00:14:41,720 --> 00:14:44,560
And I think we're seeing that. 
I think the US government in 

235
00:14:44,560 --> 00:14:48,960
particular has been incredibly 
forward lean has really 

236
00:14:48,960 --> 00:14:54,840
understood that AI has is almost
a national strategic asset and 

237
00:14:54,840 --> 00:14:56,880
that we really need to be 
thoughtful when we think about 

238
00:14:56,880 --> 00:14:59,360
competitive competition with, 
for example, China. 

239
00:14:59,640 --> 00:15:04,080
Are we doing all the right 
things to grow our AI ecosystem 

240
00:15:04,400 --> 00:15:08,040
as fast as possible? 
Essentially, the problem is that

241
00:15:08,040 --> 00:15:12,120
they want to lever up their bet 
on AI, but banks wouldn't want 

242
00:15:12,120 --> 00:15:15,640
to lend, and the interest rate 
on a loan backed by rapidly 

243
00:15:15,640 --> 00:15:19,680
depreciating chips would be so 
high that you would need the 

244
00:15:19,680 --> 00:15:22,840
government to back the loan. 
Now I can tell that this will 

245
00:15:22,840 --> 00:15:26,480
make some of my viewers angry, 
but there's actually no need to 

246
00:15:26,480 --> 00:15:30,320
get angry about something like 
this, as both Sam Altman and 

247
00:15:30,320 --> 00:15:34,760
Elon Musk have both explained in
the past that AGI will soon make

248
00:15:34,760 --> 00:15:38,040
money obsolete, so who cares 
now? 

249
00:15:38,120 --> 00:15:42,000
Even if the money materializes 
and then suddenly doesn't matter

250
00:15:42,000 --> 00:15:45,760
anymore, the electrons may not 
open AI. 

251
00:15:45,760 --> 00:15:50,080
Stargate project alone would 
require 10 gigawatts of power, 

252
00:15:50,200 --> 00:15:52,760
which is roughly 10 nuclear 
power plants. 

253
00:15:52,960 --> 00:15:57,920
It's full build out implies 23. 
And that's just open AI. 

254
00:15:58,040 --> 00:16:01,400
Google has a model. 
Facebook, or whatever they call 

255
00:16:01,400 --> 00:16:05,800
themselves has one too. 
There's Grog, good old Grog, and

256
00:16:05,800 --> 00:16:10,040
Tropic, and lots, lots more. 
What I'm saying is we're going 

257
00:16:10,040 --> 00:16:13,480
to need a lot of power plants. 
We're going to need a bigger 

258
00:16:13,640 --> 00:16:18,760
boat, and we also have to plug 
in our cars and robots. 

259
00:16:19,040 --> 00:16:22,080
Only one new nuclear power 
station has been built in the 

260
00:16:22,080 --> 00:16:24,360
United States in the last 30 
years. 

261
00:16:24,640 --> 00:16:28,640
It took a decade to complete and
was the most expensive power 

262
00:16:28,640 --> 00:16:32,480
plant ever built. 
Bloomberg estimates that AI 

263
00:16:32,480 --> 00:16:36,240
driven electricity demand will 
more than double over the next 

264
00:16:36,240 --> 00:16:39,720
10 years. 
Utilities are already balking. 

265
00:16:39,960 --> 00:16:43,600
Amazon has filed A complaint 
against Pacificorp for failing 

266
00:16:43,600 --> 00:16:47,320
to deliver promised power to 4 
Oregon data centres. 

267
00:16:47,560 --> 00:16:51,280
Pacificorp says that it's 
protecting other customers from 

268
00:16:51,280 --> 00:16:55,800
indirect harms or translation. 
We can't turn the lights off in 

269
00:16:55,800 --> 00:17:00,000
Portland so that Jeff Bezos can 
train a chatbot behind the 

270
00:17:00,000 --> 00:17:02,720
meter. 
Gas turbines are proliferating, 

271
00:17:02,720 --> 00:17:06,599
too, as stop gaps. 
Some operators are whispering 

272
00:17:06,599 --> 00:17:10,720
about nuclear partnerships. 
These fixes create stranded 

273
00:17:10,720 --> 00:17:15,920
asset risk, as a natural gas 
plant lasts 30 years and a GPU 

274
00:17:15,920 --> 00:17:18,440
cluster might be obsolete in 18 
months. 

