1
00:00:06,019 --> 00:00:09,979
Speaker 1: Welcome to CIT, a podcast series on markets and economies

2
00:00:09,978 --> 00:00:13,299
Speaker 1: from DBS Group Research. I'm Tamurbek, Chief economist. Welcoming you

3
00:00:13,300 --> 00:00:17,899
Speaker 1: to our 167th episode. Now, this is a first. We

4
00:00:17,899 --> 00:00:22,319
Speaker 1: are recording with someone in Hawaii. Noah Doyle, managing director

5
00:00:22,319 --> 00:00:25,779
Speaker 1: at Javelin Venture Partners, is based in Northern California, but

6
00:00:25,780 --> 00:00:27,649
Speaker 1: he's kind enough to take time out of his Hawaii

7
00:00:27,649 --> 00:00:31,120
Speaker 1: vacation to join us today. Noah has over 20 years

8
00:00:31,120 --> 00:00:34,700
Speaker 1: of background in entrepreneurship and management of innovation.

9
00:00:35,168 --> 00:00:38,019
Speaker 1: He's been a part of the genesis and growth phases

10
00:00:38,020 --> 00:00:42,060
Speaker 1: of numerous consequential tech firms. We will discuss the world

11
00:00:42,060 --> 00:00:45,380
Speaker 1: of investment in tech innovation with him today. Noah Doyle,

12
00:00:45,619 --> 00:00:47,099
Speaker 1: a warm welcome to Kobe time.

13
00:00:48,259 --> 00:00:49,869
Speaker 2: Thank you, Tr. It's great to be here.

14
00:00:50,598 --> 00:00:54,000
Speaker 1: It's super great to have you. I have a bunch

15
00:00:54,000 --> 00:00:58,389
Speaker 1: of questions around the tech ecosystem, but first, a general one, Noah,

16
00:00:58,520 --> 00:01:02,400
Speaker 1: what's the vibe in Silicon Valley as 2025 draws to

17
00:01:02,400 --> 00:01:02,840
Speaker 1: an end?

18
00:01:05,349 --> 00:01:10,150
Speaker 2: Yeah, great topic to start with. Um, it's, it's definitely a, um,

19
00:01:10,300 --> 00:01:14,949
Speaker 2: a vibe where, um, ah, you know, it, it's hard

20
00:01:14,949 --> 00:01:20,949
Speaker 2: to really put the enthusiasm level into words. Um, you've had, ah,

21
00:01:21,150 --> 00:01:24,730
Speaker 2: just incredible, um, breakout growth.

22
00:01:25,959 --> 00:01:32,120
Speaker 2: Across many different companies and um almost every high-tech startup,

23
00:01:32,279 --> 00:01:36,479
Speaker 2: high growth, high tech startup has been impacted by the

24
00:01:36,480 --> 00:01:40,199
Speaker 2: new AI tools that are available. They're using them to

25
00:01:40,199 --> 00:01:45,139
Speaker 2: become more efficient, to, to grow faster, to, to innovate

26
00:01:45,639 --> 00:01:48,099
Speaker 2: in ways that, uh, you know, we're really just

27
00:01:48,379 --> 00:01:51,620
Speaker 2: It feels like we're we're in the early innings or

28
00:01:51,620 --> 00:01:55,220
Speaker 2: at the tip of the iceberg of what's possible, and

29
00:01:55,220 --> 00:01:59,580
Speaker 2: I think that enthusiasm, you know, is, is, is just

30
00:01:59,580 --> 00:02:03,040
Speaker 2: palpable in, in the tech world.

31
00:02:03,500 --> 00:02:07,589
Speaker 2: And it, it's, uh, it really transcends, you know, uh,

32
00:02:07,819 --> 00:02:11,500
Speaker 2: almost every sector, you know, from, um, you know, manufacturing

33
00:02:11,500 --> 00:02:14,978
Speaker 2: to life science to the classic, you know, software and

34
00:02:14,979 --> 00:02:19,500
Speaker 2: internet spaces. So, um, uh, I think that, uh, you know,

35
00:02:19,600 --> 00:02:23,119
Speaker 2: it's attracting a lot of mainstream attention.

36
00:02:23,940 --> 00:02:27,979
Speaker 2: You know, and, uh, a lot of investment, um, and, uh,

37
00:02:28,020 --> 00:02:32,220
Speaker 2: I'm sure you're going to go there, but, uh, you know, um, it's, it's, uh,

38
00:02:32,300 --> 00:02:37,179
Speaker 2: it's been hard to contain the enthusiasm level in some respects.

39
00:02:38,508 --> 00:02:42,519
Speaker 1: Um, give us a sense of this year's sort of

40
00:02:42,520 --> 00:02:47,470
Speaker 1: the growth stories or the returns that people are looking at. Uh, I,

41
00:02:47,600 --> 00:02:49,359
Speaker 1: I want to head into the direction of, you know,

42
00:02:49,479 --> 00:02:51,320
Speaker 1: are we in a bubble, but not yet. Just help

43
00:02:51,320 --> 00:02:53,559
Speaker 1: us set the background a little bit about what sort

44
00:02:53,559 --> 00:02:56,990
Speaker 1: of returns are people looking at. Are people actually getting exits?

45
00:02:57,309 --> 00:03:00,160
Speaker 1: So you're in the VC world. Are you guys sort

46
00:03:00,160 --> 00:03:04,478
Speaker 1: of satisfied with the deal flow that's taking place right now?

47
00:03:05,940 --> 00:03:09,299
Speaker 2: Um, yeah, that's a great question. So, um, I think that,

48
00:03:09,380 --> 00:03:11,809
Speaker 2: you know, maybe to answer the first part of it,

49
00:03:12,100 --> 00:03:17,418
Speaker 2: you know, we are seeing, um, rapidly escalating valuations for

50
00:03:17,419 --> 00:03:21,779
Speaker 2: the leading companies. Um, so, um, on the heels of

51
00:03:21,779 --> 00:03:28,850
Speaker 2: a valuation in the $60 billion range for data bricks,

52
00:03:28,899 --> 00:03:29,519
Speaker 2: you had

53
00:03:29,758 --> 00:03:37,270
Speaker 2: A, um, uh, uh, you know, series of fundraisers by OpenAI, um,

54
00:03:37,679 --> 00:03:42,720
Speaker 2: most recently, you know, um, uh, in the $300 billion range,

55
00:03:42,759 --> 00:03:45,779
Speaker 2: and then, you know, I guess a rumor that they're

56
00:03:46,919 --> 00:03:49,919
Speaker 2: doing it or looking to do a transaction at a

57
00:03:49,919 --> 00:03:54,330
Speaker 2: step up from that, you know, possibly $500 billion. Anthropic

58
00:03:54,679 --> 00:03:59,539
Speaker 2: just completed a fundraise, um, over $100 billion and then

59
00:04:00,009 --> 00:04:06,070
Speaker 2: Did another quick follow-on transaction at the $130 billion range.

60
00:04:07,410 --> 00:04:11,550
Speaker 2: So these are, these are just really incredible numbers, um,

61
00:04:11,889 --> 00:04:15,919
Speaker 2: you know, totally unprecedented, you know, kinds of levels of

62
00:04:15,919 --> 00:04:20,690
Speaker 2: valuations for, you know, private companies, especially to be escalating

63
00:04:20,690 --> 00:04:21,390
Speaker 2: so quickly.

64
00:04:22,970 --> 00:04:26,809
Speaker 2: But you know, the revenue growth rates have been stupendous,

65
00:04:26,890 --> 00:04:31,250
Speaker 2: and the companies are operating, you know, with revenues in

66
00:04:31,250 --> 00:04:35,409
Speaker 2: the billions of dollars, you know, much faster than businesses

67
00:04:35,410 --> 00:04:39,329
Speaker 2: have historically been able to achieve. So it does feel

68
00:04:39,329 --> 00:04:41,970
Speaker 2: like there's, there's, uh, you know, it's hard to say

69
00:04:41,970 --> 00:04:43,928
Speaker 2: what's justified.

70
00:04:44,529 --> 00:04:48,019
Speaker 2: Or what isn't, you know, it's a guessing game, of course, and, ah,

71
00:04:48,359 --> 00:04:51,269
Speaker 2: you know, investors are, are voting, you know, with, with

72
00:04:51,269 --> 00:04:55,640
Speaker 2: their wallets, um, but the votes just, ah, keep, keep

73
00:04:55,640 --> 00:04:59,390
Speaker 2: getting bigger and bigger tallies. Um, so I think that's,

74
00:04:59,450 --> 00:05:03,260
Speaker 2: that's kind of the backdrop, but the, the, the other, um,

75
00:05:03,279 --> 00:05:06,000
Speaker 2: aspect of it is that, you know, each, each of

76
00:05:06,000 --> 00:05:08,640
Speaker 2: these companies started, um,

77
00:05:08,959 --> 00:05:12,399
Speaker 2: With a small team of people and a few million dollars,

78
00:05:12,720 --> 00:05:17,079
Speaker 2: you know, and that's really what, um, you know, where

79
00:05:17,079 --> 00:05:20,440
Speaker 2: the value is created in venture capital, you know, is

80
00:05:20,440 --> 00:05:25,950
Speaker 2: in identifying those teams and getting those, those early ownership stakes,

81
00:05:25,959 --> 00:05:29,250
Speaker 2: and that's what Javelin is focused on. So, um, we're

82
00:05:29,250 --> 00:05:33,678
Speaker 2: really investing in what we call pre-traction, signal before traction

83
00:05:33,678 --> 00:05:35,039
Speaker 2: is our motto.

84
00:05:35,428 --> 00:05:39,690
Speaker 2: And we're looking for companies where there's an indication that

85
00:05:40,109 --> 00:05:45,029
Speaker 2: there's enormous potential to achieve breakout growth, um, um, but

86
00:05:45,029 --> 00:05:49,709
Speaker 2: it hasn't yet been demonstrated, so we're putting the pieces

87
00:05:49,709 --> 00:05:53,750
Speaker 2: together to form a puzzle that ultimately, you know, paints

88
00:05:53,750 --> 00:05:58,029
Speaker 2: a picture of a company that can become very successful, um,

89
00:05:58,269 --> 00:06:02,989
Speaker 2: and then we're, we're getting a comparatively high ownership stake

90
00:06:02,988 --> 00:06:05,149
Speaker 2: for a few, a few millions of dollars.

91
00:06:05,345 --> 00:06:10,635
Speaker 2: Hours, um, you know, sub $10 million funding rounds, and

92
00:06:10,635 --> 00:06:13,953
Speaker 2: we're usually leading it, taking the board seat and working

93
00:06:13,954 --> 00:06:17,505
Speaker 2: with the companies closely to get to that next level, um,

94
00:06:17,915 --> 00:06:24,183
Speaker 2: but I think that's that whole arena still remains very vibrant. Um,

95
00:06:24,234 --> 00:06:28,295
Speaker 2: we're seeing more deal flow, higher quality deal flow, companies

96
00:06:28,295 --> 00:06:30,015
Speaker 2: that have achieved more with less.

97
00:06:30,440 --> 00:06:34,959
Speaker 2: Than ever before, and uh I think that's, um, also,

98
00:06:35,029 --> 00:06:37,399
Speaker 2: you know, because that's in many ways the bread and

99
00:06:37,399 --> 00:06:42,320
Speaker 2: butter of the venture capital industry, um, it's, it's, uh,

100
00:06:42,399 --> 00:06:43,200
Speaker 2: it's encouraging.

101
00:06:44,130 --> 00:06:47,269
Speaker 2: And how it's supporting the whole ecosystem, yeah.

102
00:06:47,649 --> 00:06:48,010
Speaker 1: And how

103
00:06:48,010 --> 00:06:51,488
Speaker 1: is this different from the past cycles that you've been

104
00:06:51,488 --> 00:06:52,010
Speaker 1: involved in?

