1
00:00:03,160 --> 00:00:17,960
Speaker 1: Media greetums. I'm at xytron and this is better offline.

2
00:00:20,160 --> 00:00:22,280
Speaker 1: In my last episode, I started by telling you about

3
00:00:22,280 --> 00:00:24,319
Speaker 1: a report from the analyst wing of a major bank,

4
00:00:24,600 --> 00:00:27,880
Speaker 1: TD Cohen that revealed how Microsoft had drastically pulled back

5
00:00:27,880 --> 00:00:30,480
Speaker 1: on its plans to build new data center capacity. At

6
00:00:30,480 --> 00:00:33,280
Speaker 1: a bare minimum, Microsoft had canceled the equivalent of every

7
00:00:33,360 --> 00:00:36,800
Speaker 1: data center in London or Tokyo at least, that the

8
00:00:36,840 --> 00:00:39,559
Speaker 1: real figure is actually probably much much higher, and as

9
00:00:39,600 --> 00:00:42,120
Speaker 1: I pointed out, that this is a real pale horse,

10
00:00:42,159 --> 00:00:45,080
Speaker 1: a harbinger of bad times through generative AI and open

11
00:00:45,080 --> 00:00:48,200
Speaker 1: AI especially, and an indicator that the bubble is imploding

12
00:00:48,320 --> 00:00:51,360
Speaker 1: or popping or I really don't want to say it's popped,

13
00:00:51,400 --> 00:00:54,360
Speaker 1: and I actually don't know if it's necessary to say

14
00:00:54,360 --> 00:00:56,600
Speaker 1: when it does, it's never going to be one thing.

15
00:00:57,480 --> 00:00:59,720
Speaker 1: I also mentioned that some may interpret this movie as

16
00:00:59,720 --> 00:01:02,600
Speaker 1: a sponds to open AI's Stargate project, which aims to

17
00:01:02,640 --> 00:01:04,920
Speaker 1: build hundreds of billions of dollars of data centers and

18
00:01:04,920 --> 00:01:08,119
Speaker 1: power generation facilities in the US, all to power generator

19
00:01:08,160 --> 00:01:11,440
Speaker 1: FAI apps and tools that I'm not sure actually there's

20
00:01:11,480 --> 00:01:14,840
Speaker 1: really the demand for. But this begs the question how

21
00:01:14,840 --> 00:01:17,480
Speaker 1: feasible is the Stargate project. Let's start with a bit

22
00:01:17,520 --> 00:01:20,600
Speaker 1: of background. The Stargate project was officially announced that the

23
00:01:20,640 --> 00:01:23,000
Speaker 1: very beginning of the Trump presidency. Is open AI tried

24
00:01:23,000 --> 00:01:25,800
Speaker 1: to court favor from the notoriously transactional and praise hungry

25
00:01:25,920 --> 00:01:28,959
Speaker 1: person of the United States of America, despite that the

26
00:01:28,959 --> 00:01:30,600
Speaker 1: project has been in the works for some time and

27
00:01:30,640 --> 00:01:34,679
Speaker 1: open AI had previously courted the Biden administration. Sam Allman's

28
00:01:34,720 --> 00:01:37,240
Speaker 1: previous pitch to that administration late last year was that

29
00:01:37,280 --> 00:01:39,720
Speaker 1: it was necessary to build a five gigaw what data center.

30
00:01:40,280 --> 00:01:42,440
Speaker 1: We don't know how big Stargate will be, just that

31
00:01:42,480 --> 00:01:44,959
Speaker 1: it will initially involve spending one hundred billion dollars due

32
00:01:45,040 --> 00:01:48,200
Speaker 1: and I quote developed data centers for artificial intelligence in

33
00:01:48,200 --> 00:01:51,040
Speaker 1: the US according to the information, with the project potentially

34
00:01:51,080 --> 00:01:54,480
Speaker 1: scaling to five hundred billion dollars, a truly fucking, astoundingly

35
00:01:54,520 --> 00:01:58,520
Speaker 1: stupid number. Stargate's first and only data center deal currently

36
00:01:58,600 --> 00:02:01,120
Speaker 1: signed is in Abilene, Texas, and it's expected to be

37
00:02:01,160 --> 00:02:04,520
Speaker 1: operational in mid twenty twenty six. Though these centers usually

38
00:02:04,520 --> 00:02:08,280
Speaker 1: become operational in phases, this is especially likely to be

39
00:02:08,320 --> 00:02:11,320
Speaker 1: the case here considering that, according to the information. Open

40
00:02:11,360 --> 00:02:14,360
Speaker 1: Ai plans to have access to a gigawat of power

41
00:02:14,360 --> 00:02:17,200
Speaker 1: on hundreds of thousands of GPUs. As part of this,

42
00:02:17,280 --> 00:02:19,640
Speaker 1: the Stargate project will construct a three hundred and sixty

43
00:02:19,680 --> 00:02:22,440
Speaker 1: point five megawatt natural gas power station. And as I

44
00:02:22,480 --> 00:02:25,720
Speaker 1: said last episode, that one's about power. This power station is,

45
00:02:25,760 --> 00:02:28,000
Speaker 1: as far as I can tell, still in the permitting phase.

46
00:02:28,240 --> 00:02:31,200
Speaker 1: It'll be some time before Stargate breaks ground on the facility,

47
00:02:31,560 --> 00:02:35,720
Speaker 1: let alone starts to actually generate power. Now, as for funding,

48
00:02:36,160 --> 00:02:38,520
Speaker 1: things have got a little weird. Both open ai and

49
00:02:38,560 --> 00:02:41,000
Speaker 1: soft Bank have committed to putting in either eighteen billion

50
00:02:41,120 --> 00:02:43,919
Speaker 1: or nineteen million dollars each into the project. I've seen

51
00:02:43,960 --> 00:02:47,040
Speaker 1: both numbers reported, by the way. Regardless, it's not that

52
00:02:47,120 --> 00:02:51,520
Speaker 1: big a difference to worry about, especially with well the

53
00:02:51,560 --> 00:02:53,560
Speaker 1: fact that open ai does not really have the money

54
00:02:53,600 --> 00:02:56,360
Speaker 1: either way. What's a billion dollars when you don't have anything.

55
00:02:56,720 --> 00:02:59,160
Speaker 1: The company is currently trying to raise forty billion dollars

56
00:02:59,160 --> 00:03:01,639
Speaker 1: at a two hundred sixty billion dollar valuation, with the

57
00:03:01,720 --> 00:03:04,760
Speaker 1: quote CNBC part of the funding expected to be used

58
00:03:04,760 --> 00:03:08,480
Speaker 1: for open AI's commitment to Stargate. Now, I'm old enough

59
00:03:08,480 --> 00:03:10,919
Speaker 1: to remember when this round was previously rumored to be

60
00:03:11,040 --> 00:03:13,400
Speaker 1: valuing open Ai at three hundred and forty billion dollars

61
00:03:13,480 --> 00:03:16,280
Speaker 1: and also at three hundred billion dollars, and soft Bank

62
00:03:16,280 --> 00:03:18,600
Speaker 1: appears to be taking full responsibility for raising the round,

63
00:03:18,639 --> 00:03:21,880
Speaker 1: including syndicating as much as ten billion dollars of the amount,

64
00:03:22,120 --> 00:03:25,440
Speaker 1: which means that it would include a group of other investors. Nevertheless,

65
00:03:25,480 --> 00:03:28,040
Speaker 1: it certainly seems SoftBank will provide the majority of the

66
00:03:28,040 --> 00:03:31,200
Speaker 1: capital thirty billion dollars, with CNBC reporting that it will

67
00:03:31,240 --> 00:03:33,480
Speaker 1: be paid out over the next twelve to twenty four months,

68
00:03:33,560 --> 00:03:36,440
Speaker 1: with the first payment coming as soon as spring. Soft

69
00:03:36,480 --> 00:03:39,520
Speaker 1: Bank also has another problem, and this one's this one.

70
00:03:39,560 --> 00:03:41,880
Speaker 1: I think even the least technical of you can understand.

71
00:03:41,920 --> 00:03:45,280
Speaker 1: They also appear to not have the money. They don't

72
00:03:45,280 --> 00:03:47,560
Speaker 1: have the money. They not the money to give open Ai,

73
00:03:47,760 --> 00:03:50,160
Speaker 1: not the money for Stargate. It's a little bit of

74
00:03:50,200 --> 00:03:54,000
Speaker 1: worrying math issues with this. According to the Information, SoftBank

75
00:03:54,080 --> 00:03:56,960
Speaker 1: CEO Masayoshi's Sun is planning to borrow sixteen billion dollars

76
00:03:57,000 --> 00:03:59,400
Speaker 1: to invest in Ai, and may borrow another eight billion

77
00:03:59,400 --> 00:04:02,120
Speaker 1: dollars next year. The following points, by the way, are

78
00:04:02,160 --> 00:04:04,560
Speaker 1: drawn from the Information's reporting, and I give serious props

79
00:04:04,560 --> 00:04:07,400
Speaker 1: to Duro Osawa and Corey Weinberg for their hard work here.