275
00:17:18,680 --> 00:17:21,920
Lenders see the mismatch, and 
they flinch. 

276
00:17:22,319 --> 00:17:26,640
Tech firms that promise to 
dematerialize the economy now 

277
00:17:26,640 --> 00:17:30,520
need more concrete, copper, and 
electricity than the steel mills

278
00:17:30,520 --> 00:17:32,800
did. 
The cloud, which was supposed to

279
00:17:32,800 --> 00:17:35,840
be weightless, turns out to be 
very heavy. 

280
00:17:36,480 --> 00:17:41,040
So why keep spending? 
Well, because the game is framed

281
00:17:41,040 --> 00:17:45,640
as being existential. 
US labs talk about sovereign AI 

282
00:17:45,840 --> 00:17:48,840
and competition with China. 
Once you call something 

283
00:17:48,840 --> 00:17:52,760
existential, the limit on 
spending becomes unlimited. 

284
00:17:53,240 --> 00:17:57,080
Polkidoski described it as a 
meta bubble on the Odd Lots 

285
00:17:57,080 --> 00:17:59,680
podcast. 
Tech hype, Real estate, state 

286
00:17:59,680 --> 00:18:04,040
speculation, loose credit and a 
potential government backstop 

287
00:18:04,320 --> 00:18:07,720
all in one. 
There are some bubbly signs. 

288
00:18:07,960 --> 00:18:12,600
I remember in 1999 seeing 
adverts on CNBC for a company 

289
00:18:12,600 --> 00:18:16,800
that manufactured equipment used
in the way for fabrication steps

290
00:18:16,800 --> 00:18:19,880
of making semiconductors. 
I couldn't understand at the 

291
00:18:19,880 --> 00:18:24,040
time why they were paying for TV
adverts when their customers 

292
00:18:24,040 --> 00:18:26,800
would all know who they were and
what they sell. 

293
00:18:27,000 --> 00:18:31,160
No one watches CNBC and decides 
to start manufacturing computer 

294
00:18:31,160 --> 00:18:34,600
chips in their garage. 
I later worked out that they 

295
00:18:34,600 --> 00:18:37,600
were advertising the stock, not 
the products. 

296
00:18:37,880 --> 00:18:41,600
The stock fell around 80% over 
the next three years. 

297
00:18:42,000 --> 00:18:46,400
Recently, I've seen a tech CEO 
being interviewed wearing AT 

298
00:18:46,400 --> 00:18:48,840
shirt with his company's ticker 
symbol on it. 

299
00:18:49,080 --> 00:18:51,840
Not the company name, the ticker
symbol. 

300
00:18:52,120 --> 00:18:55,800
I've noticed that every podcast 
I listened to over the last few 

301
00:18:55,800 --> 00:18:59,480
weeks seems to have adverts for 
an AI military tech company. 

302
00:18:59,680 --> 00:19:02,680
And once again, I wonder if they
think that their potential 

303
00:19:02,680 --> 00:19:06,760
customers might be listening to 
a Bloomberg podcast or if they 

304
00:19:06,760 --> 00:19:10,480
just want to pump the stock. 
I'll note that the CEO of that 

305
00:19:10,480 --> 00:19:13,480
company constantly. 
He talks about burning short 

306
00:19:13,480 --> 00:19:15,920
sellers while dumping his own 
stock. 

307
00:19:16,480 --> 00:19:20,280
Even if there is a bubble, it 
can be impossible to know when 

308
00:19:20,280 --> 00:19:23,080
it'll pop. 
As I mentioned a few weeks ago, 

309
00:19:23,320 --> 00:19:27,320
the big tech firms funding a lot
of the AI spending are so 

310
00:19:27,320 --> 00:19:30,600
profitable in their core 
businesses that they can afford 

311
00:19:30,600 --> 00:19:33,880
this gamble. 
So should investors cash out of 

312
00:19:33,880 --> 00:19:37,440
the stock market then? 
Well, probably not, unless they 

313
00:19:37,440 --> 00:19:39,880
know how they'll get back in 
again. 