105
00:06:54,720 --> 00:06:58,350
Speaker 2: Yeah, that's a great question. I think in some cases, in, in,

106
00:06:58,428 --> 00:07:04,350
Speaker 2: in some respects, um, you have more experienced founders, um,

107
00:07:04,440 --> 00:07:08,959
Speaker 2: you have, uh, much better tools for them to build

108
00:07:08,959 --> 00:07:13,700
Speaker 2: their products. Um, you have better established paths to market

109
00:07:14,119 --> 00:07:16,779
Speaker 2: to introduce them and start to generate revenue.

110
00:07:17,140 --> 00:07:22,059
Speaker 2: Um, you, you have a lot more of the kind

111
00:07:22,059 --> 00:07:26,980
Speaker 2: of pieces needed to create a great business are more

112
00:07:26,980 --> 00:07:31,220
Speaker 2: mature and better defined than they ever have been, and

113
00:07:31,220 --> 00:07:35,230
Speaker 2: that's just enabling companies to grow faster, you know, once they,

114
00:07:35,579 --> 00:07:39,959
Speaker 2: once they get established, you know, we, we were investors

115
00:07:40,140 --> 00:07:44,660
Speaker 2: in Niantic, the company that created Pokemon Go.

116
00:07:45,220 --> 00:07:50,790
Speaker 2: And when they launched Pokemon Go, because of that mobile

117
00:07:50,790 --> 00:07:54,089
Speaker 2: app ecosystem, you know, they, they were able to achieve

118
00:07:54,089 --> 00:07:57,989
Speaker 2: hundreds of millions of users in a matter of months,

119
00:07:58,179 --> 00:08:01,290
Speaker 2: you know, it's something that just wasn't physically really possible,

120
00:08:01,950 --> 00:08:05,549
Speaker 2: you know, before that ecosystem was in place. So if

121
00:08:05,549 --> 00:08:09,690
Speaker 2: companies are able to achieve that lightning in a bottle,

122
00:08:10,390 --> 00:08:10,809
Speaker 2: then

123
00:08:11,149 --> 00:08:15,380
Speaker 2: They really can grow faster than, you know, the, the

124
00:08:15,390 --> 00:08:19,709
Speaker 2: new records are being set regularly in terms of, you know,

125
00:08:19,829 --> 00:08:22,609
Speaker 2: how quickly can a company get to $100 million.

126
00:08:24,089 --> 00:08:26,369
Speaker 1: Right, I guess, you know, OpenAI really has a sort

127
00:08:26,369 --> 00:08:29,600
Speaker 1: of trailed uh in, in that area or that really

128
00:08:29,600 --> 00:08:32,669
Speaker 1: been a trailblazer there, uh, because I see all these charts,

129
00:08:33,210 --> 00:08:36,010
Speaker 1: your former employer, you know, Google, for example, the time

130
00:08:36,010 --> 00:08:38,409
Speaker 1: it took them to reach scale versus how long it

131
00:08:38,409 --> 00:08:41,090
Speaker 1: takes Open Air to reach scale. We thought that was

132
00:08:41,090 --> 00:08:43,770
Speaker 1: super fast, and now they're redefining what superfast

133
00:08:43,770 --> 00:08:44,090
Speaker 1: is.

134
00:08:46,270 --> 00:08:50,599
Speaker 2: Yeah, it's, it's, it's, um, it, it, and it's the record,

135
00:08:50,760 --> 00:08:54,000
Speaker 2: you know, has been broken, like the time to break it,

136
00:08:54,080 --> 00:08:59,580
Speaker 2: you know, it seems faster and faster each time around, um, and, uh,

137
00:08:59,679 --> 00:09:04,330
Speaker 2: you know, that's, that's definitely getting mainstream attention. So the

138
00:09:04,599 --> 00:09:09,719
Speaker 2: You know, public investing world, you know, is looking very

139
00:09:09,719 --> 00:09:14,270
Speaker 2: closely at what's happening on the private side, and, you know,

140
00:09:14,559 --> 00:09:18,770
Speaker 2: they're definitely knocking on the door, so it's, I think,

141
00:09:18,919 --> 00:09:19,640
Speaker 2: I think it's no.

142
00:09:19,784 --> 00:09:23,824
Speaker 2: Surprise, you know, that the dollar, you know, levels of

143
00:09:23,825 --> 00:09:27,905
Speaker 2: investment in venture are starting to go up, um, go

144
00:09:27,905 --> 00:09:31,984
Speaker 2: up again, and, uh, um, you know, you had a, you,

145
00:09:32,184 --> 00:09:36,184
Speaker 2: you have a number of factors that have been challenging

146
00:09:36,184 --> 00:09:40,304
Speaker 2: for venture in the past, um, 3 years since the, uh,

147
00:09:40,344 --> 00:09:45,585
Speaker 2: since the end of COVID, um, where, uh, during COVID, um,

148
00:09:45,775 --> 00:09:49,304
Speaker 2: you know, all, all digital businesses, right, all tech.

149
00:09:49,630 --> 00:09:54,619
Speaker 2: You know, expanded very rapidly in the post-COVID environment, um,

150
00:09:54,830 --> 00:09:56,969
Speaker 2: you know, with the increase in interest rates and the

151
00:09:56,969 --> 00:10:01,468
Speaker 2: reset of valuations and multiples and tech companies, a lot

152
00:10:01,469 --> 00:10:08,689
Speaker 2: of the investment interest in venture capital waned, uh, particularly because, um,

153
00:10:08,869 --> 00:10:12,209
Speaker 2: valuations had run up, but the distributions were behind.

154
00:10:12,669 --> 00:10:17,460
Speaker 2: Because of the the difficulty of going public today, a

155
00:10:17,460 --> 00:10:19,348
Speaker 2: company needs to be a lot bigger than it used

156
00:10:19,349 --> 00:10:22,869
Speaker 2: to be to be able to get public, and the

157
00:10:22,869 --> 00:10:30,089
Speaker 2: slowdown in M&A with stricter antitrust enforcement and

158
00:10:31,000 --> 00:10:33,849
Speaker 2: Ah, you know, really just, uh, I think as the,

159
00:10:33,950 --> 00:10:37,559
Speaker 2: the large tech companies have gotten bigger and bigger and

160
00:10:37,559 --> 00:10:40,239
Speaker 2: they're looking for acquisitions that really moved the needle for them,

161
00:10:40,320 --> 00:10:42,669
Speaker 2: there are fewer and fewer of those companies that are

162
00:10:42,669 --> 00:10:48,380
Speaker 2: also of comparable scale. So it's, it's, uh, it's led to, um,

163
00:10:48,679 --> 00:10:53,840
Speaker 2: you know, a pullback, you know, from LPs in the

164
00:10:53,840 --> 00:10:57,479
Speaker 2: sector in the past 3 years, and that's now getting

165
00:10:57,479 --> 00:10:58,650
Speaker 2: kind of counteracted.

166
00:10:59,190 --> 00:11:03,228
Speaker 2: By a lot of new interest from newer investors who

167
00:11:03,229 --> 00:11:05,669
Speaker 2: hadn't been in the venture capital sector before who are

168
00:11:05,669 --> 00:11:08,989
Speaker 2: coming in. Um, so we're, we're starting to see things

169
00:11:08,989 --> 00:11:09,869
Speaker 2: turn in that regard.

170
00:11:11,099 --> 00:11:15,140
Speaker 1: Who are these new investors? Are they international or are

171
00:11:15,140 --> 00:11:19,098
Speaker 1: they US or North America based uh real money who

172
00:11:19,099 --> 00:11:20,640
Speaker 1: were not into VC before?

173
00:11:21,590 --> 00:11:26,109
Speaker 2: It's, it's, yeah, it's, it's, it's mostly international, I would say.

174
00:11:26,349 --> 00:11:33,590
Speaker 2: So actually European investors historically have a relatively small percentage

175
00:11:33,590 --> 00:11:38,348
Speaker 2: of assets allocated to alternative and venture capital was the

176
00:11:38,349 --> 00:11:40,049
Speaker 2: smallest of that allocation.

177
00:11:40,950 --> 00:11:48,419
Speaker 2: So many, many large family office groups and multi, multi-sector

178
00:11:48,419 --> 00:11:53,020
Speaker 2: investment groups had limited to no venture capital exposure at

179
00:11:53,020 --> 00:11:57,780
Speaker 2: all in Europe and probably nothing in North America, even

180
00:11:57,780 --> 00:12:00,780
Speaker 2: if they had some in Europe. So that's a big,

181
00:12:00,900 --> 00:12:06,500
Speaker 2: that's a very big pool of capital, and then definitely,

182
00:12:06,619 --> 00:12:09,468
Speaker 2: you know, across Asia as well.

183
00:12:09,840 --> 00:12:15,229
Speaker 2: Um, you know, in economies like India, India and, um, uh,

184
00:12:15,559 --> 00:12:19,409
Speaker 2: Hong Kong, Singapore, um, you know, you had a, uh,

185
00:12:19,770 --> 00:12:24,989
Speaker 2: in general preference toward, um, a more traditional investment style,

186
00:12:25,090 --> 00:12:29,689
Speaker 2: cash flow centric, you know, value-based, um, you know, venture

187
00:12:29,690 --> 00:12:32,728
Speaker 2: was always a little bit, a little bit perceived as

188
00:12:32,729 --> 00:12:33,750
Speaker 2: risky and out there.

189
00:12:34,479 --> 00:12:38,530
Speaker 2: And I think a lot of those, especially the larger

190
00:12:38,530 --> 00:12:41,609
Speaker 2: family offices that haven't had that exposure are now getting

191
00:12:41,609 --> 00:12:46,929
Speaker 2: very interested in it. So that that's happening for sure,

192
00:12:48,770 --> 00:12:54,169
Speaker 2: and you're also seeing established groups. You're seeing sovereign sovereign funds,

193
00:12:54,210 --> 00:12:59,510
Speaker 2: you're seeing hedge funds crossover, you know, from private to public,

194
00:12:59,849 --> 00:13:04,130
Speaker 2: more so. So, you know, that's all, that's all driving it.

195
00:13:04,559 --> 00:13:08,659
Speaker 2: Um, uh, you know, I think if you look at the, the,

196
00:13:08,729 --> 00:13:12,630
Speaker 2: the pace of fundraising, you know, it's, it's, it's still below,

197
00:13:13,200 --> 00:13:16,969
Speaker 2: you know, where it was 34 years ago, um, but it's, it's, uh,

198
00:13:17,130 --> 00:13:18,929
Speaker 2: I think the signs that it will turn and it'll

199
00:13:18,929 --> 00:13:20,210
Speaker 2: start to pick up again are there.

200
00:13:21,460 --> 00:13:24,500
Speaker 1: It's fascinating. On, on one hand, we don't have like

201
00:13:24,500 --> 00:13:30,059
Speaker 1: the greatest ever capital market momentum. I read articles about

202
00:13:30,059 --> 00:13:35,299
Speaker 1: private equity firms are lengthening their exit average duration. Uh, I,

203
00:13:35,349 --> 00:13:38,859
Speaker 1: I just what you're saying that, you know, you have

204
00:13:38,859 --> 00:13:41,020
Speaker 1: a crop of investors coming in, but it's still not

205
00:13:41,020 --> 00:13:43,900
Speaker 1: like an all-time record high in terms of deployment. But

206
00:13:43,900 --> 00:13:45,979
Speaker 1: at the same time, when we read the headlines from

207
00:13:45,979 --> 00:13:47,900
Speaker 1: the opening eyes of the world and the kinds of

208
00:13:47,900 --> 00:13:50,419
Speaker 1: investment they're doing, it's just absolutely jaw-dropping.