80
00:04:07,960 --> 00:04:10,160
Speaker 1: I attack the information sometimes with some of the framing,

81
00:04:10,200 --> 00:04:12,160
Speaker 1: but they actually do some of the best reporting in

82
00:04:12,200 --> 00:04:15,880
Speaker 1: tech journalism. Now let's do some maths. Soft Bank currently

83
00:04:15,880 --> 00:04:17,760
Speaker 1: only has thirty one billion dollars in cash and its

84
00:04:17,800 --> 00:04:20,480
Speaker 1: balance sheet as of December twenty twenty four. It's net debt,

85
00:04:20,520 --> 00:04:22,680
Speaker 1: which despite what you think, does not measure total debt,

86
00:04:22,680 --> 00:04:25,919
Speaker 1: but rather represents its cash minus any debt liabilities, sounds

87
00:04:25,920 --> 00:04:27,960
Speaker 1: at twenty nine billion dollars. They plan to use the

88
00:04:28,000 --> 00:04:30,760
Speaker 1: loaning question to finance part of their investment in open Ai,

89
00:04:31,040 --> 00:04:34,480
Speaker 1: and there are acquisition of chip design firm Ampere. According

90
00:04:34,480 --> 00:04:36,800
Speaker 1: to SoftBank's reported assets, its holdings are worth about two

91
00:04:36,880 --> 00:04:39,559
Speaker 1: hundred and nineteen billion dollars, so thirty three point sixty

92
00:04:39,839 --> 00:04:42,719
Speaker 1: six trillion yen for those of you who deal with yen,

93
00:04:43,040 --> 00:04:46,680
Speaker 1: including stock in companies like Alibaba and ARM. On a

94
00:04:46,760 --> 00:04:49,480
Speaker 1: side note, not every company in soft Bank's portfolio is

95
00:04:49,520 --> 00:04:52,159
Speaker 1: an arm or an Alibaba or like a good one.

96
00:04:52,320 --> 00:04:54,599
Speaker 1: There are some real stinkers too, and I'm not just

97
00:04:54,600 --> 00:04:57,400
Speaker 1: talking about the Morrobon carcass of we work. I recommend

98
00:04:57,480 --> 00:04:59,240
Speaker 1: you look at and I'll link to the list in

99
00:04:59,240 --> 00:05:02,320
Speaker 1: the episode of notes SoftBank's recent group report, which is

100
00:05:02,360 --> 00:05:04,719
Speaker 1: linked in the spreadsheet. Like I just said, which is

101
00:05:04,760 --> 00:05:06,880
Speaker 1: me reading a script out a little behind the curtain

102
00:05:06,880 --> 00:05:09,760
Speaker 1: for you there. In particular, got to twenty nine in

103
00:05:09,839 --> 00:05:12,679
Speaker 1: the report, which lists the ten largest publicly traded companies

104
00:05:12,680 --> 00:05:16,120
Speaker 1: in soft Bank's Vision Fund portfolio. Now how all of them,

105
00:05:16,120 --> 00:05:19,920
Speaker 1: without exception trade at a significant fraction of their peak market cap.

106
00:05:20,400 --> 00:05:22,880
Speaker 1: In simpler terms, the ten biggest companies in soft Bank's

107
00:05:22,880 --> 00:05:25,440
Speaker 1: flagship tech fund are worth far far less than their

108
00:05:25,480 --> 00:05:27,680
Speaker 1: all time high, and in some cases with less than

109
00:05:27,720 --> 00:05:30,719
Speaker 1: one fifth of their all time high. If soft Bank

110
00:05:30,720 --> 00:05:32,960
Speaker 1: liquidated its assets, and I admit this is a big

111
00:05:33,000 --> 00:05:35,760
Speaker 1: ear for most likely a worst case scenario situation, how

112
00:05:35,760 --> 00:05:39,720
Speaker 1: big would their losses be? Separately, SoftBank has committed to

113
00:05:39,800 --> 00:05:42,640
Speaker 1: a joint venture called SBO open Ai Japan and to

114
00:05:42,680 --> 00:05:44,680
Speaker 1: spend three billion dollars a year on open Ai. Is

115
00:05:44,720 --> 00:05:47,560
Speaker 1: take for the various companies across its group. We'll talk

116
00:05:47,560 --> 00:05:50,240
Speaker 1: about that later, but there's some napkin mass. It is

117
00:05:50,279 --> 00:05:53,200
Speaker 1: what soft Banks agreed to sixteen billion dollars sorry, eighteen

118
00:05:53,240 --> 00:05:55,880
Speaker 1: billion dollars or nineteen billion dollars really do not know

119
00:05:55,960 --> 00:05:58,839
Speaker 1: in funding for the Stargate Data Center project, three billion

120
00:05:58,880 --> 00:06:01,880
Speaker 1: dollars a year in spend on open ai software, and

121
00:06:01,920 --> 00:06:04,280
Speaker 1: as much as thirty billion dollars in funding for open

122
00:06:04,279 --> 00:06:06,600
Speaker 1: Ai paid over twelve to twenty four months, according to

123
00:06:06,640 --> 00:06:09,760
Speaker 1: the Information, and by the way, twenty five billion has

124
00:06:09,800 --> 00:06:12,120
Speaker 1: also been reported. But the Information reports that open Ai

125
00:06:12,160 --> 00:06:14,159
Speaker 1: has told investors that soft Bank will provide it at

126
00:06:14,279 --> 00:06:16,839
Speaker 1: least thirty billion dollars of the forty billion dollars they need.

127
00:06:17,040 --> 00:06:19,080
Speaker 1: Jesus fucking Christ. Can you imagine if this went It's

128
00:06:19,080 --> 00:06:23,560
Speaker 1: something useful anyway, What I'm getting at is that soft

129
00:06:23,560 --> 00:06:26,400
Speaker 1: Bank has effectively agreed to bankroll the entirety of open

130
00:06:26,400 --> 00:06:29,520
Speaker 1: AI's future, signing up for over forty six billion dollars

131
00:06:29,520 --> 00:06:32,080
Speaker 1: of investments over the next few years, and does not

132
00:06:32,120 --> 00:06:34,080
Speaker 1: appear to be able to do so without selling its

133
00:06:34,120 --> 00:06:36,680
Speaker 1: current holdings in valuable companies like arm or taking at

134
00:06:36,760 --> 00:06:39,680
Speaker 1: least sixteen billion dollars of debt this year, representing a

135
00:06:39,680 --> 00:06:43,039
Speaker 1: fifty five percent increase of its current liabilities. Work still,

136
00:06:43,080 --> 00:06:46,160
Speaker 1: thanks to disagreement, open AI's future both in its ability

137
00:06:46,200 --> 00:06:49,760
Speaker 1: to expand its infrastructure, which appears to be entirely contingent

138
00:06:49,839 --> 00:06:53,040
Speaker 1: on the construction of Stargate with Microsoft pulling out, and

139
00:06:53,080 --> 00:06:56,039
Speaker 1: its ability to raise funding, which is also entirely dependent

140
00:06:56,120 --> 00:06:59,000
Speaker 1: on soft bank and soft Bank in this case, where

141
00:06:59,040 --> 00:07:01,640
Speaker 1: open ai is entirely dependent on soft Bank to live,

142
00:07:01,839 --> 00:07:06,000
Speaker 1: must borrow money to give to open Ai, a company

143
00:07:06,040 --> 00:07:11,640
Speaker 1: that only only loses money. On top of that, open

144
00:07:11,680 --> 00:07:14,800
Speaker 1: Ai also on the money losing front, anticipates it will

145
00:07:14,840 --> 00:07:16,840
Speaker 1: burn as much as forty billion dollars a year by

146
00:07:16,880 --> 00:07:19,200
Speaker 1: twenty twenty eight, and projects to only turn a profit

147
00:07:19,240 --> 00:07:22,440
Speaker 1: by the end of the decade after the build out

148
00:07:22,480 --> 00:07:26,920
Speaker 1: of Stargate, which I add is, like I said, almost

149
00:07:27,040 --> 00:07:29,600
Speaker 1: entirely dependent on soft bank, which has to take on

150
00:07:29,720 --> 00:07:33,880
Speaker 1: debt to fund both open Ai and the project required

151
00:07:33,920 --> 00:07:37,920
Speaker 1: to theoretically make open ai profitable. How the fuck does

152
00:07:37,920 --> 00:07:40,600
Speaker 1: this work? How does this work? How does this work?

153
00:07:40,640 --> 00:07:42,560
Speaker 1: How does this work? How the fuck does this work?

154
00:07:42,600 --> 00:07:45,680
Speaker 1: Open Ai, a company that spent nine billion dollars to

155
00:07:45,760 --> 00:07:48,640
Speaker 1: lose five billion dollars in twenty twenty four, requires so

156
00:07:48,720 --> 00:07:52,520
Speaker 1: much money to be its obligations, both to cover its stupid, ruinous, unprofitable,

157
00:07:52,600 --> 00:07:55,560
Speaker 1: unsustainable operations and the eighteen billion to nineteen billion dollars.

158
00:07:55,560 --> 00:07:58,440
Speaker 1: A committed to keep growing that it has to raise

159
00:07:58,480 --> 00:08:01,679
Speaker 1: more money than any startup ever raised in history, forty

160
00:08:01,760 --> 00:08:04,480
Speaker 1: billion fucking dollars, with the cast iron guarantee that it

161
00:08:04,480 --> 00:08:08,320
Speaker 1: will need even more money within a year. Every goddamn month,

162
00:08:08,480 --> 00:08:10,679
Speaker 1: Sam Wilton has to go to someone and say, my huge,

163
00:08:10,680 --> 00:08:13,320
Speaker 1: beautiful company is so powerful, but it's weak and sick

164
00:08:13,360 --> 00:08:15,360
Speaker 1: and frail. I need more money than you have today,

165
00:08:15,480 --> 00:08:17,480
Speaker 1: and I'll need more money than you'll have tomorrow. My

166
00:08:17,560 --> 00:08:19,400
Speaker 1: company is so sick and weak, but it's the most

167
00:08:19,440 --> 00:08:24,160
Speaker 1: powerful company of all time. Sam flipping off pisces me off.