314
00:19:40,120 --> 00:19:44,120
If you're a diversified investor
with a long holding period, even

315
00:19:44,120 --> 00:19:48,760
if you invested the day before 
the 1987 crash, right before the

316
00:19:48,760 --> 00:19:52,400
credit crunch, or right before 
the COVID sell off, you might 

317
00:19:52,400 --> 00:19:54,400
have been uncomfortable for a 
while. 

318
00:19:54,480 --> 00:19:57,880
But if you stayed invested, you 
weren't good returns over time. 

319
00:19:58,400 --> 00:20:01,880
The Economist estimates that 
should an AI crash occur. 

320
00:20:01,960 --> 00:20:06,400
Her it could erase 8% of US 
household wealth and cut 

321
00:20:06,400 --> 00:20:11,760
consumption by $500 billion, or 
1.6% of GDP. 

322
00:20:12,120 --> 00:20:16,160
They show that at the 
peakofthe.com bubble, the market

323
00:20:16,160 --> 00:20:20,400
cap of the S&P was 124% of 
USGDP. 

324
00:20:20,880 --> 00:20:25,640
When the bubble burst, tech 
stocks lost on average 76% of 

325
00:20:25,640 --> 00:20:30,240
their value. 
Since CHAT GP TS launch in 2022,

326
00:20:30,440 --> 00:20:37,640
American stocks are up 71% and 
the S&P is worth 175% of GDP. 

327
00:20:38,560 --> 00:20:42,080
They point out that a crash 
today would have a bigger effect

328
00:20:42,080 --> 00:20:46,320
on ordinary Americans than it 
did 25 years ago, as the share 

329
00:20:46,320 --> 00:20:50,920
of household wealth in the stock
market has climbed from 17% back

330
00:20:50,920 --> 00:20:55,440
then to 21% today. 
If the stock market fell as much

331
00:20:55,440 --> 00:20:59,720
as it did back then, it would 
wipe out as much as 8% of US 

332
00:20:59,720 --> 00:21:02,880
household wealth. 
Foreign investors who are 

333
00:21:02,880 --> 00:21:06,400
heavily invested in US tech 
would take a significant hit, 

334
00:21:06,400 --> 00:21:09,720
too. 
The fallout wouldn't stop at 

335
00:21:09,720 --> 00:21:12,560
Silicon Valley. 
Pension funds, REITs, and 

336
00:21:12,560 --> 00:21:16,000
private credit vehicles are 
exposed to AI investment, too. 

337
00:21:16,320 --> 00:21:19,960
Utilities that built gas plants 
for data centers could be left 

338
00:21:19,960 --> 00:21:23,680
with standard assets. 
The last time America overbuilt 

339
00:21:23,680 --> 00:21:28,040
infrastructure this aggressively
was the telecom boom, and much 

340
00:21:28,040 --> 00:21:31,880
of the dark fibre that was laid 
back then was never lit. 

341
00:21:32,320 --> 00:21:36,080
As I said a few weeks ago, the 
tech boom today is very 

342
00:21:36,080 --> 00:21:40,120
different to the.com bubble of 
the late 90s where unprofitable 

343
00:21:40,120 --> 00:21:44,360
start-ups were racing to IPO 
after a few months in business, 

344
00:21:44,640 --> 00:21:48,840
burning cash on vague promises 
of eyeballs and banner ads. 

345
00:21:49,360 --> 00:21:53,480
Today's big tech firms, 
Microsoft, Amazon, Google, Meta 

346
00:21:53,720 --> 00:21:57,560
are highly profitable, well run 
businesses with entrenched 

347
00:21:57,560 --> 00:22:00,320
revenue streams. 
They may be pouring 10s of 

348
00:22:00,320 --> 00:22:04,440
billions into AI, but if these 
bets fail, their core 

349
00:22:04,440 --> 00:22:09,440
businesses, cloud, advertising 
and e-commerce remain intact and

350
00:22:09,440 --> 00:22:12,600
cash flow positive. 
The real risk sits with the 

351
00:22:12,600 --> 00:22:16,920
private AI labs and their 
venture backers, not really with

352
00:22:16,920 --> 00:22:20,720
the hyperscalers. 
If anything resembles the fraud 

353
00:22:20,720 --> 00:22:25,240
of 1999, it's crypto, not 
trillion dollar companies with 

354
00:22:25,240 --> 00:22:30,440
fortress balance sheets. 
For AI users, this frenzy is a 

355
00:22:30,440 --> 00:22:32,640
gift. 
Competition has meant that the 

356
00:22:32,640 --> 00:22:36,320
models improve rapidly and their
prices start low. 