209
00:13:50,989 --> 00:13:53,280
Speaker 1: So there is this whole ecosystem, Noah, it seems to

210
00:13:53,280 --> 00:13:57,839
Speaker 1: me of fundraising beyond the whole VC ecosystem, which is

211
00:13:57,840 --> 00:14:01,119
Speaker 1: just the private credit world, private debt world where the

212
00:14:01,119 --> 00:14:03,640
Speaker 1: opening eyes of the world are going and creating syndicates

213
00:14:03,640 --> 00:14:06,590
Speaker 1: of banks and so on. So, is that just a

214
00:14:06,590 --> 00:14:09,039
Speaker 1: separate world from your world or you guys also get

215
00:14:09,039 --> 00:14:10,020
Speaker 1: to be part of that?

216
00:14:11,820 --> 00:14:14,859
Speaker 2: Um, that, that's a good question. I think, you know, it's, it's,

217
00:14:15,099 --> 00:14:18,489
Speaker 2: those are really direct investments in the companies, you know,

218
00:14:18,650 --> 00:14:21,849
Speaker 2: and I think the companies, as they, you know, um,

219
00:14:22,059 --> 00:14:27,099
Speaker 2: get scale, they're able to tap much more established pools

220
00:14:27,099 --> 00:14:30,640
Speaker 2: of capital, in many cases, um, public pools of capital

221
00:14:31,099 --> 00:14:38,940
Speaker 2: where you have, um, uh, you know, big crossover investors like, um, Fidelity, um, uh, Wellington.

222
00:14:39,330 --> 00:14:45,400
Speaker 2: You know, um, Vanguard, you know, they're, they're actually trying to, they, they,

223
00:14:45,409 --> 00:14:49,679
Speaker 2: they have a specific strategy to take stakes in tech

224
00:14:49,679 --> 00:14:52,770
Speaker 2: companies that are likely to go public in order to

225
00:14:52,770 --> 00:14:55,130
Speaker 2: start building a stake that then they would add to

226
00:14:55,130 --> 00:15:00,820
Speaker 2: once they're public. Um, so for the companies themselves, those

227
00:15:00,820 --> 00:15:03,809
Speaker 2: sources do open up for the funds, it's, it's a

228
00:15:03,809 --> 00:15:06,500
Speaker 2: little bit of a different ballgame because

229
00:15:06,859 --> 00:15:10,340
Speaker 2: You're looking at LP commitments that get drawn down over

230
00:15:10,340 --> 00:15:14,580
Speaker 2: the life of a fund, so you need institutions that

231
00:15:14,580 --> 00:15:19,719
Speaker 2: have a, um, you know, capital planning cycle and, you know, have, um, uh,

232
00:15:20,140 --> 00:15:21,020
Speaker 2: you know, uh.

233
00:15:21,419 --> 00:15:25,830
Speaker 2: Have, have a committed capital model and it's it's, it's a,

234
00:15:26,020 --> 00:15:29,010
Speaker 2: it's a bit of a different marketplace, um, you know,

235
00:15:29,059 --> 00:15:32,140
Speaker 2: venture capital, you know, even though it gets a lot

236
00:15:32,140 --> 00:15:36,460
Speaker 2: of attention, it's still relatively small in the scheme of things,

237
00:15:36,570 --> 00:15:40,099
Speaker 2: you know, compared to even growth equity or, you know,

238
00:15:40,219 --> 00:15:44,380
Speaker 2: certainly LBO, you know, it's still like less, it's probably

239
00:15:44,380 --> 00:15:48,919
Speaker 2: just 6, 7%, you know, of the total private equity universe.

240
00:15:49,429 --> 00:15:49,830
Speaker 2: Right,

241
00:15:50,150 --> 00:15:53,109
Speaker 1: but a very critical part because that's where the

242
00:15:53,109 --> 00:15:55,549
Speaker 2: seeds becomes very influential, yeah, when it, when it works,

243
00:15:55,630 --> 00:15:56,890
Speaker 2: it's very influential.

244
00:15:57,469 --> 00:16:01,419
Speaker 1: Indeed, indeed. Uh, OK, Noah, the billion dollars or is

245
00:16:01,419 --> 00:16:04,549
Speaker 1: it a trillion dollar question? Are we in an AI bubble?

246
00:16:05,969 --> 00:16:05,979
Speaker 2: Man

247
00:16:07,669 --> 00:16:10,409
Speaker 2: Well, it's, yeah, it's, it's a, it's a, it is

248
00:16:10,549 --> 00:16:14,890
Speaker 2: a complicated question. It certainly feels like it on some days,

249
00:16:16,229 --> 00:16:21,510
Speaker 2: and one of the things we've been in discussions with

250
00:16:21,510 --> 00:16:25,770
Speaker 2: our investor base around this question that they're, they're getting

251
00:16:26,190 --> 00:16:30,510
Speaker 2: a bit concerned about, um, you know, how capital is

252
00:16:30,510 --> 00:16:32,090
Speaker 2: getting deployed and, you know,

253
00:16:32,510 --> 00:16:35,390
Speaker 2: How valuations are changing and, you know, how to interpret

254
00:16:35,390 --> 00:16:37,630
Speaker 2: all this. Um, so we did a little bit of

255
00:16:37,630 --> 00:16:43,989
Speaker 2: an analysis of, um, venture capital deployments and um what

256
00:16:43,989 --> 00:16:50,619
Speaker 2: they are, um, uh, what they're trending towards, um, and, uh,

257
00:16:50,630 --> 00:16:54,710
Speaker 2: one of the, you know, um, developments of the, you know,

258
00:16:54,950 --> 00:16:59,320
Speaker 2: past six months is that 64% of venture investment.

259
00:16:59,979 --> 00:17:04,979
Speaker 2: Went into AI companies, um, you know, in, in the,

260
00:17:05,140 --> 00:17:08,060
Speaker 2: in the, you know, in the first half of um

261
00:17:09,040 --> 00:17:14,718
Speaker 2: Of 2025 and, you know, I think that I'm just

262
00:17:14,719 --> 00:17:20,479
Speaker 2: sharing this slide with you and um uh actually is it.

263
00:17:21,849 --> 00:17:24,349
Speaker 2: This screen sharing is paused.

264
00:17:27,339 --> 00:17:29,829
Speaker 2: Well, I can send the slide to you later. Um,

265
00:17:30,270 --> 00:17:32,169
Speaker 2: maybe I'll just describe it for now. OK.

266
00:17:32,349 --> 00:17:35,410
Speaker 1: Uh, yeah, it seems like I've got some firewall issue.

267
00:17:35,510 --> 00:17:37,430
Speaker 1: So although you have the right to share some, it's

268
00:17:37,430 --> 00:17:39,910
Speaker 1: not happening, just read it out for us, Noah, and

269
00:17:39,910 --> 00:17:42,109
Speaker 1: I'll make sure that I'll read it out viewers get to.

270
00:17:42,310 --> 00:17:43,228
Speaker 1: Oh no, here we go. I can

271
00:17:43,229 --> 00:17:45,040
Speaker 2: see it. Oh, there it goes. OK, OK, great. Uh,

272
00:17:45,069 --> 00:17:47,349
Speaker 2: I just selected on my side. OK, um.

273
00:17:47,459 --> 00:17:53,188
Speaker 2: So if, if you look at total investment dollars that

274
00:17:53,250 --> 00:17:58,349
Speaker 2: were deployed in venture capital, um, you know, there was

275
00:17:58,770 --> 00:18:04,208
Speaker 2: um a, a run rate um in, in, you know,

276
00:18:04,290 --> 00:18:06,859
Speaker 2: the late 20, late.

277
00:18:07,510 --> 00:18:13,448
Speaker 2: Um, 2010, right, so kind of 2015 to 2020, venture

278
00:18:13,449 --> 00:18:18,609
Speaker 2: capital moved up from the from the historic range of

279
00:18:18,609 --> 00:18:21,889
Speaker 2: the $30 to $40 billion dollar level, and it, it,

280
00:18:21,969 --> 00:18:25,770
Speaker 2: you know, stepped up to about $100 billion give or take.

281
00:18:26,209 --> 00:18:32,089
Speaker 2: Then post-COVID, there was a massive run up to $359

282
00:18:32,089 --> 00:18:36,119
Speaker 2: billion you know, total deployed, yeah, and, um, it's, it's

283
00:18:36,119 --> 00:18:36,869
Speaker 2: since then.

284
00:18:37,219 --> 00:18:42,300
Speaker 2: Um, basically reverted to that 100 billion level. Um, so

285
00:18:42,300 --> 00:18:51,790
Speaker 2: you had, um, in 2024, 2023, 165 billion, 2024, 215 billion, um,

286
00:18:52,420 --> 00:18:57,459
Speaker 2: but in that 2023 period, the AI boom, you know,

287
00:18:57,579 --> 00:18:58,640
Speaker 2: really started.

288
00:18:59,069 --> 00:19:03,910
Speaker 2: And a significant portion of those dollars were going to

289
00:19:03,910 --> 00:19:07,750
Speaker 2: AI companies, and one of the unique aspects of this

290
00:19:07,750 --> 00:19:14,469
Speaker 2: particular investment cycle is that these companies take enormous resources

291
00:19:14,469 --> 00:19:17,550
Speaker 2: to build their products. So, um, to build a state

292
00:19:17,550 --> 00:19:21,569
Speaker 2: of the art LLM today does require

293
00:19:22,689 --> 00:19:25,380
Speaker 2: Billions of dollars, you know, it's probably a minimum of

294
00:19:25,380 --> 00:19:30,020
Speaker 2: $1 billion to build a, a model that compete, can

295
00:19:30,020 --> 00:19:33,938
Speaker 2: compete at the top level and, uh, arguably, you know, could,

296
00:19:34,060 --> 00:19:37,579
Speaker 2: could be more like 2 to $2.5 billion required. Um,

297
00:19:38,459 --> 00:19:42,739
Speaker 2: so the companies have been showing enormous growth, but they're

298
00:19:42,739 --> 00:19:47,599
Speaker 2: also incredibly capital hungry, um, and a lot of the, the,

299
00:19:47,619 --> 00:19:50,920
Speaker 2: the funding interest in the companies.

300
00:19:51,369 --> 00:19:57,349
Speaker 2: has also been coming from the hyperscalers where the companies

301
00:19:57,589 --> 00:20:02,650
Speaker 2: will receive investment from Google, Microsoft, Nvidia, then turn around

302
00:20:02,650 --> 00:20:06,000
Speaker 2: and spend that money with them to actually build their

303
00:20:06,000 --> 00:20:08,150
Speaker 2: models and do the, do the um

304
00:20:09,030 --> 00:20:13,030
Speaker 2: Do the training and utilize their compute capacity to create

305
00:20:13,030 --> 00:20:17,709
Speaker 2: and serve their product. Um, so it's, it's created really

306
00:20:17,709 --> 00:20:23,750
Speaker 2: an ecosystem that has concentrated a lot of investment in

307
00:20:23,750 --> 00:20:31,250
Speaker 2: this AI sector. And um if, if you take that 64%

308
00:20:31,510 --> 00:20:35,109
Speaker 2: of investment in the first half of 2025.

309
00:20:35,569 --> 00:20:39,229
Speaker 2: That went into AI companies, but you sub you

310
00:20:40,229 --> 00:20:45,000
Speaker 2: Sub-segmented by the amount that went into very, very large rounds,

311
00:20:45,079 --> 00:20:49,420
Speaker 2: so rounds of $250 million or more, which, you know, historically,

312
00:20:50,020 --> 00:20:52,060
Speaker 2: a $250 million round.

313
00:20:52,650 --> 00:20:55,409
Speaker 2: In venture capital was was extremely rare, you know, it

314
00:20:55,410 --> 00:20:58,129
Speaker 2: would have been a pre-IPO type mezzanine round for a

315
00:20:58,130 --> 00:21:02,010
Speaker 2: very large late stage company. So if you, if you

316
00:21:02,010 --> 00:21:08,530
Speaker 2: subsegment by larger than $250 less than 250, 89% of

317
00:21:08,530 --> 00:21:13,050
Speaker 2: those AI dollars went into those large rounds, those kind

318
00:21:13,050 --> 00:21:17,650
Speaker 2: of super sized rounds, and that's in a very small

319
00:21:17,650 --> 00:21:21,260
Speaker 2: number of companies as well, so just about a dozen companies.