168
00:08:24,880 --> 00:08:27,520
Speaker 1: Imagine if this money went literally anywhere else, you could

169
00:08:27,560 --> 00:08:30,560
Speaker 1: set it on fire at least be fucking warm. Soft Bank.

170
00:08:31,000 --> 00:08:33,080
Speaker 1: On top of the thirty billion dollars of funding and

171
00:08:33,120 --> 00:08:35,400
Speaker 1: three billion dollars a year of revenue it's committed to,

172
00:08:35,440 --> 00:08:38,240
Speaker 1: open ai itself also has to cough up eighteen billion

173
00:08:38,240 --> 00:08:41,000
Speaker 1: dollars for Stargate, the data center project that open ai

174
00:08:41,120 --> 00:08:45,080
Speaker 1: will run and get this. Soft Bank will take financial

175
00:08:45,120 --> 00:08:49,240
Speaker 1: responsibility for forty eight billion dollars in cash three billion

176
00:08:49,280 --> 00:08:51,560
Speaker 1: dollars in revenue. You the latter of which, like all

177
00:08:51,559 --> 00:08:55,200
Speaker 1: open AI's offerings, will lose the company money. Open Ai

178
00:08:55,360 --> 00:08:58,559
Speaker 1: has no path to profitability guaranteeing it will need more

179
00:08:58,600 --> 00:09:00,400
Speaker 1: cash and right now at the time time, and it

180
00:09:00,440 --> 00:09:02,920
Speaker 1: needs it more than it's ever needed it. Soft Bank,

181
00:09:02,960 --> 00:09:05,080
Speaker 1: the only company willing to provide it and possibly the

182
00:09:05,080 --> 00:09:07,120
Speaker 1: only company with the money to do so, has proven

183
00:09:07,160 --> 00:09:09,760
Speaker 1: that it will have to go to greet possibly ruinous

184
00:09:09,840 --> 00:09:12,840
Speaker 1: lengths to do so. If open ai needs forty billion

185
00:09:12,880 --> 00:09:14,800
Speaker 1: dollars in twenty twenty five, how much will it need

186
00:09:14,840 --> 00:09:17,880
Speaker 1: in twenty twenty six fifty billion, one hundred billion dollars.

187
00:09:18,000 --> 00:09:21,040
Speaker 1: Where is that money going to come from? While soft

188
00:09:21,040 --> 00:09:23,120
Speaker 1: Bank might be able to do this once, what happens

189
00:09:23,120 --> 00:09:25,480
Speaker 1: when open ai needs money in six to twelve months.

190
00:09:25,600 --> 00:09:27,960
Speaker 1: Soft Bank made about fifteen billion dollars a profit in

191
00:09:27,960 --> 00:09:31,240
Speaker 1: the last year on about forty six billion dollars of revenue.

192
00:09:31,440 --> 00:09:34,079
Speaker 1: Three billion dollars is an absolutely obscene amount to commit

193
00:09:34,080 --> 00:09:37,040
Speaker 1: to buying open ai software annually, especially when some of

194
00:09:37,080 --> 00:09:39,760
Speaker 1: it is allegedly for access to open AI's barely functional

195
00:09:39,920 --> 00:09:44,160
Speaker 1: operator and mediocre deep research products. As per my previous

196
00:09:44,280 --> 00:09:47,760
Speaker 1: podcasts and pieces, I do not see how open ai survives,

197
00:09:47,960 --> 00:09:51,240
Speaker 1: and soft Bank's involvement only gives me further concern. While

198
00:09:51,240 --> 00:09:54,280
Speaker 1: soft banker theoretically burn its current holdings to fund open

199
00:09:54,320 --> 00:09:56,960
Speaker 1: Ai in perpetuity, its ability to do so is cast

200
00:09:57,000 --> 00:09:59,000
Speaker 1: into doubt by them having to borrow money from other

201
00:09:59,080 --> 00:10:01,720
Speaker 1: banks to get into this funding ground and to get

202
00:10:01,760 --> 00:10:05,720
Speaker 1: Stargate done. Open Ai burned five billion dollars in twenty

203
00:10:05,760 --> 00:10:08,400
Speaker 1: twenty four, a number it will likely double in twenty

204
00:10:08,440 --> 00:10:11,360
Speaker 1: twenty five. And remember the information reported that open ai

205
00:10:11,559 --> 00:10:14,200
Speaker 1: was projected to spend thirteen billion dollars on compute alone

206
00:10:14,240 --> 00:10:17,360
Speaker 1: with Microsoft in twenty twenty five, and has no path

207
00:10:17,400 --> 00:10:20,880
Speaker 1: to profitability. SoftBank has already had to borrow to fund

208
00:10:20,960 --> 00:10:22,800
Speaker 1: this round, and the fact that it had to do

209
00:10:22,840 --> 00:10:26,200
Speaker 1: so suggests its inability to continue supporting open Ai without

210
00:10:26,200 --> 00:10:30,080
Speaker 1: accruing further debt. Open Ai. As a result of Microsoft's cuts,

211
00:10:30,080 --> 00:10:34,040
Speaker 1: the data center capacity now only has one path expansion

212
00:10:34,600 --> 00:10:37,840
Speaker 1: once it runs through whatever remaining buildout Microsoft has planned.

213
00:10:37,840 --> 00:10:40,680
Speaker 1: That is, and that's Stargate, a project funded by open

214
00:10:40,760 --> 00:10:44,440
Speaker 1: AI's contribution, which it's receiving from SoftBank and SoftBank which

215
00:10:44,480 --> 00:10:47,079
Speaker 1: is also having to take out loans to meet its share.

216
00:10:47,760 --> 00:10:51,240
Speaker 1: How does this work exactly? How does this continue? Do

217
00:10:51,360 --> 00:10:53,840
Speaker 1: you see anyone else stepping up to fund this. Who

218
00:10:53,920 --> 00:10:56,400
Speaker 1: else has got thirty to forty billion dollars to shit

219
00:10:56,440 --> 00:10:59,840
Speaker 1: out every year? While the answer is Saudi Arabia. SoftBank's

220
00:11:00,000 --> 00:11:02,920
Speaker 1: CEO Massioshi Son recently said that he had had and

221
00:11:03,000 --> 00:11:05,640
Speaker 1: I quote not given Saudi ruler Muhammed bin Salmon in

222
00:11:05,720 --> 00:11:10,199
Speaker 1: our return, adding that he still owed him. That's really

223
00:11:10,240 --> 00:11:13,720
Speaker 1: not the ideal thing you want to say after naming MBS.

224
00:11:14,200 --> 00:11:16,640
Speaker 1: Nothing about this suggests that Saudi money will follow soft

225
00:11:16,640 --> 00:11:19,280
Speaker 1: banks in anywhere near the volume necessary. As to the

226
00:11:19,360 --> 00:11:21,599
Speaker 1: m ratis. They're already involved through the MGX fund, and

227
00:11:21,640 --> 00:11:27,160
Speaker 1: it's unclear how much more they'd be willing to commit. Really, really,

228
00:11:27,240 --> 00:11:31,920
Speaker 1: are really, mate, buddy fellow? How does this? How does

229
00:11:31,920 --> 00:11:35,000
Speaker 1: this work? In my opinion, this open Ai soft Bank

230
00:11:35,040 --> 00:11:38,040
Speaker 1: deal is wildly unsustainable, dependent on soft Bank continuing to

231
00:11:38,120 --> 00:11:41,080
Speaker 1: raise both debt and funnel money directly into a company

232
00:11:41,080 --> 00:11:43,760
Speaker 1: that burns it burns it by the billions every year,

233
00:11:43,840 --> 00:11:45,520
Speaker 1: and it's said to only burn more thanks to the

234
00:11:45,600 --> 00:11:48,360
Speaker 1: arrival of its latest bullshit model. And if it had

235
00:11:48,400 --> 00:11:50,880
Speaker 1: a huge breakthrough that would change everything. Wouldn't Microsoft want

236
00:11:50,880 --> 00:11:52,760
Speaker 1: to make sure they were building the data center capacity

237
00:11:52,800 --> 00:12:08,080
Speaker 1: to support him? Hey, maybe they're not. Now. Perhaps this

238
00:12:08,360 --> 00:12:11,280
Speaker 1: crazy level of spending would be necessary if open ai

239
00:12:11,440 --> 00:12:13,640
Speaker 1: was still the leader in generator of AI and it

240
00:12:13,679 --> 00:12:16,960
Speaker 1: was still meaningfully improving its capabilities year after year or

241
00:12:17,280 --> 00:12:19,559
Speaker 1: two year after two years. And I think we all

242
00:12:19,559 --> 00:12:21,960
Speaker 1: know that that isn't the case. A few weeks ago,

243
00:12:22,000 --> 00:12:24,800
Speaker 1: open ai launched GPT four point five, it's latest model

244
00:12:24,840 --> 00:12:29,560
Speaker 1: that well, this isn't brilliant, sam Alman says, and I quote,

245
00:12:29,600 --> 00:12:31,600
Speaker 1: is the first model that feels like talking to a

246
00:12:31,640 --> 00:12:35,000
Speaker 1: thoughtful person, which, by the way, is really funny. It's

247
00:12:35,040 --> 00:12:37,280
Speaker 1: really funny to say it, to be like, yeah, I