357
00:22:36,600 --> 00:22:39,440
There's no reason not to use 
these products while they're 

358
00:22:39,440 --> 00:22:43,160
free or almost free for AI 
investors. 

359
00:22:43,280 --> 00:22:47,680
The economics are unforgiving. 
Better chips make models faster 

360
00:22:47,680 --> 00:22:50,000
and make yesterday space chips 
worthless. 

361
00:22:50,240 --> 00:22:54,320
Every leap forward accelerates 
depreciation on the collateral 

362
00:22:54,320 --> 00:22:58,760
lenders are asked to finance. 
That's why banks refuse to lend.

363
00:22:59,000 --> 00:23:02,480
They prefer assets that last 
longer than a news cycle. 

364
00:23:02,760 --> 00:23:07,600
Sam Altman says that Open AI 
isn't and wasn't pitching for a 

365
00:23:07,600 --> 00:23:11,400
government backstop, that he 
thinks government should build 

366
00:23:11,400 --> 00:23:16,120
their own AI infrastructure. 
That might happen, but it does 

367
00:23:16,120 --> 00:23:21,160
nothing to solve Open AI's 
problem financing $1.4 trillion 

368
00:23:21,160 --> 00:23:25,280
of private data centres with non
government guaranteed bonds. 

369
00:23:25,600 --> 00:23:29,200
For now, the company is betting 
that the capital markets will 

370
00:23:29,200 --> 00:23:32,520
keep playing along. 
If they don't, the build out 

371
00:23:32,520 --> 00:23:37,280
debate FIRE stumbled into will 
return louder, sharper, and 

372
00:23:37,280 --> 00:23:41,080
harder to ignore. 
I almost forgot to include this 

373
00:23:41,080 --> 00:23:44,120
piece, but one of the funnier 
news stories of the week was 

374
00:23:44,120 --> 00:23:48,240
about Grok, Elon Musk's maximum 
truth seeking chatbot. 

375
00:23:48,560 --> 00:23:51,920
It seems that the code must have
been tweaked a bit this week and

376
00:23:51,920 --> 00:23:56,120
adjusted such that Grok's output
is more in line with Musk's way 

377
00:23:56,120 --> 00:23:59,040
of thinking. 
Grok began claiming that Elon 

378
00:23:59,040 --> 00:24:02,760
Musk is more physically fit than
LeBron James, a better role 

379
00:24:02,760 --> 00:24:06,560
model than Jesus, and that his 
intellect is in the same bracket

380
00:24:06,560 --> 00:24:10,400
as Isaac Newton's, that he was a
better fighter than Mike Tyson, 

381
00:24:10,560 --> 00:24:13,000
and that he's funnier than Jerry
Seinfeld. 

382
00:24:13,320 --> 00:24:16,720
People quickly worked out that 
Grok would say that Musk was 

383
00:24:16,720 --> 00:24:20,720
amazing at everything, which led
to some inappropriate questions 

384
00:24:20,720 --> 00:24:24,040
and this headline at Four O 4 
Media. 

385
00:24:24,400 --> 00:24:28,720
Many of the Grok responses were 
quietly deleted on Friday, and 

386
00:24:28,720 --> 00:24:33,160
Musk tweeted that someone had 
manipulated Grok into saying 

387
00:24:33,160 --> 00:24:35,440
absurdly positive things about 
him. 

388
00:24:35,800 --> 00:24:39,400
I'm sure if he ever catches that
guy, he'll be in a world of of 

389
00:24:39,400 --> 00:24:42,000
trouble. 
Thanks again for tuning into the

390
00:24:42,000 --> 00:24:45,600
podcast, which is entirely 
supported by viewers like you on

391
00:24:45,600 --> 00:24:47,840
Patreon. 
If you'd like to support the 

392
00:24:47,840 --> 00:24:50,760
podcast, I'll leave a link in 
the description. 

393
00:24:50,920 --> 00:24:53,360
Have a great day and talk to you
again soon. 

394
00:24:53,640 --> 00:24:54,000
Bye.