320
00:21:21,729 --> 00:21:25,640
Speaker 2: If you look at what's remaining in the venture capital

321
00:21:27,500 --> 00:21:32,739
Speaker 2: investment level, um, you still have a total deployment into

322
00:21:32,739 --> 00:21:38,020
Speaker 2: general tech, um, excluding those bigger AI rounds, that's actually

323
00:21:38,020 --> 00:21:42,959
Speaker 2: quite healthy, so it's still running at about $110 billion annually,

324
00:21:43,900 --> 00:21:46,000
Speaker 2: and what that means is that

325
00:21:46,439 --> 00:21:50,560
Speaker 2: The AI boom, so to speak, has not sucked the

326
00:21:50,560 --> 00:21:55,389
Speaker 2: air out of the room for other high growth tech companies. Well,

327
00:21:55,680 --> 00:22:00,399
Speaker 2: the entire ecosystem actually has been benefiting, and the investment

328
00:22:00,400 --> 00:22:04,718
Speaker 2: levels are remaining pretty healthy across the board. So I

329
00:22:04,719 --> 00:22:06,739
Speaker 2: think that although there's a concentration.

330
00:22:07,479 --> 00:22:10,680
Speaker 2: Of investment in a small number of companies and the

331
00:22:10,680 --> 00:22:16,709
Speaker 2: levels are unprecedented. It doesn't have that bubble characteristic of

332
00:22:17,010 --> 00:22:21,129
Speaker 2: um kind of um sucking all the air out of

333
00:22:21,130 --> 00:22:24,209
Speaker 2: the room for everybody else. So that's kind of a

334
00:22:24,209 --> 00:22:25,448
Speaker 2: healthy sign, I think.

335
00:22:26,739 --> 00:22:31,010
Speaker 2: Um, you know, I think the other area of analysis

336
00:22:31,010 --> 00:22:36,619
Speaker 2: around the bubble question is what percent of GDP, you know,

337
00:22:36,819 --> 00:22:41,459
Speaker 2: has the data center investment level gotten to, and although

338
00:22:41,459 --> 00:22:44,719
Speaker 2: it's definitely like the numbers are really large, you know,

339
00:22:45,060 --> 00:22:49,500
Speaker 2: $344 billion you know, to be invested this year by

340
00:22:49,500 --> 00:22:52,718
Speaker 2: the hyperscalers, you know, in new data center buildouts.

341
00:22:53,060 --> 00:22:57,239
Speaker 2: You know, probably about over $400 billion projected for next year.

342
00:22:57,619 --> 00:23:04,660
Speaker 2: You know, it's still only about, you know, 5-6% of

343
00:23:04,660 --> 00:23:08,219
Speaker 2: total capex, you know, across the economy. You know, it's

344
00:23:08,219 --> 00:23:10,939
Speaker 2: about 1.2% of GDP.

345
00:23:11,619 --> 00:23:15,389
Speaker 2: You know, it's growing very rapidly and that growth has aided,

346
00:23:15,709 --> 00:23:19,369
Speaker 2: you know, US GDP growth. So probably, and, you know,

347
00:23:19,800 --> 00:23:23,119
Speaker 2: for a variety of factors or reasons, you know, GDP

348
00:23:23,119 --> 00:23:26,099
Speaker 2: growth in the US has been slowing, you know, the

349
00:23:26,099 --> 00:23:29,479
Speaker 2: rise of interest rates, you know, the biggest reason, um,

350
00:23:29,839 --> 00:23:34,349
Speaker 2: you know, so the belief is that the AI investment wave,

351
00:23:34,599 --> 00:23:38,938
Speaker 2: or at least AI-driven investment wave in tech infrastructure, was

352
00:23:38,939 --> 00:23:40,859
Speaker 2: responsible for about 50% of growth.

353
00:23:41,310 --> 00:23:45,140
Speaker 2: So far in 2025 in the US, but that said,

354
00:23:45,430 --> 00:23:49,829
Speaker 2: it's still relatively small in the context of a very

355
00:23:49,829 --> 00:23:56,449
Speaker 2: large diversified economy, so it, it hasn't, you know, that, that, um, uh, the, the,

356
00:23:56,550 --> 00:23:57,000
Speaker 2: the

357
00:23:57,670 --> 00:24:02,750
Speaker 2: Dependence or concentration of spending that has happened in past

358
00:24:02,750 --> 00:24:07,390
Speaker 2: bubbles where a particular sector got to be 5 to 10%

359
00:24:07,390 --> 00:24:11,750
Speaker 2: of total GDP, you know, and then overinvestment led to

360
00:24:11,750 --> 00:24:15,739
Speaker 2: a retraction where it sank the entire economy, you know,

361
00:24:16,790 --> 00:24:19,149
Speaker 2: it doesn't look like we're anywhere near that kind of

362
00:24:19,150 --> 00:24:22,790
Speaker 2: territory today. You know, there is a lot of stock

363
00:24:22,790 --> 00:24:24,599
Speaker 2: market concentration.

364
00:24:25,089 --> 00:24:28,489
Speaker 2: Of value, you know, in, in the MG 7, I

365
00:24:28,489 --> 00:24:32,170
Speaker 2: guess they crossed about 35% or they're, they're hovering right

366
00:24:32,170 --> 00:24:35,689
Speaker 2: at like 35% today, you know, they crossed over it,

367
00:24:35,810 --> 00:24:37,609
Speaker 2: you know, a few weeks ago and then gave a

368
00:24:37,609 --> 00:24:41,208
Speaker 2: little back, um, you know, so that, that's certainly a

369
00:24:41,209 --> 00:24:45,770
Speaker 2: concern that if, you know, if, if the growth cycles,

370
00:24:46,050 --> 00:24:49,250
Speaker 2: you know, for the leaders in the, you know, the

371
00:24:49,250 --> 00:24:53,089
Speaker 2: high valued companies in the stock market, you know, start

372
00:24:53,089 --> 00:24:54,229
Speaker 2: to show signs of weakness.

373
00:24:54,900 --> 00:24:57,969
Speaker 2: You could see a retraction in the market and, you know,

374
00:24:58,339 --> 00:25:01,819
Speaker 2: you could see, um, you know, some wealth retraction there,

375
00:25:01,900 --> 00:25:07,899
Speaker 2: but I think it, it certainly hasn't happened, and you know,

376
00:25:07,979 --> 00:25:11,419
Speaker 2: these companies are, they're very, very profitable. They're charting at

377
00:25:11,420 --> 00:25:17,260
Speaker 2: relatively low PE to growth rate ratios, and the fundamentals

378
00:25:17,260 --> 00:25:20,900
Speaker 2: look pretty good. So, um, you know, I think it'd

379
00:25:20,900 --> 00:25:23,150
Speaker 2: be easy to kind of try to call it a bubble.

380
00:25:24,150 --> 00:25:26,589
Speaker 2: You probably get better headlines, you know, by calling it

381
00:25:26,589 --> 00:25:27,239
Speaker 2: a bubble.

382
00:25:27,589 --> 00:25:27,909
Speaker 1: Every

383
00:25:27,910 --> 00:25:28,290
Speaker 1: day,

384
00:25:29,150 --> 00:25:32,989
Speaker 2: yeah, I'd say the underlying factors don't, don't really point

385
00:25:32,989 --> 00:25:33,669
Speaker 2: in that direction.

386
00:25:34,589 --> 00:25:38,079
Speaker 1: Um, no, maybe since we're talking about headlines, might as

387
00:25:38,079 --> 00:25:40,079
Speaker 1: well talk about the two big headlines. One is the

388
00:25:40,079 --> 00:25:43,790
Speaker 1: circularity aspect that the big ones are buying things off

389
00:25:43,790 --> 00:25:47,060
Speaker 1: each other, and the second is the Mike Burry critique,

390
00:25:47,160 --> 00:25:51,359
Speaker 1: which is that the high-end chips have a very short

391
00:25:51,359 --> 00:25:55,438
Speaker 1: period of obsolescence and therefore they need to sort of

392
00:25:55,439 --> 00:26:00,670
Speaker 1: depreciate them much faster and that would create certain accounting dislocations.

393
00:26:01,030 --> 00:26:03,079
Speaker 1: Do you have any view on either of those issues?

394
00:26:05,290 --> 00:26:08,560
Speaker 2: Yeah, so I, um, take this, the uh sharing off

395
00:26:08,560 --> 00:26:09,000
Speaker 2: for now,

396
00:26:09,099 --> 00:26:09,448
Speaker 1: um.

397
00:26:10,400 --> 00:26:12,640
Speaker 2: I may come back to some of these little slides

398
00:26:12,640 --> 00:26:13,719
Speaker 2: as we talk about them.

399
00:26:14,329 --> 00:26:14,770
Speaker 2: Um,

400
00:26:16,260 --> 00:26:19,449
Speaker 2: Uh, yeah, I think, I think so the, um, to the,

401
00:26:19,699 --> 00:26:23,089
Speaker 2: to the second point about depreciation, um.

402
00:26:23,949 --> 00:26:26,790
Speaker 2: You know, it, it, so there are a couple of

403
00:26:26,790 --> 00:26:32,189
Speaker 2: factors there that, um, uh, I mean, yes, like it

404
00:26:32,189 --> 00:26:36,770
Speaker 2: would seem on the surface that because these

405
00:26:37,719 --> 00:26:40,939
Speaker 2: High performance GPUs, you know, the chips that are used

406
00:26:41,400 --> 00:26:45,469
Speaker 2: to drive AI data processing and, you know, um, they're being,

407
00:26:45,560 --> 00:26:50,199
Speaker 2: you know, purchased at incredible scale, and newer ones are

408
00:26:50,199 --> 00:26:54,280
Speaker 2: coming out every year, or, you know, um, they were with,

409
00:26:54,319 --> 00:26:56,438
Speaker 2: if you count the delays, been a little bit more

410
00:26:56,439 --> 00:26:59,260
Speaker 2: than a year, you know, since the last generation, so

411
00:26:59,260 --> 00:27:03,959
Speaker 2: maybe the timelines are extending a little bit for the generations. Um, it's,

412
00:27:04,040 --> 00:27:04,800
Speaker 2: it's a.

413
00:27:05,189 --> 00:27:09,069
Speaker 2: You know, and the demand tends to shift, you know,

414
00:27:09,270 --> 00:27:12,630
Speaker 2: toward the latest ones, although it takes time for the

415
00:27:12,630 --> 00:27:15,988
Speaker 2: newest generation to get to production scale and get to,

416
00:27:16,380 --> 00:27:21,750
Speaker 2: you know, delivery and deployment scale, that, you know, what's

417
00:27:21,750 --> 00:27:25,369
Speaker 2: the value of the last generation? It must be like 0,

418
00:27:25,630 --> 00:27:29,650
Speaker 2: but actually that's not the case because you can virtualize

419
00:27:29,979 --> 00:27:32,929
Speaker 2: the previous, the older chips.

420
00:27:33,390 --> 00:27:38,729
Speaker 2: And still deliver a very good service, um, and, um, a,

421
00:27:38,739 --> 00:27:42,229
Speaker 2: a pretty good performance level. Um, you really need the

422
00:27:42,229 --> 00:27:49,930
Speaker 2: latest generation only to do the, um, the kind of headline, um, top-ranked, um,

423
00:27:49,949 --> 00:27:55,280
Speaker 2: first-year model development, um, which, um, you know, is, is the,

424
00:27:55,459 --> 00:27:58,910
Speaker 2: like the newest, greatest models that are gonna get the

425
00:27:58,910 --> 00:28:01,478
Speaker 2: attention or kind of leading, you know, the marketing push.