248
00:12:37,320 --> 00:12:40,480
Speaker 1: have spent billions of dollars telling you that this bullshit

249
00:12:40,600 --> 00:12:43,200
Speaker 1: is like a person, but this one really is. The

250
00:12:43,320 --> 00:12:45,560
Speaker 1: other one's not so much. I did not think that

251
00:12:45,640 --> 00:12:48,880
Speaker 1: in the past. And it's not obvious what GPT four

252
00:12:48,880 --> 00:12:52,439
Speaker 1: point five does better, or even really what it does,

253
00:12:52,480 --> 00:12:55,040
Speaker 1: other than Altman saying it is a different kind of

254
00:12:55,040 --> 00:12:58,120
Speaker 1: intelligence and there's a magic to it that he has

255
00:12:58,160 --> 00:13:03,760
Speaker 1: not felt before. Wow, wow, really really inspirational stuff. Sammy

256
00:13:04,720 --> 00:13:07,319
Speaker 1: fucking idiot. This was, by the way, in Altman's words,

257
00:13:07,400 --> 00:13:10,160
Speaker 1: the good news. The bad news was that and I

258
00:13:10,280 --> 00:13:13,480
Speaker 1: quote GPT four point five is a giant expensive model,

259
00:13:13,600 --> 00:13:16,520
Speaker 1: adding the open Ai was out of GPUs, but proudly

260
00:13:16,559 --> 00:13:19,160
Speaker 1: declaring that it had added tens of thousands of GPUs

261
00:13:19,200 --> 00:13:21,680
Speaker 1: in the week following and would roll out to open aiyes,

262
00:13:21,720 --> 00:13:24,120
Speaker 1: twenty dollars a month plus here and that they would

263
00:13:24,120 --> 00:13:28,240
Speaker 1: be adding hundreds of thousands of GPUs soon. Excited. Well,

264
00:13:28,280 --> 00:13:30,560
Speaker 1: you shouldn't be on top of a vague product set

265
00:13:30,559 --> 00:13:33,840
Speaker 1: and interterminately high compute cost. GPT four point five costs

266
00:13:33,880 --> 00:13:37,880
Speaker 1: developers an incredible seventy five dollars put million input tokens

267
00:13:37,920 --> 00:13:40,400
Speaker 1: in those are the prompts and data pushed into the model,

268
00:13:40,640 --> 00:13:43,600
Speaker 1: and of absolutely astounding one hundred and fifty dollars per

269
00:13:43,600 --> 00:13:46,640
Speaker 1: million output tokens. That's the output it creates. And that's

270
00:13:46,720 --> 00:13:49,600
Speaker 1: roughly three thousand percent more for input tokens and fifteen

271
00:13:49,679 --> 00:13:52,319
Speaker 1: hundred percent more expensive for output tokens than GPT four

272
00:13:52,320 --> 00:13:55,600
Speaker 1: to zero For results that open Aiyes, andre J. Caapathy

273
00:13:55,720 --> 00:13:59,000
Speaker 1: described as a little bit better and awesome, but also

274
00:13:59,240 --> 00:14:02,040
Speaker 1: not exactly that are trivial to point to, and one

275
00:14:02,080 --> 00:14:05,320
Speaker 1: developer described GPT four point five to our technica as

276
00:14:05,360 --> 00:14:08,520
Speaker 1: a lemon when comparing its reported performance to its price.

277
00:14:10,600 --> 00:14:14,240
Speaker 1: Arstentnaca also reported the GPT four point five was terrible

278
00:14:14,240 --> 00:14:16,600
Speaker 1: for coding relatively speaking, and other tests showed that the

279
00:14:16,600 --> 00:14:19,760
Speaker 1: model's performance was either slightly better or slightly worse across

280
00:14:19,760 --> 00:14:22,440
Speaker 1: the board. With that according to Ours Technica, one success

281
00:14:22,480 --> 00:14:26,240
Speaker 1: metric being the open ai found open human evaluators preferred

282
00:14:26,240 --> 00:14:29,360
Speaker 1: GPT four point five's responses of a GPT four to

283
00:14:29,400 --> 00:14:34,640
Speaker 1: zero in about fifty seven percent of interactions. Wow, that's

284
00:14:35,160 --> 00:14:38,160
Speaker 1: very underwhelming. So, just to be crystal clear, the biggest

285
00:14:38,160 --> 00:14:40,960
Speaker 1: AI company's latest model appears to be even more ruinously

286
00:14:41,000 --> 00:14:43,400
Speaker 1: expensive than its last one, while providing modest at best

287
00:14:43,400 --> 00:14:46,840
Speaker 1: improvements and performing worse than several benchmarks done competing models

288
00:14:48,760 --> 00:14:51,840
Speaker 1: very good. Despite these piss poor results, Sam Wultman's reaction

289
00:14:51,960 --> 00:14:54,040
Speaker 1: was to bring in hundreds of thousands of GPUs as

290
00:14:54,040 --> 00:14:56,080
Speaker 1: a means of exposing as many as people as possible

291
00:14:56,080 --> 00:14:58,960
Speaker 1: to his mediocre, ultra expensive model. And the best that

292
00:14:59,000 --> 00:15:01,040
Speaker 1: Altman has to offer is that this is the first

293
00:15:01,120 --> 00:15:04,080
Speaker 1: time people have been emailing with such passion asking open

294
00:15:04,120 --> 00:15:06,800
Speaker 1: ai to promise to never stop offering a specific model.

295
00:15:07,160 --> 00:15:09,480
Speaker 1: I am just going to say this. That is a tweet,

296
00:15:09,520 --> 00:15:13,000
Speaker 1: and it never happened, or it happened like once. This

297
00:15:13,080 --> 00:15:16,480
Speaker 1: is some girlfriend in Canada share Sam Altman is washed

298
00:15:16,880 --> 00:15:19,800
Speaker 1: when all of this falls apart, Remember I said he

299
00:15:19,920 --> 00:15:22,880
Speaker 1: was washed? Now, Remember how I talked about open AI's

300
00:15:22,920 --> 00:15:26,000
Speaker 1: lack of meaningful improvement as a reminder, GPT four point

301
00:15:26,000 --> 00:15:28,360
Speaker 1: five was meant to be GPT five, but, according to

302
00:15:28,400 --> 00:15:30,960
Speaker 1: the Wall Street Journal, continually failed to make a model

303
00:15:31,000 --> 00:15:33,840
Speaker 1: that advanced enough to justify the enormous cost, with a

304
00:15:33,880 --> 00:15:37,680
Speaker 1: six month training run costing five hundred million dollars and

305
00:15:37,760 --> 00:15:41,360
Speaker 1: GPT four point five requiring multiple runs of different sizes.

306
00:15:41,640 --> 00:15:44,160
Speaker 1: So yeah, open ai spent hundreds of millions of dollars

307
00:15:44,200 --> 00:15:47,320
Speaker 1: to make this great stuff. And I haven't even mentioned

308
00:15:47,360 --> 00:15:49,800
Speaker 1: the company's supported agent products. And no, I'm not talking

309
00:15:49,840 --> 00:15:52,760
Speaker 1: about Operator, which is also dogshit. By the way, open

310
00:15:52,800 --> 00:15:55,640
Speaker 1: ai wants to create tiers of AI agents, with the

311
00:15:55,720 --> 00:15:59,480
Speaker 1: cheapest costing two thousand dollars a month and capable of

312
00:15:59,480 --> 00:16:02,480
Speaker 1: handling it administrative tasks, and the most expensive costing twenty

313
00:16:02,520 --> 00:16:05,840
Speaker 1: thousand dollars in having PhD level, Look, I wrote the

314
00:16:05,880 --> 00:16:08,760
Speaker 1: script right, and I'm gonna be honest. I can't even

315
00:16:08,840 --> 00:16:15,080
Speaker 1: read that sentence with a straight face. Operator cannot even

316
00:16:15,120 --> 00:16:18,320
Speaker 1: search trip Advisor properly. It can't even do a thing

317
00:16:18,320 --> 00:16:20,680
Speaker 1: that let me google that for you does. And these

318
00:16:20,760 --> 00:16:23,000
Speaker 1: chunder fucks want to charge two thousand dollars a month

319
00:16:23,040 --> 00:16:26,040
Speaker 1: for an agent that does what does some sort of

320
00:16:26,360 --> 00:16:26,880
Speaker 1: what's it dou?

321
00:16:27,000 --> 00:16:27,160
Speaker 2: Oh?