426
00:28:01,839 --> 00:28:05,699
Speaker 2: You know, um, for, for alphabet and, you know, um,

427
00:28:06,040 --> 00:28:09,550
Speaker 2: Microsoft and Meta and OpenAI, everybody's trying to, you know,

428
00:28:09,719 --> 00:28:13,198
Speaker 2: showcase what they're doing at the, at the, um, at

429
00:28:13,199 --> 00:28:18,660
Speaker 2: the leading edge and certainly having the most performant chip

430
00:28:19,520 --> 00:28:22,319
Speaker 2: set gives you the best, you know, chance of delivering

431
00:28:22,319 --> 00:28:27,650
Speaker 2: something that's eye opening that really, you know, tops the benchmarks. However,

432
00:28:28,060 --> 00:28:33,698
Speaker 2: Um, each of these companies has many products, and a

433
00:28:33,699 --> 00:28:38,099
Speaker 2: lot of those products don't require that latest generation to

434
00:28:38,099 --> 00:28:41,540
Speaker 2: build or to serve to end users, so you can

435
00:28:41,540 --> 00:28:45,219
Speaker 2: use the older products, they work just as well, you know,

436
00:28:45,290 --> 00:28:48,939
Speaker 2: they're less efficient, but, you know, they still, they still

437
00:28:48,939 --> 00:28:52,819
Speaker 2: work fine, and over time you can add them to

438
00:28:52,819 --> 00:28:55,680
Speaker 2: a pool of resources that gets virtualized.

439
00:28:56,290 --> 00:28:59,640
Speaker 2: And you can still get pretty good value out of them.

440
00:29:00,380 --> 00:29:04,219
Speaker 2: You can, you charge less to the end customer or

441
00:29:04,219 --> 00:29:07,719
Speaker 2: you're using them for, you know, lower value use cases

442
00:29:07,979 --> 00:29:12,060
Speaker 2: if it's an internal, you know, calculation, but, but they're

443
00:29:12,060 --> 00:29:15,060
Speaker 2: still very useful. So I think there's, there's a little

444
00:29:15,060 --> 00:29:20,140
Speaker 2: bit of a misconception around what's that value changed from

445
00:29:20,140 --> 00:29:24,339
Speaker 2: the latest generation to the next, you know.

446
00:29:24,869 --> 00:29:30,140
Speaker 2: That some of the depreciation assumptions of 5 to 7 years,

447
00:29:30,150 --> 00:29:34,849
Speaker 2: depending on the company, they're based on that expectation that

448
00:29:35,030 --> 00:29:38,910
Speaker 2: the older generation will get shift to lower use cases

449
00:29:38,910 --> 00:29:42,310
Speaker 2: but will still, you know, have value. They actually have

450
00:29:42,310 --> 00:29:46,130
Speaker 2: a lot of experience with that with CPUs, so it's

451
00:29:46,510 --> 00:29:48,750
Speaker 2: really the same with the CPU, you know, when you

452
00:29:48,750 --> 00:29:52,930
Speaker 2: buy a new server, you put the latest gen CPU.

453
00:29:53,260 --> 00:29:56,369
Speaker 2: And it um and then you run it for a

454
00:29:56,369 --> 00:30:01,359
Speaker 2: long time, so you know as you expand, you're expanding

455
00:30:01,359 --> 00:30:05,939
Speaker 2: with the newest stuff you generally are not really replacing

456
00:30:06,300 --> 00:30:12,180
Speaker 2: older servers with the latest CPUs that frequently, so that,

457
00:30:12,260 --> 00:30:15,839
Speaker 2: you know, those assumptions aren't necessarily crazy, I think.

458
00:30:16,250 --> 00:30:19,069
Speaker 2: You know, they're they're rooted in some level of experience.

459
00:30:19,770 --> 00:30:21,849
Speaker 2: The other thing is that it's a, it's a, you know,

460
00:30:21,920 --> 00:30:26,089
Speaker 2: the depreciation is just a book value factor, and the

461
00:30:26,089 --> 00:30:30,250
Speaker 2: cash flows, you know, all this is factored into the

462
00:30:30,250 --> 00:30:33,130
Speaker 2: cash flow. So when you look at the EBITDA performance

463
00:30:33,130 --> 00:30:38,530
Speaker 2: of the companies, um, you know, those gap book value assumptions.

464
00:30:38,959 --> 00:30:43,160
Speaker 2: Aren't, you know, aren't subtracted out, so, um, you know,

465
00:30:43,239 --> 00:30:47,170
Speaker 2: the cash flows have been good for these businesses. So,

466
00:30:47,479 --> 00:30:50,520
Speaker 2: you know, uh, you know, they're, they're, and they're, they're

467
00:30:50,520 --> 00:30:53,719
Speaker 2: in many cases they have, you know, customers, you know,

468
00:30:53,839 --> 00:30:56,359
Speaker 2: on the hook waiting to pay for these services and

469
00:30:56,359 --> 00:30:59,439
Speaker 2: make these big cash commitments, you know, and contract, you know,

470
00:30:59,729 --> 00:31:03,760
Speaker 2: contracting commitments, um, and that's really how it's being priced, so.

471
00:31:04,219 --> 00:31:10,520
Speaker 2: Uh, the, the, the book value issue, I don't think it's, um, uh, it's, it's, it's,

472
00:31:10,550 --> 00:31:16,640
Speaker 2: it's not like you're creating a, you know, um, a CDO,

473
00:31:16,819 --> 00:31:21,020
Speaker 2: you know, out of mortgages, right, where you're, you know,

474
00:31:21,260 --> 00:31:24,900
Speaker 2: slicing and dicing it and making a zillion assumptions, and then,

475
00:31:25,140 --> 00:31:27,180
Speaker 2: you know, it has to have, you know, value that's

476
00:31:27,180 --> 00:31:29,739
Speaker 2: going to perform for 30 years. It's not like that

477
00:31:29,739 --> 00:31:33,680
Speaker 2: at all, you know, it's actually very like.

478
00:31:34,010 --> 00:31:37,670
Speaker 2: Open book, very clear what the cash flows are, um,

479
00:31:37,780 --> 00:31:42,530
Speaker 2: and these depreciation assumptions, um, are relatively, I think they're

480
00:31:42,530 --> 00:31:46,189
Speaker 2: relatively minor in the scheme of the valuations of the companies.

481
00:31:46,930 --> 00:31:48,959
Speaker 1: All right, very good and very useful. As long as

482
00:31:48,959 --> 00:31:52,520
Speaker 1: I suppose the inner operability part and the backward compatibility

483
00:31:52,520 --> 00:31:56,050
Speaker 1: part remains intact, you can certainly rely on your older

484
00:31:56,050 --> 00:32:00,050
Speaker 1: chips for much longer than what Mike Burry is suggesting. Uh,

485
00:32:00,160 --> 00:32:02,689
Speaker 1: what about that first point, the circularity aspect?

486
00:32:04,219 --> 00:32:07,699
Speaker 2: But yeah, I think the circularity aspect, it's been, it's

487
00:32:07,699 --> 00:32:11,579
Speaker 2: been growing, which I think is a concern, it has

488
00:32:11,579 --> 00:32:16,300
Speaker 2: not been that significant to date in terms of driving

489
00:32:16,300 --> 00:32:21,180
Speaker 2: the revenues for Nvidia in particular, you know, you know,

490
00:32:21,260 --> 00:32:26,560
Speaker 2: to some extent the investments that Microsoft have made.

491
00:32:26,930 --> 00:32:30,050
Speaker 2: You know, Microsoft made an OpenAI and the, you know,

492
00:32:30,290 --> 00:32:34,810
Speaker 2: Alphabet made in anthropic, you know, and Amazon made in anthropic,

493
00:32:35,130 --> 00:32:38,689
Speaker 2: you know, had circularity aspects to them that supported their

494
00:32:38,689 --> 00:32:44,670
Speaker 2: cloud businesses, but those cloud businesses are so enormous.

495
00:32:45,150 --> 00:32:49,410
Speaker 2: Today and have such huge diversified customer bases that the

496
00:32:49,410 --> 00:32:54,060
Speaker 2: revenue impact, you know, to Alphabet and Amazon and Microsoft

497
00:32:54,329 --> 00:32:57,930
Speaker 2: is really quite minor. It's maybe a few percent, you know,

498
00:32:58,050 --> 00:33:02,010
Speaker 2: that gets that gets recycled or circled back, you know,

499
00:33:02,089 --> 00:33:07,209
Speaker 2: and spent. So, you know, it's really not, it's, it's,

500
00:33:07,280 --> 00:33:11,810
Speaker 2: it's not really that significant. It, it's also something that

501
00:33:11,810 --> 00:33:13,750
Speaker 2: historically those

502
00:33:14,140 --> 00:33:17,260
Speaker 2: The hyper scalers were not the lead investors, so the

503
00:33:17,260 --> 00:33:22,180
Speaker 2: rounds would come together, driven by financial investors, would get

504
00:33:22,180 --> 00:33:26,459
Speaker 2: priced and mostly filled out, and then the strategics would

505
00:33:26,459 --> 00:33:30,699
Speaker 2: come in and and top it off, you know, add

506
00:33:30,699 --> 00:33:34,140
Speaker 2: some additional capital, and that's very much how Nvidia has

507
00:33:34,140 --> 00:33:38,280
Speaker 2: been treating these funding rounds. So they generally have not

508
00:33:38,280 --> 00:33:40,699
Speaker 2: been the lead or the largest source of the capital.

509
00:33:40,979 --> 00:33:42,319
Speaker 2: I think if that were to flip.

510
00:33:42,770 --> 00:33:45,849
Speaker 2: And you were to start to see the hyper scalers

511
00:33:45,849 --> 00:33:49,569
Speaker 2: be the primary source of the capital into the companies

512
00:33:49,569 --> 00:33:52,000
Speaker 2: that then needed to turn around and spend it with them,

513
00:33:52,319 --> 00:33:55,770
Speaker 2: you know, then, then I think it would be reason

514
00:33:55,770 --> 00:34:01,310
Speaker 2: for concern, but it hasn't been the case, so I think,

515
00:34:01,410 --> 00:34:04,890
Speaker 2: I think it's not, it's definitely an area to keep

516
00:34:04,890 --> 00:34:05,540
Speaker 2: an eye on.

517
00:34:06,079 --> 00:34:08,729
Speaker 2: And you know, over time it could be a concern

518
00:34:08,729 --> 00:34:12,689
Speaker 2: if it grows, but I don't think historically it's been

519
00:34:12,689 --> 00:34:15,610
Speaker 2: a pattern. There's, there's not really a pattern there that

520
00:34:15,610 --> 00:34:21,290
Speaker 2: would indicate fragility or, you know, an overdependence on that

521
00:34:21,290 --> 00:34:23,770
Speaker 2: circular funding for the ecosystem.

522
00:34:25,259 --> 00:34:28,618
Speaker 1: Noah, you and I began our careers under the shadow

523
00:34:28,618 --> 00:34:31,098
Speaker 1: of the dot com boom burst. Uh, one of the

524
00:34:31,099 --> 00:34:34,299
Speaker 1: things that we saw there was that the original infrastructure

525
00:34:34,299 --> 00:34:37,458
Speaker 1: companies that lay the pipes for the internet, you know,

526
00:34:37,539 --> 00:34:40,019
Speaker 1: we're not the big profitable ones. The ones who build

527
00:34:40,018 --> 00:34:41,498
Speaker 1: applications and on top of that are the ones who

528
00:34:41,498 --> 00:34:45,178
Speaker 1: really hit it out of the ballpark. I guess my

529
00:34:45,178 --> 00:34:48,258
Speaker 1: question is trying to draw a parallel between then and now,

530
00:34:48,309 --> 00:34:53,328
Speaker 1: which is, are LLMs becoming ubiquitous or from your vantage point,

531
00:34:53,378 --> 00:34:54,979
Speaker 1: you see sufficient differentiation?