322
00:16:27,480 --> 00:16:31,040
Speaker 1: Twenty thousand dollars for something with PhD level capabilities. I

323
00:16:31,040 --> 00:16:33,320
Speaker 1: think all the people with PhDs listen to this have

324
00:16:33,480 --> 00:16:36,960
Speaker 1: just stood up and gone, I have an idea. And

325
00:16:37,000 --> 00:16:39,320
Speaker 1: this is insane on many levels, not simply because the

326
00:16:39,360 --> 00:16:41,920
Speaker 1: base product is undercut by actual human workers in many

327
00:16:41,960 --> 00:16:44,760
Speaker 1: parts of the world, and even PhD students are typically

328
00:16:44,800 --> 00:16:46,880
Speaker 1: I only paid twenty thousand to thirty thousand a year

329
00:16:47,000 --> 00:16:50,120
Speaker 1: on average, and even people with actual doctorates in industries

330
00:16:50,160 --> 00:16:52,800
Speaker 1: really earn twenty thousand dollars a month unless they're working

331
00:16:52,800 --> 00:16:56,040
Speaker 1: in Silicon Valley occupying a C suite job. But forget

332
00:16:56,080 --> 00:16:57,880
Speaker 1: all about that. What does it mean to have a

333
00:16:57,880 --> 00:17:01,440
Speaker 1: PhD level agent? Remember? Other lambs are guessing machines. They

334
00:17:01,480 --> 00:17:05,159
Speaker 1: don't know anything or even understand the concepts behind the

335
00:17:05,200 --> 00:17:08,520
Speaker 1: words they spit onto a page. No, seriously, Sammy, what

336
00:17:08,600 --> 00:17:15,239
Speaker 1: does it mean? I'm fucking waiting you damp Goblin you pissant. Ah,

337
00:17:15,320 --> 00:17:19,199
Speaker 1: maybe I shouldn't just site here instarting him wanker. Anyway, This,

338
00:17:19,280 --> 00:17:20,879
Speaker 1: by the way, is the company that is about to

339
00:17:20,960 --> 00:17:23,159
Speaker 1: raise forty billion dollars led by a Japanese bank that

340
00:17:23,160 --> 00:17:25,280
Speaker 1: has to go into debt to fund both their operations

341
00:17:25,280 --> 00:17:31,520
Speaker 1: and the infrastructure necessary for them to grow any further. Again,

342
00:17:32,480 --> 00:17:36,880
Speaker 1: as we started with, Microsoft is canceling plans to massively

343
00:17:36,920 --> 00:17:39,360
Speaker 1: expand its data center capacity right at the time when

344
00:17:39,359 --> 00:17:42,560
Speaker 1: open ai just released its most computationally demanding model. Ever,

345
00:17:43,280 --> 00:17:46,520
Speaker 1: how do you reconcile those two things without concluding either

346
00:17:46,520 --> 00:17:49,080
Speaker 1: that Microsoft expects GBT four point five to be a

347
00:17:49,080 --> 00:17:52,560
Speaker 1: flop although it's simply unwilling to continue bankrolling open AI's

348
00:17:52,600 --> 00:17:55,400
Speaker 1: continued growth, or perhaps it's having doubts about the future

349
00:17:55,440 --> 00:18:01,800
Speaker 1: of generative AI in general. Hmm. Maybe now I have

350
00:18:02,040 --> 00:18:04,800
Speaker 1: been and remain hesitant to call the bubble bursting, because

351
00:18:04,840 --> 00:18:07,879
Speaker 1: bubbles do not burst, really, they certainly don't burst in

352
00:18:07,960 --> 00:18:11,560
Speaker 1: neat little events. Nevertheless, my pale horses I've predicted in

353
00:18:11,560 --> 00:18:14,159
Speaker 1: the past were led by one specific call that reduction

354
00:18:14,240 --> 00:18:16,760
Speaker 1: and capital expenditures by a hyperscaler was a sign that

355
00:18:16,800 --> 00:18:20,600
Speaker 1: things were collapsing. Microsoft walking away from over a gigawatt

356
00:18:20,600 --> 00:18:23,080
Speaker 1: of data center plans equivalent to as much as fourteen

357
00:18:23,119 --> 00:18:26,240
Speaker 1: percent of its current data center capacity, is a direct

358
00:18:26,359 --> 00:18:28,680
Speaker 1: sign that it does not believe that growth is there

359
00:18:28,880 --> 00:18:31,199
Speaker 1: in generative AI, and thus they are not building the

360
00:18:31,240 --> 00:18:34,600
Speaker 1: infrastructure to support it, and indeed may have overbuilt something,

361
00:18:34,720 --> 00:18:38,400
Speaker 1: as I've mentioned that Microsoft CEO Sachinidella has directly foreshadowed

362
00:18:38,480 --> 00:18:42,439
Speaker 1: in his interview with Dwarkesh and otherwise extremely boring waste

363
00:18:42,480 --> 00:18:45,080
Speaker 1: of an hour of your life. The entirety of the

364
00:18:45,080 --> 00:18:47,600
Speaker 1: tech industry and the AI bubble has been built on

365
00:18:47,640 --> 00:18:50,359
Speaker 1: the assumption that generative AI was the next big growth

366
00:18:50,440 --> 00:18:53,720
Speaker 1: vehicle for tech, and if Microsoft, the largest purchaser of

367
00:18:53,800 --> 00:18:57,119
Speaker 1: Nvidia GPUs in the most aggressive builder of AI infrastructure,

368
00:18:57,160 --> 00:19:00,280
Speaker 1: is reducing capacity, it heavily suggests that the growth is

369
00:19:00,440 --> 00:19:04,000
Speaker 1: not there. Microsoft has, by the looks of things, effectively

370
00:19:04,040 --> 00:19:07,000
Speaker 1: given up on further data center expansion, at least at

371
00:19:07,000 --> 00:19:10,520
Speaker 1: the breakneck pay suit runs promised, and that even suggests

372
00:19:10,520 --> 00:19:12,199
Speaker 1: that generator if AI will be a thing in a

373
00:19:12,200 --> 00:19:14,880
Speaker 1: few years, definitely not the scale it is right now.

374
00:19:15,520 --> 00:19:18,760
Speaker 1: AI boosters will email me and they'll say there's something

375
00:19:18,800 --> 00:19:20,879
Speaker 1: that I don't know that in fact, Microsoft has some

376
00:19:20,920 --> 00:19:24,119
Speaker 1: greater strategy and some efficiency play. But answer me this,

377
00:19:24,280 --> 00:19:26,520
Speaker 1: Why is Microsoft canceling over a giga? What of data

378
00:19:26,560 --> 00:19:29,440
Speaker 1: set expansion? And again this is the most conservative estimate.

379
00:19:30,119 --> 00:19:32,040
Speaker 1: The realistic number is much much higher. Do you think

380
00:19:32,040 --> 00:19:34,479
Speaker 1: it's because it expects there to be this dramatic demand

381
00:19:34,480 --> 00:19:37,520
Speaker 1: for the AI services? Do you think it's reducing supply

382
00:19:37,640 --> 00:19:41,520
Speaker 1: because of all the demand? Now you might think that

383
00:19:41,600 --> 00:19:46,760
Speaker 1: this is an efficiency play they're playing with deep SIG right, No, sorry,

384
00:19:46,880 --> 00:19:50,439
Speaker 1: that doesn't matter. Even if deep CG was this magical

385
00:19:50,440 --> 00:19:52,399
Speaker 1: efficiency play, which it may or may not be. I

386
00:19:52,480 --> 00:19:54,480
Speaker 1: actually think it is more efficient, Like that's true, but

387
00:19:54,520 --> 00:19:57,800
Speaker 1: we actually don't know if it's profitable. Even then, they've

388
00:19:57,840 --> 00:20:00,720
Speaker 1: been talking about not having the capacity to deal with demand.

389
00:20:00,720 --> 00:20:03,520
Speaker 1: They've been talking about how incredible this is. They've been

390
00:20:03,520 --> 00:20:06,439
Speaker 1: talking about how big this is gonna be. This sounds

391
00:20:06,440 --> 00:20:09,360
Speaker 1: like they don't think it's gonna big, gonna be big.

392
00:20:09,400 --> 00:20:11,280
Speaker 1: Even it's gonna big, they're gonna get my ass in

393
00:20:11,320 --> 00:20:14,719
Speaker 1: the comments on that one. Now, one might argue that

394
00:20:14,760 --> 00:20:17,160
Speaker 1: Microsoft's reduction and capacity build out it's just the sign

395
00:20:17,160 --> 00:20:20,919
Speaker 1: that open AI is moving its compute elsewhere. Maybe that's true.

396
00:20:21,840 --> 00:20:24,520
Speaker 1: And if starguate ever gets built, which I question anyway,

397
00:20:25,040 --> 00:20:28,159
Speaker 1: here are some questions to ask. Microsoft still sells access

398
00:20:28,200 --> 00:20:30,920
Speaker 1: to open AI's API through asure. Does it not see

399
00:20:30,960 --> 00:20:33,879
Speaker 1: the growth in that product? Do they not see it?

400
00:20:34,040 --> 00:20:37,240
Speaker 1: They're not expanding? Is the growth not there? And Microsoft still,

401
00:20:37,280 --> 00:20:40,560
Speaker 1: one would assume, makes money off of open AI's compute expenses,

402
00:20:40,640 --> 00:20:42,320
Speaker 1: right or is that not the case due to the

403
00:20:42,400 --> 00:20:44,600
Speaker 1: vast seventy five percent discount the open ai gets on

404
00:20:44,720 --> 00:20:47,280
Speaker 1: using its services. I have been told that it's very

405
00:20:47,280 --> 00:20:50,679
Speaker 1: close to the wire by sources, but I can't say it.

406
00:20:51,240 --> 00:20:53,159
Speaker 1: But you'd have to just look at the fact that

407
00:20:53,240 --> 00:20:55,280
Speaker 1: they do actually do that was reported by the information.

408
00:20:56,400 --> 00:21:00,800
Speaker 1: Look look what that Microsoft making. So which a material

409
00:21:00,880 --> 00:21:03,840
Speaker 1: pullback on data center expansions suggests that the growth in

410
00:21:03,920 --> 00:21:07,359
Speaker 1: generative AI products, both those run on Microsoft servers and

411
00:21:07,400 --> 00:21:10,160
Speaker 1: those sold as part of Microsoft's products, do not have

412
00:21:10,200 --> 00:21:14,320
Speaker 1: the revolutionary growth trajectory that both CFO Amhood and CEO

413
00:21:14,320 --> 00:21:16,840
Speaker 1: of sach Inadella have been claiming. And this is all

414
00:21:16,880 --> 00:21:21,199
Speaker 1: deeply concerning, while also calling into consideration the viability of

415
00:21:21,200 --> 00:21:24,840
Speaker 1: generative AI as a growth vehicle for any hyperscaler. If

416
00:21:24,880 --> 00:21:27,400
Speaker 1: I am correct, Microsoft is walking away not just from

417
00:21:27,440 --> 00:21:30,720
Speaker 1: expansion of its current data center operations, but from generative AI,

418
00:21:30,800 --> 00:21:34,040
Speaker 1: writ large. I actually believe it will continue selling this unprofitable,

419
00:21:34,119 --> 00:21:36,920
Speaker 1: unsustainable software because the capacity it has right now is

420
00:21:36,960 --> 00:21:40,080
Speaker 1: more than sufficient to deal with the incredible lack of demand.

421
00:21:40,720 --> 00:21:43,320
Speaker 1: It's time for investors in the general public to begin

422
00:21:43,400 --> 00:21:47,320
Speaker 1: demanding tangible, direct numbers on the revenue and profits related

423
00:21:47,359 --> 00:21:50,359
Speaker 1: to GENERATIVEAI, as it is becoming increasingly obvious that the

424
00:21:50,400 --> 00:21:53,560
Speaker 1: revenues are small and the profits are non existent. A

425
00:21:53,680 --> 00:21:56,479
Speaker 1: gigawatt of capacity is huge, and walking away from that

426
00:21:56,560 --> 00:21:59,240
Speaker 1: much capacity is a direct signal that Microsoft's long term

427
00:21:59,280 --> 00:22:02,000
Speaker 1: plans do not include needing a great deal of compute.

428
00:22:02,640 --> 00:22:04,520
Speaker 1: One counter could be that it's waiting for more of

429
00:22:04,520 --> 00:22:07,440
Speaker 1: the specialized nvideo GPUs to arrive, to which the response

430
00:22:07,520 --> 00:22:09,800
Speaker 1: is Microsoft still wants to build the capacity, so it

431
00:22:09,800 --> 00:22:12,679
Speaker 1: has somewhere to fucking put them again, These facilities take

432
00:22:12,680 --> 00:22:15,159
Speaker 1: anywhere between three and six years to build. Do you

433
00:22:15,200 --> 00:22:17,080
Speaker 1: really think black will be so delayed that they won't

434
00:22:17,160 --> 00:22:19,960
Speaker 1: arrive until what twenty twenty eight? Ways, what are you

435
00:22:20,080 --> 00:22:24,000
Speaker 1: talking about? From even alive then? Anyway, One counter could

436
00:22:24,040 --> 00:22:25,879
Speaker 1: be that there isn't the power necessary to power these

437
00:22:25,960 --> 00:22:28,560
Speaker 1: data centers, and if that's the case, it isn't but

438
00:22:28,680 --> 00:22:30,040
Speaker 1: let me human the idea.

439
00:22:30,240 --> 00:22:30,359
Speaker 2: Then.

440
00:22:30,400 --> 00:22:32,840
Speaker 1: The suggestion is that Microsoft is currently changing its entire

441
00:22:32,920 --> 00:22:36,880
Speaker 1: data center's strategy so significantly they now has to issue

442
00:22:37,040 --> 00:22:39,160
Speaker 1: over a giga what's worth of statements of intent across

443
00:22:39,160 --> 00:22:44,200
Speaker 1: the country to different places because of more power. There's

444
00:22:44,280 --> 00:22:47,160
Speaker 1: more power in those places, less than did they make

445
00:22:47,640 --> 00:23:08,080
Speaker 1: like a giga what worth of mistakes? Now? Okay. Another

446
00:23:08,160 --> 00:23:10,760
Speaker 1: counter is that I'm only talking about leases and not purchases.

447
00:23:10,920 --> 00:23:13,440
Speaker 1: In that case, I'll refer you to this article from CBRE,

448
00:23:13,480 --> 00:23:15,720
Speaker 1: which is linked in the spreadsheet for this episode, and

449
00:23:15,760 --> 00:23:18,760
Speaker 1: which includes a looser dating read on how hyperscalers actually

450
00:23:18,840 --> 00:23:22,080
Speaker 1: invest in data center infrastructure. Leases tend to account for

451
00:23:22,080 --> 00:23:24,720
Speaker 1: the majority of spending simply because it's less risky. A

452
00:23:24,760 --> 00:23:28,680
Speaker 1: specialist takes care of the tough stuff location buying, land handling, construction,

453
00:23:29,000 --> 00:23:30,840
Speaker 1: and the hyperscaler isn't left trying to figure out what

454
00:23:30,880 --> 00:23:32,359
Speaker 1: to do with the facility when it reaches the end

455
00:23:32,359 --> 00:23:35,920
Speaker 1: of it's the useful life cycle. I also expect someone

456
00:23:35,960 --> 00:23:38,320
Speaker 1: to chime in and say well, that's this Microsoft, What

457
00:23:38,359 --> 00:23:41,600
Speaker 1: about Google and Amazon? Get out of my house. I'd

458
00:23:41,600 --> 00:23:44,440
Speaker 1: counter and say that these companies are comparatively less exposed

459
00:23:44,440 --> 00:23:48,280
Speaker 1: to general IFAI. Amazon has invested eight billion dollars in Anthropic,

460
00:23:48,280 --> 00:23:50,200
Speaker 1: which is a bit less than half of what Microsoft

461
00:23:50,280 --> 00:23:53,160
Speaker 1: is reportedly invested in open Ai, which amounted to about

462
00:23:53,160 --> 00:23:55,920
Speaker 1: fourteen billion dollars as of December. When you consider the

463
00:23:55,960 --> 00:23:58,960
Speaker 1: discounted asure rates Microsoft offers to open ai too, the

464
00:23:59,000 --> 00:24:01,600
Speaker 1: real number is probably my much much higher. Google also

465
00:24:01,680 --> 00:24:03,960
Speaker 1: has three billion dollars in Anthropic, in addition to its

466
00:24:03,960 --> 00:24:07,639
Speaker 1: own AI services like Gemini. Open Ai, as I noted

467
00:24:07,640 --> 00:24:10,600
Speaker 1: in my last newsletter, is pretty much the only real

468
00:24:10,640 --> 00:24:14,119
Speaker 1: generative AI company with market share and significant revenue, although

469
00:24:14,119 --> 00:24:16,320
Speaker 1: I once again remind you that revenue is not the

470
00:24:16,320 --> 00:24:19,000
Speaker 1: same thing as profim and this is true across mobile

471
00:24:19,000 --> 00:24:22,760
Speaker 1: web and likely it's APIs two. Similarly, nobody has quite

472
00:24:22,840 --> 00:24:26,320
Speaker 1: pushed generative AI as aggressively as Microsoft, which has introduced

473
00:24:26,320 --> 00:24:29,399
Speaker 1: it to an overwhelming number of its paid products, hiking

474
00:24:29,440 --> 00:24:32,000
Speaker 1: prices for customers as it goes. I suppose you could

475
00:24:32,000 --> 00:24:34,720
Speaker 1: say that Google has pushed GENII into its workspace products

476
00:24:34,720 --> 00:24:37,159
Speaker 1: as well as its sarch products, but the scale and

477
00:24:37,160 --> 00:24:40,880
Speaker 1: aggression of Microsoft's push feels different that, and as I've

478
00:24:40,920 --> 00:24:44,040
Speaker 1: mentioned repeatedly, they are the largest purchaser of Nvidia GPUs

479
00:24:44,080 --> 00:24:46,160
Speaker 1: by nearly twice as many four hundred and eighty five

480
00:24:46,200 --> 00:24:48,639
Speaker 1: thousand as its nearest competitor to Meta, which brought two

481
00:24:48,720 --> 00:24:53,440
Speaker 1: hundred and twenty four thousand in twenty twenty four. Ultimately,

482
00:24:53,480 --> 00:24:55,920
Speaker 1: Microsoft has positioned itself at the heart of generative AI,

483
00:24:56,000 --> 00:24:58,960
Speaker 1: both through its own strategic product decisions and its partnership

484
00:24:59,000 --> 00:25:01,359
Speaker 1: with open AI. And the fact that it's now scaling

485
00:25:01,440 --> 00:25:04,240
Speaker 1: back on the investment required to maintain that momentum is

486
00:25:04,280 --> 00:25:08,040
Speaker 1: I believe pretty significant. I also recognize that all of

487
00:25:08,040 --> 00:25:10,920
Speaker 1: this is a big juicy steake for someone some people

488
00:25:10,960 --> 00:25:13,040
Speaker 1: call a pig or an animal or a monster or

489
00:25:13,040 --> 00:25:16,600
Speaker 1: an AI cynic. Look, I've pulled over this data repeatedly

490
00:25:16,640 --> 00:25:19,520
Speaker 1: and done all that I can to find less convenient

491
00:25:19,600 --> 00:25:22,600
Speaker 1: or satisfying conclusions. Let us off intent and likely the

492
00:25:22,640 --> 00:25:25,119
Speaker 1: weakest part of my argument. These are serious documents, by

493
00:25:25,160 --> 00:25:27,840
Speaker 1: the way, but they're not always legally binding. Neither are

494
00:25:27,840 --> 00:25:31,600
Speaker 1: those statements of qualifications. But as Tdcoen pointed out, sqqs

495
00:25:31,600 --> 00:25:33,760
Speaker 1: are generally treated as the green lights stop working on

496
00:25:33,800 --> 00:25:36,760
Speaker 1: construction even though a formal lease agreement hasn't yet been signed,

497
00:25:37,119 --> 00:25:39,760
Speaker 1: and to be clear, Microsoft let an interminate amount of

498
00:25:39,800 --> 00:25:44,359
Speaker 1: sqqs go. Nevertheless, it's incredibly significant that Microsoft is letting

499
00:25:44,400 --> 00:25:46,720
Speaker 1: so many the equivalent of as much as fourteen percent

500
00:25:46,760 --> 00:25:49,200
Speaker 1: of its current data set of capacity at a bare minimum.