532
00:34:55,560 --> 00:34:58,010
Speaker 1: That all the big ones that take billions of dollars

533
00:34:58,010 --> 00:35:02,239
Speaker 1: to build out and scale can operate next to each other.

534
00:35:02,330 --> 00:35:05,529
Speaker 1: It doesn't have to be winner take all or winner take,

535
00:35:05,840 --> 00:35:06,689
Speaker 1: you know, most.

536
00:35:08,810 --> 00:35:13,969
Speaker 2: Great, yeah, great, great question. Um, I, I, I think, yes, there's,

537
00:35:14,050 --> 00:35:19,790
Speaker 2: there's a, a very significant difference in between dark fiber

538
00:35:20,449 --> 00:35:25,350
Speaker 2: and LLMs in the sense that LLMs, um, and, and

539
00:35:25,360 --> 00:35:30,080
Speaker 2: really AI technology, you know, more broadly has been

540
00:35:30,760 --> 00:35:36,590
Speaker 2: Just, you know, it's, it's anywhere there's knowledge creation, knowledge management,

541
00:35:36,989 --> 00:35:41,830
Speaker 2: knowledge and manipulation, information manipulation, AI has an opportunity to

542
00:35:41,830 --> 00:35:46,709
Speaker 2: make people more efficient, make businesses more efficient, so it's,

543
00:35:46,790 --> 00:35:51,850
Speaker 2: it's in some sense a much vaster market opportunity than

544
00:35:52,070 --> 00:35:55,070
Speaker 2: just the, you know, the opportunity to move bits around

545
00:35:55,070 --> 00:35:57,060
Speaker 2: through a pipe, um, and

546
00:35:57,429 --> 00:36:03,149
Speaker 2: Yeah, that, yes, the telecom industry like vastly over overestimated,

547
00:36:03,229 --> 00:36:07,850
Speaker 2: you know, what the demand would be, you know, and

548
00:36:08,060 --> 00:36:10,770
Speaker 2: there was certainly, you know, a

549
00:36:11,399 --> 00:36:15,229
Speaker 2: Insane collapse, you know, a drastic collapse, you know, um,

550
00:36:15,679 --> 00:36:19,399
Speaker 2: but there also wasn't as much opportunity for differentiation there,

551
00:36:19,520 --> 00:36:23,639
Speaker 2: you know, it, it turned out that connectivity, you know,

552
00:36:24,040 --> 00:36:27,638
Speaker 2: became a commodity and, as, as, you know, to some

553
00:36:27,639 --> 00:36:32,639
Speaker 2: extent was predicted, and, uh, you know, the opportunities to

554
00:36:32,639 --> 00:36:36,719
Speaker 2: create value add, you know, um, were never really realized,

555
00:36:37,000 --> 00:36:39,979
Speaker 2: you know, despite everyone's, you know, all the telecom players.

556
00:36:40,469 --> 00:36:43,830
Speaker 2: Talked about value added services and really worked hard to

557
00:36:43,830 --> 00:36:46,969
Speaker 2: do it, but in the end, you know, it was

558
00:36:46,969 --> 00:36:47,409
Speaker 2: really

559
00:36:49,010 --> 00:36:54,699
Speaker 2: Connectivity was a commodity with AI, it's really the opposite

560
00:36:54,699 --> 00:37:02,839
Speaker 2: because the ability to utilize these products in different contexts

561
00:37:03,739 --> 00:37:08,979
Speaker 2: is variegating, you know, very rapidly. You're seeing a number

562
00:37:08,979 --> 00:37:12,449
Speaker 2: of products that are much better for code development. You're

563
00:37:12,449 --> 00:37:18,300
Speaker 2: seeing areas of specialization around science and

564
00:37:19,060 --> 00:37:23,629
Speaker 2: Research in life science and physical world, you're seeing a

565
00:37:23,629 --> 00:37:28,489
Speaker 2: set of companies, including the company that spun out of

566
00:37:28,909 --> 00:37:33,469
Speaker 2: our investment in Nanic is when Nanic sold its gaming

567
00:37:33,469 --> 00:37:37,229
Speaker 2: business to the PIF-backed.

568
00:37:37,669 --> 00:37:42,709
Speaker 2: A company Scopey, they spun out the platform business that

569
00:37:42,709 --> 00:37:47,189
Speaker 2: had enabled Pokemon Go to operate called Niantic Spatial, and

570
00:37:47,189 --> 00:37:51,229
Speaker 2: they're building a real-world, real-world model that will enable a

571
00:37:51,229 --> 00:37:55,350
Speaker 2: visual GPS. So without a GPS sensor, you'll be able

572
00:37:55,350 --> 00:37:58,939
Speaker 2: to point your phone anywhere you are and get an

573
00:37:58,939 --> 00:38:06,370
Speaker 2: extremely precise geolocation and be able to create real world experiences.

574
00:38:06,810 --> 00:38:13,459
Speaker 2: Using highly, highly detailed geospatial information via navigation or logistics

575
00:38:13,459 --> 00:38:19,689
Speaker 2: or experiences like gaming, you know, that's a, that's an

576
00:38:19,689 --> 00:38:26,340
Speaker 2: AI enabled and AI powered set of capabilities that are, um,

577
00:38:26,489 --> 00:38:29,320
Speaker 2: you know, moving into the physical world and, you know,

578
00:38:29,810 --> 00:38:32,370
Speaker 2: have all these new use cases.

579
00:38:32,800 --> 00:38:36,159
Speaker 2: That just weren't possible before. So I think, I think

580
00:38:36,159 --> 00:38:38,280
Speaker 2: it's a very different kind of thing, you know, it's something,

581
00:38:38,439 --> 00:38:41,800
Speaker 2: it's something AI will, you know, already is and, you know,

582
00:38:41,879 --> 00:38:47,239
Speaker 2: increasingly will impact everyone in their own life and career

583
00:38:47,239 --> 00:38:51,399
Speaker 2: and how they live, and it's, it's, it's hard to

584
00:38:51,399 --> 00:38:56,429
Speaker 2: really pinpoint the potential for value there, but it's really vast. It's,

585
00:38:56,479 --> 00:38:59,560
Speaker 2: it's much bigger than any, any one industry.

586
00:39:01,030 --> 00:39:04,310
Speaker 1: Excellent. Ah, OK. I want to ask you something that

587
00:39:04,310 --> 00:39:06,310
Speaker 1: you and I discussed while we're having a cup of

588
00:39:06,310 --> 00:39:08,250
Speaker 1: coffee a few weeks ago in Singapore.

589
00:39:09,110 --> 00:39:11,780
Speaker 1: Is the path to AGI through LLMs?

590
00:39:15,030 --> 00:39:21,750
Speaker 2: Yes, great question. It's, it's an increasingly hot topic, as,

591
00:39:22,149 --> 00:39:28,000
Speaker 2: you know, I think the biggest AI players are making

592
00:39:28,000 --> 00:39:33,310
Speaker 2: big bets. The LLMs are the path, and there's, there's

593
00:39:33,310 --> 00:39:37,209
Speaker 2: been a, uh, I would describe it as an undercurrent

594
00:39:37,429 --> 00:39:42,750
Speaker 2: of doubt or at least suspicion that there's an alternate path.

595
00:39:43,149 --> 00:39:47,580
Speaker 2: It's rooted in what has been described as the the

596
00:39:48,610 --> 00:39:49,800
Speaker 2: the the

597
00:39:50,629 --> 00:39:54,870
Speaker 2: Paradigm of neurosymbolic or symbolic AI.

598
00:39:55,560 --> 00:40:03,489
Speaker 2: Intersecting with neuro-enabled AI where neuro is the deep learning

599
00:40:03,679 --> 00:40:11,080
Speaker 2: neural net-based LLM model that's very tech-centric and the kind

600
00:40:11,080 --> 00:40:15,279
Speaker 2: of fundamental, since the real, you know, that technology really

601
00:40:15,280 --> 00:40:20,520
Speaker 2: broke out and has shown enormous promise, many of the

602
00:40:20,520 --> 00:40:24,479
Speaker 2: researchers and the engineering engineers driving it have come to

603
00:40:24,479 --> 00:40:25,139
Speaker 2: believe that

604
00:40:25,590 --> 00:40:30,469
Speaker 2: If you throw enough data at these deep learning neural networks,

605
00:40:30,820 --> 00:40:34,149
Speaker 2: they can start to teach themselves, and eventually you'll kind

606
00:40:34,149 --> 00:40:35,159
Speaker 2: of recreate.

607
00:40:36,070 --> 00:40:38,899
Speaker 2: Knowledge, you know, they will be able to recreate knowledge,

608
00:40:39,030 --> 00:40:41,310
Speaker 2: you know, on their own independently, and it's really a

609
00:40:41,310 --> 00:40:45,429
Speaker 2: matter of getting to the scale needed, um, to, uh,

610
00:40:45,590 --> 00:40:48,629
Speaker 2: to be able to enable them to start to mimic

611
00:40:48,629 --> 00:40:49,570
Speaker 2: the human brain.

612
00:40:49,889 --> 00:40:53,810
Speaker 2: Um, and it's a little bit of a belief, you know,

613
00:40:53,929 --> 00:40:58,489
Speaker 2: nobody really knows if that's possible. It looked like there

614
00:40:58,489 --> 00:41:02,529
Speaker 2: were some signs that these systems were like teaching themselves

615
00:41:02,530 --> 00:41:05,569
Speaker 2: to do math and things like that. Those have not

616
00:41:05,570 --> 00:41:12,399
Speaker 2: really played out actually, and I think this, this undercurrent

617
00:41:12,689 --> 00:41:15,409
Speaker 2: of different perspective.

618
00:41:15,830 --> 00:41:21,549
Speaker 2: Around the intersection of symbolic AI techniques with the neuro

619
00:41:21,550 --> 00:41:26,469
Speaker 2: side has shown a lot more promise. We recently made

620
00:41:26,469 --> 00:41:30,790
Speaker 2: an investment in a stealth company that's focused on this

621
00:41:30,790 --> 00:41:36,969
Speaker 2: particular area, and they have developed some early proof points.

622
00:41:37,320 --> 00:41:43,469
Speaker 2: Of using logical reasoning and a true understanding of language

623
00:41:44,370 --> 00:41:49,129
Speaker 2: and the logical relationships inside language to show 10x speed

624
00:41:49,129 --> 00:41:53,629
Speaker 2: ups in processing time and training time for AI models

625
00:41:53,909 --> 00:41:56,850
Speaker 2: and show the ability to solve problems that have stumped

626
00:41:56,850 --> 00:42:01,449
Speaker 2: the LLMs to date. So we really, we, we really

627
00:42:01,449 --> 00:42:04,510
Speaker 2: on the investment side see it as a huge opportunity,

628
00:42:05,169 --> 00:42:06,709
Speaker 2: but I think in a broader sense.

629
00:42:07,000 --> 00:42:10,139
Speaker 2: Um, you know, it, it, if you think about it

630
00:42:10,399 --> 00:42:15,070
Speaker 2: at the philosophical level, you know, um, if, if there's

631
00:42:15,070 --> 00:42:19,600
Speaker 2: all this, you know, established knowledge, you know, we have mathematics,

632
00:42:19,709 --> 00:42:23,439
Speaker 2: we have physics, um, you know, we have, um, we

633
00:42:23,439 --> 00:42:26,600
Speaker 2: have logic, we have, um, you know, uh,

634
00:42:26,979 --> 00:42:31,149
Speaker 2: Um, uh, we have very structured ways of thinking about things,

635
00:42:31,229 --> 00:42:35,379
Speaker 2: you know, we have economics. Uh, well, maybe economics is,

636
00:42:35,459 --> 00:42:38,330
Speaker 2: is harder, but, you know, so there's a lot.