501
00:25:49,240 --> 00:25:51,159
Speaker 1: On top of the couple hundred so at least two

502
00:25:51,200 --> 00:25:55,160
Speaker 1: hundred megabats of data center leases become canceled. I do

503
00:25:55,200 --> 00:25:58,920
Speaker 1: not know why nobody else has done this analysis. I've

504
00:25:58,920 --> 00:26:01,680
Speaker 1: now read every single piece about the td co and

505
00:26:01,760 --> 00:26:04,359
Speaker 1: report from every single outlet the covenant. I've read some

506
00:26:04,400 --> 00:26:08,000
Speaker 1: weird seo stuff. It's it's not good, and I'm under

507
00:26:08,400 --> 00:26:11,800
Speaker 1: I'm just kind of astounded by the lack of curiosity

508
00:26:11,800 --> 00:26:14,439
Speaker 1: as to what one GW plus means and the report

509
00:26:14,680 --> 00:26:17,879
Speaker 1: the meaningfully moved markets, as I'm equally astonished by the

510
00:26:17,960 --> 00:26:21,600
Speaker 1: lack of curiosity to contextualize most tech news. It's as

511
00:26:21,640 --> 00:26:24,080
Speaker 1: if nobody wants to think about this too hard, like

512
00:26:24,160 --> 00:26:27,160
Speaker 1: nobody wants to stop the pie. Nobody wants to accept

513
00:26:27,200 --> 00:26:29,920
Speaker 1: what's being staring us in the face since last year,

514
00:26:30,040 --> 00:26:33,640
Speaker 1: if not earlier, and we're given the most egregious, glaring evidence.

515
00:26:34,160 --> 00:26:36,480
Speaker 1: People still must find ways to dismiss it or ignore

516
00:26:36,560 --> 00:26:39,000
Speaker 1: it rather than give it the energy it deserves. Far

517
00:26:39,080 --> 00:26:41,399
Speaker 1: more resources were dedicated to finding ways to gussy up

518
00:26:41,400 --> 00:26:43,960
Speaker 1: the releases of anthropics claudes on at three point seven

519
00:26:44,160 --> 00:26:46,520
Speaker 1: or open ais GPT four point five. And we're given

520
00:26:46,560 --> 00:26:49,679
Speaker 1: the report from an Investment Banks research wings that the

521
00:26:49,760 --> 00:26:52,640
Speaker 1: largest spender and generityve Ai, the largest backer for now

522
00:26:52,840 --> 00:26:56,119
Speaker 1: of open ai at least, is massively reducing its expenditures

523
00:26:56,119 --> 00:26:59,120
Speaker 1: in data centers required for the industry and for open Ai,

524
00:26:59,200 --> 00:27:01,800
Speaker 1: a company of stems worth one hundred and fifty seven billion

525
00:27:01,840 --> 00:27:06,119
Speaker 1: dollars to expand. Microsoft's stake in open Ai is a

526
00:27:06,160 --> 00:27:09,040
Speaker 1: bit fuzzy, is open Ai doesn't issue traditional equity, and

527
00:27:09,080 --> 00:27:11,440
Speaker 1: there's a likelihood it may be diluted as more money

528
00:27:11,520 --> 00:27:15,399
Speaker 1: comes in. It reportedly owned forty nine percent in twenty

529
00:27:15,400 --> 00:27:17,879
Speaker 1: twenty three, though assuming that's still the case, are we

530
00:27:18,000 --> 00:27:20,160
Speaker 1: to believe that Microsoft is willing to strangle and that's

531
00:27:20,160 --> 00:27:23,040
Speaker 1: it worth at least seventy five billion dollars several times

532
00:27:23,119 --> 00:27:25,840
Speaker 1: more than its investment today by canceling a few leases.

533
00:27:26,560 --> 00:27:28,520
Speaker 1: How many more alarms do we need to go off

534
00:27:28,560 --> 00:27:32,240
Speaker 1: before people recognize that something bad is happening. Why is

535
00:27:32,359 --> 00:27:35,439
Speaker 1: that tangible, meaningful evidence that we're in a bubble and

536
00:27:35,480 --> 00:27:38,480
Speaker 1: possibly a sign that it might be popping less interesting

537
00:27:38,520 --> 00:27:40,560
Speaker 1: than the fact that Claudes on it three point seven

538
00:27:40,600 --> 00:27:42,879
Speaker 1: can think longer. And if you're listening to this and

539
00:27:42,920 --> 00:27:45,359
Speaker 1: you think I'm talking about you, I fucking am. I

540
00:27:45,400 --> 00:27:49,080
Speaker 1: am sick of this shit. I am absolutely sick of

541
00:27:49,119 --> 00:27:51,680
Speaker 1: this shit. I'm sick of reading articles like that when

542
00:27:51,720 --> 00:27:54,280
Speaker 1: far more important and scary and damning things are happening,

543
00:27:54,440 --> 00:27:57,120
Speaker 1: things that actually matter. I don't care if Warrio Ama

544
00:27:57,200 --> 00:27:59,359
Speaker 1: Day has allowed you to make it compute for longer.

545
00:27:59,480 --> 00:28:01,760
Speaker 1: It doesn't are compared to this. It does not matter

546
00:28:01,760 --> 00:28:04,439
Speaker 1: compared to a gig or what or more of capacity going.

547
00:28:04,680 --> 00:28:07,200
Speaker 1: And I'm sick of this. I'm sick of me having

548
00:28:07,200 --> 00:28:10,080
Speaker 1: to be the guy. Sometimes. I do not say these

549
00:28:10,119 --> 00:28:12,199
Speaker 1: things to be right. I don't want to be a

550
00:28:12,240 --> 00:28:14,919
Speaker 1: cynic or a hater. I say this all because I

551
00:28:14,920 --> 00:28:16,919
Speaker 1: am trying to understand what's going on, and if I

552
00:28:16,960 --> 00:28:20,080
Speaker 1: do not, I will actually go insane. Every time I

553
00:28:20,119 --> 00:28:22,760
Speaker 1: sit down to write my newsletter or record this podcast,

554
00:28:23,240 --> 00:28:25,480
Speaker 1: I'm doing it because I'm trying to understand what's happening

555
00:28:25,480 --> 00:28:27,159
Speaker 1: and how I feel about it, And these are the

556
00:28:27,200 --> 00:28:30,320
Speaker 1: only terms to dictate my creativity. It just happens that

557
00:28:30,359 --> 00:28:33,280
Speaker 1: I've stared at the tech industry for too long and

558
00:28:33,359 --> 00:28:35,520
Speaker 1: now I can't look away. Perhaps it's driving me mad,

559
00:28:35,600 --> 00:28:37,399
Speaker 1: or maybe I'm getting smarter, or maybe it's an arm

560
00:28:37,480 --> 00:28:39,840
Speaker 1: or palmer of the two. But what comes out of

561
00:28:39,880 --> 00:28:41,960
Speaker 1: my work is not driven by wanting to go viral

562
00:28:42,000 --> 00:28:44,560
Speaker 1: or having a hot take, or be a renowned skeptic,

563
00:28:44,680 --> 00:28:48,400
Speaker 1: or being right or being anything, because such things suggest

564
00:28:48,400 --> 00:28:52,200
Speaker 1: that I would do this differently if three people listened

565
00:28:52,280 --> 00:28:55,760
Speaker 1: versus actually can't say the amount of people that do.

566
00:28:55,800 --> 00:28:58,680
Speaker 1: I have rules, but fifty seven thousand people subscribe to

567
00:28:58,720 --> 00:29:02,200
Speaker 1: my newsletter extra. From there, I would do the same

568
00:29:02,240 --> 00:29:04,080
Speaker 1: goddamn thing. And in fact, if you look back in

569
00:29:04,160 --> 00:29:07,520
Speaker 1: my work, I've done the same thing from the beginning. Now,

570
00:29:07,520 --> 00:29:09,560
Speaker 1: I'm not saying that Microsoft is dying or making any

571
00:29:09,560 --> 00:29:13,080
Speaker 1: grandiose claims about what happens next. What I am describing, however,

572
00:29:13,200 --> 00:29:16,120
Speaker 1: is the material contraction of the largest investor in data centers,

573
00:29:16,120 --> 00:29:19,160
Speaker 1: according to TD Cohen, potentially at a scale that suggests

574
00:29:19,440 --> 00:29:22,680
Speaker 1: the Microsoft has meaningfully reduced its interest in further expansion

575
00:29:22,680 --> 00:29:26,320
Speaker 1: of data centers. Writ large, this is a deeply concerning move,

576
00:29:26,520 --> 00:29:29,160
Speaker 1: one that suggests that Microsoft does not see demands sustain

577
00:29:29,240 --> 00:29:32,640
Speaker 1: the current expansions, which has greater ramifications beyond generative AI,

578
00:29:32,960 --> 00:29:35,320
Speaker 1: because it suggests that there isn't any other reason for

579
00:29:35,400 --> 00:29:38,840
Speaker 1: it to expand the means of delivering software. What has

580
00:29:38,880 --> 00:29:43,800
Speaker 1: sat in Adella scene? What is Microsoft cfo amy Hood doing?