637
00:42:39,669 --> 00:42:42,689
Speaker 2: Still the dismal science, but yeah, there, there are a

638
00:42:42,689 --> 00:42:45,330
Speaker 2: lot of, a lot of arenas where, you know, there's

639
00:42:45,330 --> 00:42:50,699
Speaker 2: genuine knowledge that has deterministic relationships, and do we really

640
00:42:50,699 --> 00:42:55,540
Speaker 2: need to um uh take computers through the process of

641
00:42:55,540 --> 00:42:59,629
Speaker 2: teaching themselves all this knowledge without benefiting from it, you know,

642
00:42:59,899 --> 00:43:04,090
Speaker 2: Shouldn't they be benefiting from it? Um, and in that case,

643
00:43:04,219 --> 00:43:06,719
Speaker 2: can you, you know, ingest the textbook.

644
00:43:07,600 --> 00:43:12,279
Speaker 2: You know, ingest ingest the the mathematics, um, and then

645
00:43:12,280 --> 00:43:16,459
Speaker 2: apply it when it's relevant and not need to play

646
00:43:16,560 --> 00:43:21,500
Speaker 2: the guessing game, you know, of a massive statistical probabilistic, um,

647
00:43:21,959 --> 00:43:24,239
Speaker 2: you know, extremely large model.

648
00:43:25,810 --> 00:43:29,209
Speaker 1: So, on one hand, you're talking about the promise of

649
00:43:29,209 --> 00:43:34,529
Speaker 1: thinking through not the neural network type architecture that this

650
00:43:34,530 --> 00:43:38,609
Speaker 1: LLMs try to sort of foundationally build. And, and, and

651
00:43:38,610 --> 00:43:40,330
Speaker 1: the very last thing that you said, which is, I

652
00:43:40,330 --> 00:43:42,669
Speaker 1: suppose the essence of an LLM that is a probabilistic

653
00:43:42,669 --> 00:43:45,370
Speaker 1: model having ingested all the words and all the images

654
00:43:45,370 --> 00:43:47,669
Speaker 1: that it can and trying to organize them.

655
00:43:48,159 --> 00:43:51,399
Speaker 1: So, your key point is that there might be a

656
00:43:51,399 --> 00:43:56,840
Speaker 1: limitation to intelligent development out of that LLM framework.

657
00:43:59,340 --> 00:44:02,129
Speaker 2: What, well, I think we're already, we're already seeing that,

658
00:44:02,340 --> 00:44:06,820
Speaker 2: you know, um, the, the scale of training, you know,

659
00:44:06,979 --> 00:44:11,949
Speaker 2: that these models are taken through has been increasing dramatically, um,

660
00:44:12,060 --> 00:44:16,620
Speaker 2: but they're just inching forward on the benchmarks, um, you know,

661
00:44:16,750 --> 00:44:21,080
Speaker 2: by 1% or 2%, and they're still performing very poorly

662
00:44:21,179 --> 00:44:22,479
Speaker 2: at some very basic.

663
00:44:23,360 --> 00:44:27,839
Speaker 2: Tests which, you know, frankly, a 5 year old can pass,

664
00:44:27,959 --> 00:44:33,729
Speaker 2: you know, so there's something missing, you know, and no one,

665
00:44:33,800 --> 00:44:37,979
Speaker 2: of course, knows what that is, or, you know, we wouldn't,

666
00:44:38,199 --> 00:44:40,699
Speaker 2: wouldn't be here speculating about it, but

667
00:44:41,629 --> 00:44:44,409
Speaker 2: You know, I think there are quite a few, um,

668
00:44:44,669 --> 00:44:49,709
Speaker 2: certainly researchers out there who are come from that's either

669
00:44:49,709 --> 00:44:53,669
Speaker 2: come from the symbolic tradition or starting to look out

670
00:44:53,669 --> 00:44:58,709
Speaker 2: at some of the techniques used in more traditional machine

671
00:44:58,709 --> 00:45:08,189
Speaker 2: learning and expert system style rules-based logic-driven AI systems, AI techniques.

672
00:45:08,449 --> 00:45:12,570
Speaker 2: And saying how do we incorporate that into these models

673
00:45:12,570 --> 00:45:16,209
Speaker 2: to make them smarter, make them more efficient, enable them

674
00:45:16,209 --> 00:45:19,928
Speaker 2: to solve problems, truly solve problems as opposed to just

675
00:45:19,929 --> 00:45:23,060
Speaker 2: guess at the answers, and I think that's, that's a

676
00:45:23,060 --> 00:45:27,129
Speaker 2: very promising area, and I think it's, it's, you know, broadly.

677
00:45:27,479 --> 00:45:31,659
Speaker 2: You know, it's good for society because it means that

678
00:45:31,919 --> 00:45:36,520
Speaker 2: we won't necessarily need, you know, 10% of our energy

679
00:45:36,520 --> 00:45:42,060
Speaker 2: devoted to AI, you know, processing, training, you know, or inference, right,

680
00:45:42,840 --> 00:45:43,359
Speaker 2: in the future.

681
00:45:44,050 --> 00:45:46,449
Speaker 2: There'll be more efficient ways of doing things, of getting

682
00:45:46,449 --> 00:45:51,089
Speaker 2: to answers, and we'll be able to solve the hallucination problem,

683
00:45:51,179 --> 00:45:56,300
Speaker 2: you know, um, with systems that are more deterministic in nature, uh,

684
00:45:56,530 --> 00:46:00,830
Speaker 2: where there's a question that really has an answer, we should, the,

685
00:46:00,929 --> 00:46:04,250
Speaker 2: the product should be delivering the answer, you know, it

686
00:46:04,250 --> 00:46:08,209
Speaker 2: shouldn't be making, it shouldn't be playing the guessing game

687
00:46:08,209 --> 00:46:09,389
Speaker 2: and getting it wrong, you know,

688
00:46:10,040 --> 00:46:10,050
Speaker 1: right.

689
00:46:10,760 --> 00:46:14,439
Speaker 1: Uh, absolutely fascinating. I'm sure listeners will start looking up

690
00:46:14,439 --> 00:46:17,799
Speaker 1: the alternatives to LLM stuff that you're talking about. Uh, Noah,

691
00:46:18,000 --> 00:46:20,679
Speaker 1: we have managed to have this detailed discussion without mentioning

692
00:46:20,679 --> 00:46:21,939
Speaker 1: the word China once.

693
00:46:22,879 --> 00:46:26,520
Speaker 1: Uh, I was looking at some trade publication the other

694
00:46:26,520 --> 00:46:28,870
Speaker 1: day after the Gemini 3 model came out. I was

695
00:46:28,870 --> 00:46:31,419
Speaker 1: looking at the ranking of the top LLMs in the world.

696
00:46:31,879 --> 00:46:34,979
Speaker 1: And of course, now Gemini 3 gets ranked number 1,

697
00:46:34,989 --> 00:46:36,560
Speaker 1: and then a bunch of open AI stuff comes in.

698
00:46:36,679 --> 00:46:39,049
Speaker 1: But when I look at the top 20, uh, Noah,

699
00:46:39,120 --> 00:46:42,600
Speaker 1: I'm just gonna read you some, uh, model. I actually

700
00:46:42,600 --> 00:46:45,040
Speaker 1: never heard of these things. Uh, number 5 in the

701
00:46:45,040 --> 00:46:48,520
Speaker 1: world is Kimmi K2, which is a Chinese company called

702
00:46:48,520 --> 00:46:49,580
Speaker 1: Moonshot AI.

703
00:46:49,979 --> 00:46:52,780
Speaker 1: In the top 10 ranking, I also saw something called Minimax,

704
00:46:52,860 --> 00:46:55,859
Speaker 1: also from China. And this is not even including the

705
00:46:55,860 --> 00:46:57,899
Speaker 1: deep seeks of the world and the LLMs that are

706
00:46:57,899 --> 00:47:01,179
Speaker 1: the Alibaba's and ByteDance is developing. So, tell me a

707
00:47:01,179 --> 00:47:04,340
Speaker 1: little bit about your perspective on the China tech stack.

708
00:47:04,689 --> 00:47:06,459
Speaker 1: It takes two questions. One is your assessment of what

709
00:47:06,459 --> 00:47:08,929
Speaker 1: they're doing? Are they doing it cheaper, better, or is

710
00:47:08,929 --> 00:47:11,839
Speaker 1: it a, you know, whole new paradigm? And the second is,

711
00:47:12,300 --> 00:47:14,219
Speaker 1: are we really heading into a world where we have

712
00:47:14,219 --> 00:47:16,850
Speaker 1: two different tech stacks, made in China and made in

713
00:47:16,850 --> 00:47:17,379
Speaker 1: the US?

714
00:47:19,810 --> 00:47:22,779
Speaker 2: Yeah, yeah, great question, and, you know, it's, it's, it's

715
00:47:22,780 --> 00:47:27,219
Speaker 2: relevant for everyone in this field. Um, I, my, my

716
00:47:27,679 --> 00:47:30,479
Speaker 2: broader point of view on this question is that we,

717
00:47:30,560 --> 00:47:34,500
Speaker 2: we're probably not headed toward two different tech stacks because

718
00:47:34,830 --> 00:47:38,479
Speaker 2: I think that the, the Chinese really can't take a

719
00:47:38,479 --> 00:47:43,799
Speaker 2: risk of falling behind, and I think that they, um, they,

720
00:47:43,840 --> 00:47:48,888
Speaker 2: they need to, for the most part, um follow practices

721
00:47:48,889 --> 00:47:49,399
Speaker 2: and

722
00:47:49,790 --> 00:47:53,989
Speaker 2: Use the infrastructure and use the tools which run best

723
00:47:53,989 --> 00:47:57,989
Speaker 2: on the hardware, you know, that comes from the mainstream

724
00:47:58,350 --> 00:48:02,510
Speaker 2: companies and from the top models, you know, and I

725
00:48:02,510 --> 00:48:06,870
Speaker 2: think they've done that really, really well and have demonstrated

726
00:48:06,870 --> 00:48:10,428
Speaker 2: that they can be very scrappy and they can find

727
00:48:10,429 --> 00:48:14,629
Speaker 2: efficiencies and they can do great work and make, you know,

728
00:48:14,739 --> 00:48:16,929
Speaker 2: real contributions, you know, to the space.

729
00:48:17,909 --> 00:48:22,870
Speaker 2: Ah, you know, there, there are many of those Chinese researchers,

730
00:48:22,989 --> 00:48:25,790
Speaker 2: you know, based in China were trained in the US,

731
00:48:25,949 --> 00:48:29,909
Speaker 2: you know, and, um, uh, you know, it, it's actually,

732
00:48:30,100 --> 00:48:32,779
Speaker 2: you know, um, if you look at the field globally,

733
00:48:33,149 --> 00:48:36,229
Speaker 2: you know, it's a very international, you know, field in

734
00:48:36,229 --> 00:48:39,428
Speaker 2: terms of who's doing the work, um, and who's doing

735
00:48:39,429 --> 00:48:40,149
Speaker 2: the development.

736
00:48:41,199 --> 00:48:46,760
Speaker 2: So I don't, I don't really see it bifurcating, at

737
00:48:46,760 --> 00:48:50,620
Speaker 2: least in the near term, um, you know, especially while

738
00:48:51,040 --> 00:48:54,280
Speaker 2: there is so much, such a fast pace of development

739
00:48:54,280 --> 00:48:58,759
Speaker 2: and so many innovations happening so quickly, no one can

740
00:48:58,760 --> 00:49:01,459
Speaker 2: really afford to take a chance, you know, on a secondary,

741
00:49:01,639 --> 00:49:05,479
Speaker 2: you know, or, you know, a different attempt.

742
00:49:05,979 --> 00:49:09,739
Speaker 2: Um, at creating a tool set that, you know, just

743
00:49:09,739 --> 00:49:10,759
Speaker 2: may not be as good.