581
00:29:44,120 --> 00:29:47,520
Speaker 1: What is the plan here? And really what's the plan

582
00:29:47,600 --> 00:29:50,960
Speaker 1: with open Ai? SoftBank has committed over forty billion dollars

583
00:29:51,000 --> 00:29:53,800
Speaker 1: of cost that it currently cannot afford, taking on as

584
00:29:53,880 --> 00:29:55,800
Speaker 1: much as twenty four billion dollars in debt in the

585
00:29:55,800 --> 00:29:58,400
Speaker 1: next year to help sustain one more funding round and

586
00:29:58,440 --> 00:30:01,440
Speaker 1: the construction of data centers. Open Ai, a company that

587
00:30:01,480 --> 00:30:05,240
Speaker 1: loses money on literally every single customer. To survive, open

588
00:30:05,280 --> 00:30:07,880
Speaker 1: Ai must continue raising more money than any startup has

589
00:30:07,920 --> 00:30:10,200
Speaker 1: ever raised before, and they are only able to do

590
00:30:10,240 --> 00:30:12,560
Speaker 1: so from SoftBank, which in turn must take on debt.

591
00:30:13,000 --> 00:30:15,479
Speaker 1: Open Ai burned five billion dollars in twenty twenty four

592
00:30:15,520 --> 00:30:17,640
Speaker 1: and more likely burn eleven billion dollars or more in

593
00:30:17,640 --> 00:30:20,720
Speaker 1: twenty twenty five, and will continue burning money in perpetuity

594
00:30:20,880 --> 00:30:23,360
Speaker 1: and to scale further will require funding for a data

595
00:30:23,360 --> 00:30:26,120
Speaker 1: center project funded partially by a funding from a company

596
00:30:26,160 --> 00:30:28,360
Speaker 1: that's taken on debt to fund them. And when you

597
00:30:28,360 --> 00:30:31,240
Speaker 1: put this all together, all I can see is calamity.

598
00:30:31,960 --> 00:30:35,120
Speaker 1: Generative AI does not have meaningful mass market use cases,

599
00:30:35,120 --> 00:30:37,640
Speaker 1: and while chat GPT may have four hundred million weekly

600
00:30:37,680 --> 00:30:41,160
Speaker 1: active users, there doesn't appear to be meaningful consumer adoption

601
00:30:41,280 --> 00:30:44,960
Speaker 1: outside of chat GPT, mostly because almost all AI media

602
00:30:45,000 --> 00:30:49,240
Speaker 1: coverage inevitably ends up marketing one company Open AI. Argue

603
00:30:49,240 --> 00:30:51,560
Speaker 1: with me all you want about your personal experiences with

604
00:30:51,640 --> 00:30:54,400
Speaker 1: chat GPT or how you personally found it useful. I

605
00:30:54,480 --> 00:30:57,720
Speaker 1: don't care. I stopped listening a while ago. Your points

606
00:30:57,760 --> 00:31:00,200
Speaker 1: never prove anything that doesn't make it a product with

607
00:31:00,240 --> 00:31:03,120
Speaker 1: mass market utility or enterprise utility, or worth the vast

608
00:31:03,120 --> 00:31:05,920
Speaker 1: sums of money being plowed into it. Worse still, that

609
00:31:06,000 --> 00:31:09,160
Speaker 1: doesn't appear to be any meaningful revenue. As discussed in

610
00:31:09,160 --> 00:31:12,680
Speaker 1: my last episode, Microsoft claims thirteen billion dollars in annual

611
00:31:12,680 --> 00:31:16,240
Speaker 1: recurring revenue not profit on all AI products combined on

612
00:31:16,320 --> 00:31:19,080
Speaker 1: over two hundred billion dollars of capital expenditures since twenty

613
00:31:19,120 --> 00:31:21,760
Speaker 1: twenty three, and no other hyperscaler is willing to break

614
00:31:21,800 --> 00:31:25,040
Speaker 1: out any AI revenue at all. Not Amazon, no matter,

615
00:31:25,360 --> 00:31:30,080
Speaker 1: not Google, Nobody does that. Not worry anyone. Is anyone

616
00:31:30,120 --> 00:31:33,520
Speaker 1: listening to this who actually deals with the economy? Can

617
00:31:33,560 --> 00:31:36,280
Speaker 1: you please listen to me? So we don't Actually I

618
00:31:36,320 --> 00:31:38,719
Speaker 1: don't know what we do. I actually don't know. I

619
00:31:38,800 --> 00:31:44,200
Speaker 1: have no idea. Where's the growth? Where's the money? Where's

620
00:31:44,240 --> 00:31:47,840
Speaker 1: the money? Sammy? Where is it? Where's my money? Honey,

621
00:31:47,960 --> 00:31:50,720
Speaker 1: give me the money, Sam Mormon? Where's my money? Why

622
00:31:50,800 --> 00:31:53,520
Speaker 1: is Microsoft canceling a gigawat of data center capacity while

623
00:31:53,560 --> 00:31:56,000
Speaker 1: telling everybody that it didn't have enough data centers to

624
00:31:56,000 --> 00:31:59,520
Speaker 1: handle demands for its AI products. I suppose this is

625
00:31:59,520 --> 00:32:01,880
Speaker 1: one way of looking at it. Microsoft may currently have

626
00:32:01,920 --> 00:32:04,920
Speaker 1: a capacity issue, but soon one, meaning that further expansion

627
00:32:04,960 --> 00:32:08,200
Speaker 1: is unnecessary. And that's the case, It'll be interesting to

628
00:32:08,240 --> 00:32:12,680
Speaker 1: see whether their peers follow suit either way. Look, I

629
00:32:12,760 --> 00:32:15,120
Speaker 1: see nothing that suggests that there's future growth in general

630
00:32:15,200 --> 00:32:17,320
Speaker 1: of AI. In fact, I think it's time for everybody

631
00:32:17,360 --> 00:32:19,720
Speaker 1: to seriously consider the big tech burn billions of dollars

632
00:32:19,720 --> 00:32:22,240
Speaker 1: on something that nobody ever wanted or would pay for.

633
00:32:23,520 --> 00:32:25,880
Speaker 1: If you listen to this and scoff, I don't know

634
00:32:25,920 --> 00:32:28,160
Speaker 1: what should I have talked about anthropic adding a sliding

635
00:32:28,200 --> 00:32:30,600
Speaker 1: thinking bar to a model GPT four point five. Who

636
00:32:30,640 --> 00:32:32,760
Speaker 1: gives a shit? Can you even tell me what it

637
00:32:32,800 --> 00:32:36,280
Speaker 1: does differently to GPT four to zero? Can you explain

638
00:32:36,360 --> 00:32:38,600
Speaker 1: to me why it matters? Or are you more interested

639
00:32:38,600 --> 00:32:41,880
Speaker 1: in nakedly captured imbeciles like Ethan Mollock sweatly oinking about

640
00:32:41,880 --> 00:32:46,360
Speaker 1: how powerful the latest ask and shit. It's like no

641
00:32:46,440 --> 00:32:48,840
Speaker 1: one's interested in the things happening in the real world.

642
00:32:49,040 --> 00:32:52,240
Speaker 1: It's like nobody's thinking about the silicon and the infrastructure

643
00:32:52,320 --> 00:32:56,080
Speaker 1: and how things actually get built. Wake the fuck up, everybody.

644
00:32:56,440 --> 00:33:07,760
Speaker 1: Things are on fire. Thank you for listening to Better Offline.

645
00:33:07,920 --> 00:33:10,320
Speaker 1: The editor and composer of the Better Offline theme song

646
00:33:10,400 --> 00:33:13,040
Speaker 1: is Matasowski. You can check out more of his music

647
00:33:13,080 --> 00:33:16,720
Speaker 1: and audio projects at Matasowski dot com, M A T

648
00:33:16,720 --> 00:33:21,200
Speaker 1: T O S O W s ki dot com. You

649
00:33:21,200 --> 00:33:23,720
Speaker 1: can email me at easy at better offline dot com,

650
00:33:23,840 --> 00:33:26,160
Speaker 1: or visit better Offline dot com to find more podcast

651
00:33:26,200 --> 00:33:29,520
Speaker 1: links and of course, my newsletter. I also really recommend

652
00:33:29,560 --> 00:33:31,480
Speaker 1: you go to chat dot Where's youreaed dot at to

653
00:33:31,560 --> 00:33:34,360
Speaker 1: visit the discord and go to our slash Better Offline

654
00:33:34,400 --> 00:33:37,600
Speaker 1: to check out our reddit. Thank you so much for listening.

655
00:33:38,440 --> 00:33:41,160
Speaker 2: Better Offline is a production of cool Zone Media. For

656
00:33:41,240 --> 00:33:44,440
Speaker 2: more from cool Zone Media, visit our website cool zonemedia

657
00:33:44,480 --> 00:33:47,320
Speaker 2: dot com, or check us out on the iHeartRadio app,

658
00:33:47,360 --> 00:34:08,080
Speaker 2: Apple Podcasts, or wherever you get your podcasts.

659
00:34:02,720 --> 00:34:02,760
Speaker 1: F