744
00:49:11,750 --> 00:49:14,810
Speaker 2: Um, uh, you know, I do, I do think that, I,

745
00:49:15,129 --> 00:49:20,729
Speaker 2: you know, China, um, you know, and, um, the Chinese tech.

746
00:49:21,350 --> 00:49:27,110
Speaker 2: Um, you know, uh, community was without question always going

747
00:49:27,110 --> 00:49:32,979
Speaker 2: to be very important in the AI world and AI industry. Um,

748
00:49:34,010 --> 00:49:37,229
Speaker 2: there was, there was, you know, um, and they've demonstrated

749
00:49:37,229 --> 00:49:40,229
Speaker 2: that they can stay very, very close to the leading

750
00:49:40,229 --> 00:49:43,689
Speaker 2: edge and as you point out, you know, actually, you know,

751
00:49:44,389 --> 00:49:47,229
Speaker 2: of late, developed very competitive.

752
00:49:47,760 --> 00:49:50,040
Speaker 2: Models that are very low cost to run, you know,

753
00:49:50,199 --> 00:49:56,790
Speaker 2: so a lot of, especially consumer products that you know,

754
00:49:56,919 --> 00:50:02,199
Speaker 2: are trying to service large numbers of users inexpensively are

755
00:50:02,199 --> 00:50:05,280
Speaker 2: dependent on the Chinese models, and they're working really well.

756
00:50:06,280 --> 00:50:09,408
Speaker 2: Um, so, you know, I think it's, it's, you know,

757
00:50:09,600 --> 00:50:13,489
Speaker 2: it's a good thing to keep people, you know, keep

758
00:50:13,489 --> 00:50:18,569
Speaker 2: competition high. It keeps everybody honest. It keeps driving the

759
00:50:18,570 --> 00:50:23,330
Speaker 2: industry forward, um, you know, uh, it, it, it, and I,

760
00:50:23,379 --> 00:50:26,229
Speaker 2: and I think it's, uh, you know, the, the, the

761
00:50:26,370 --> 00:50:31,489
Speaker 2: sort of questions for strategic implications, um, at least, at

762
00:50:31,489 --> 00:50:34,209
Speaker 2: least today, you know, in the LLM space.

763
00:50:34,570 --> 00:50:39,179
Speaker 2: They're not really that significant. I think there's sort of

764
00:50:39,179 --> 00:50:42,570
Speaker 2: a many people draw a line between having the world's

765
00:50:42,570 --> 00:50:47,469
Speaker 2: best LLM and AGI and the ability to, you know,

766
00:50:48,530 --> 00:50:53,370
Speaker 2: develop some sort of, you know, advancement on the strategic side,

767
00:50:53,469 --> 00:50:59,370
Speaker 2: but other than maybe, you know, cybersecurity where, um, you know,

768
00:50:59,610 --> 00:51:01,469
Speaker 2: espionage or

769
00:51:01,850 --> 00:51:08,189
Speaker 2: you know, a sort of disinformation, you know, kind of stratagem,

770
00:51:08,520 --> 00:51:12,000
Speaker 2: you know, could be impactful at the strategic scale. You

771
00:51:12,000 --> 00:51:16,179
Speaker 2: don't really have a direct relationship between the LLMs and

772
00:51:17,709 --> 00:51:22,000
Speaker 2: strategic competition. It's much more really market competition as things

773
00:51:22,000 --> 00:51:22,819
Speaker 2: stand today.

774
00:51:23,709 --> 00:51:26,239
Speaker 1: OK, I'm glad you touched upon the cybersecurity issue. The

775
00:51:26,239 --> 00:51:31,799
Speaker 1: additional strategic question that I had was um ethics of AI. Um,

776
00:51:32,000 --> 00:51:34,919
Speaker 1: you know, we basically went through, you know, 15 years

777
00:51:34,919 --> 00:51:38,379
Speaker 1: of social media revolution without effectively putting guardrails around.

778
00:51:38,669 --> 00:51:41,689
Speaker 1: Social media and the cost has been the mental health

779
00:51:41,689 --> 00:51:45,540
Speaker 1: of our youth and today I'm now beginning to see

780
00:51:45,540 --> 00:51:49,580
Speaker 1: more and more articles, particularly related to meta and its

781
00:51:49,580 --> 00:51:53,610
Speaker 1: lama models that are small but are not a, you know,

782
00:51:53,699 --> 00:51:58,100
Speaker 1: insignificant portion of, again, young adults exposed to these LLMs

783
00:51:58,100 --> 00:51:59,799
Speaker 1: tend to have certain

784
00:51:59,969 --> 00:52:02,649
Speaker 1: Delusional experiences and so on. Actually, not just the youth.

785
00:52:02,729 --> 00:52:05,409
Speaker 1: I've seen stories about the, you know, much more mature

786
00:52:05,409 --> 00:52:09,129
Speaker 1: ones as well. So the Trump administration is very light

787
00:52:09,129 --> 00:52:13,560
Speaker 1: touch on crypto, on AI, on everything. They want 1000

788
00:52:13,560 --> 00:52:17,209
Speaker 1: innovations to bloom. Is that a risk that we are

789
00:52:17,209 --> 00:52:19,959
Speaker 1: going a bit on the other extreme, bit of a wild,

790
00:52:19,969 --> 00:52:20,969
Speaker 1: wild west, if you will.

791
00:52:23,040 --> 00:52:27,179
Speaker 2: I, I, I think there, there is a risk, and

792
00:52:27,239 --> 00:52:33,479
Speaker 2: it is, you know, an oft repeated argument that any technology,

793
00:52:33,600 --> 00:52:36,919
Speaker 2: you know, has its good and bad sides and could be,

794
00:52:37,040 --> 00:52:41,379
Speaker 2: can be abused, um, and, you know, the onus is on.

795
00:52:42,560 --> 00:52:47,239
Speaker 2: The creators, you know, providers of the technology, you know,

796
00:52:47,360 --> 00:52:51,040
Speaker 2: obviously the users on their own recognizance, you know, but

797
00:52:51,040 --> 00:52:53,840
Speaker 2: and also society to react to that and to try to,

798
00:52:54,159 --> 00:52:59,239
Speaker 2: you know, put safeguards in place. I think the likelihood

799
00:52:59,239 --> 00:53:06,138
Speaker 2: for government regulation is probably not realistic for the foreseeable future,

800
00:53:07,719 --> 00:53:10,879
Speaker 2: not only just, you know, not only because of the

801
00:53:10,879 --> 00:53:12,179
Speaker 2: dynamics of the current.

802
00:53:12,570 --> 00:53:17,279
Speaker 2: Administration in the US, but also um the rise of

803
00:53:17,290 --> 00:53:24,928
Speaker 2: the sovereign push for culture-specific, nation-specific investments in AI and

804
00:53:24,929 --> 00:53:32,399
Speaker 2: in LLMs in particular because um of the, the natural language, um,

805
00:53:32,610 --> 00:53:40,479
Speaker 2: you know, intersection with culture and identity, um, these particular products, really, ah,

806
00:53:40,610 --> 00:53:41,510
Speaker 2: every nation.

807
00:53:41,889 --> 00:53:45,250
Speaker 2: Um, you know, feels like they need their own, you know,

808
00:53:45,409 --> 00:53:49,250
Speaker 2: as the end result of it, and I think that's, that's,

809
00:53:49,370 --> 00:53:52,370
Speaker 2: that's happening today, but it's going to happen even more

810
00:53:52,370 --> 00:53:57,389
Speaker 2: so in the next 10 years, um, where, ah, you know,

811
00:53:57,729 --> 00:54:02,330
Speaker 2: every language, every ethnic group, certainly every nation is going

812
00:54:02,330 --> 00:54:03,489
Speaker 2: to have their own.

813
00:54:04,350 --> 00:54:10,350
Speaker 2: AI models and their own, you know, investment in creating

814
00:54:11,250 --> 00:54:16,290
Speaker 2: training data sets and corpuses of information, you know, to

815
00:54:16,290 --> 00:54:21,770
Speaker 2: build from. So all of that means that you can't

816
00:54:21,770 --> 00:54:25,509
Speaker 2: really have any kind of controls like no one country.

817
00:54:26,229 --> 00:54:29,550
Speaker 2: will be willing to put controls on its industry or

818
00:54:29,550 --> 00:54:33,870
Speaker 2: its development, right, if everyone else is pushing, you know,

819
00:54:33,959 --> 00:54:36,429
Speaker 2: as hard as they can, as fast as they can. Um,

820
00:54:36,790 --> 00:54:39,790
Speaker 2: so I think the, the, the dynamics, at least for

821
00:54:39,790 --> 00:54:43,870
Speaker 2: the foreseeable future, and that's probably, you know, something like

822
00:54:43,870 --> 00:54:47,439
Speaker 2: a decade, is that they're going to be really minimal.

823
00:54:48,139 --> 00:54:53,520
Speaker 2: regulatory controls coming in over the top. Now I think

824
00:54:53,909 --> 00:54:57,689
Speaker 2: each company, certainly the leading companies that provide these models

825
00:54:57,689 --> 00:54:58,760
Speaker 2: that are very powerful.

826
00:55:00,209 --> 00:55:05,639
Speaker 2: They are feeling a lot of heat around instituting controls,

827
00:55:06,169 --> 00:55:10,649
Speaker 2: you know, Anthropic just published a very detailed report on

828
00:55:10,649 --> 00:55:16,209
Speaker 2: how their product was used in a cybersecurity incident. what

829
00:55:16,209 --> 00:55:20,139
Speaker 2: they believe to be a state sponsored entity used quad

830
00:55:20,689 --> 00:55:28,969
Speaker 2: to basically write code on the fly to hack into

831
00:55:28,969 --> 00:55:29,689
Speaker 2: a system.

832
00:55:30,189 --> 00:55:35,489
Speaker 2: And or a set of systems and steal information and

833
00:55:36,070 --> 00:55:40,310
Speaker 2: it was it was very effective. They evaded the controls

834
00:55:40,310 --> 00:55:44,449
Speaker 2: that Claude has to try to prevent that use case

835
00:55:45,350 --> 00:55:50,189
Speaker 2: and opening since, or rather Anthropic has since tried to

836
00:55:50,189 --> 00:55:54,060
Speaker 2: strengthen those and improve upon those and to their credit,

837
00:55:54,070 --> 00:55:57,919
Speaker 2: they were very open about this and published it and it's,

838
00:55:58,270 --> 00:55:59,739
Speaker 2: it's important because

839
00:56:00,449 --> 00:56:04,689
Speaker 2: Any provider of an LLM now needs to, you know,

840
00:56:04,810 --> 00:56:06,689
Speaker 2: really think about how to do the same thing and

841
00:56:06,689 --> 00:56:10,649
Speaker 2: how to institute the same controls, but I think, I

842
00:56:10,649 --> 00:56:12,929
Speaker 2: think for now at least the onus is really going

843
00:56:12,929 --> 00:56:16,209
Speaker 2: to be on the private sector to try to prevent abuse.

844
00:56:17,080 --> 00:56:20,810
Speaker 2: And you know, hopefully, hopefully head off government regulation, you know, if,

845
00:56:20,879 --> 00:56:21,820
Speaker 2: if they do a good job.

846
00:56:22,909 --> 00:56:25,029
Speaker 1: No, um, I remember Skynet, OK.

847
00:56:26,320 --> 00:56:26,939
Speaker 1: Oh yeah.

848
00:56:27,879 --> 00:56:31,310
Speaker 1: Um, OK. Tell us, uh, from your vantage point, you

849
00:56:31,310 --> 00:56:35,850
Speaker 1: look at lots of potential investor pitches on a regular basis. Uh,

850
00:56:36,199 --> 00:56:39,479
Speaker 1: few really promising tech that you see in the pipeline,

851
00:56:39,760 --> 00:56:41,800
Speaker 1: get us excited about the future other than just the

852
00:56:41,800 --> 00:56:42,719
Speaker 1: LLM stuff, Noah.

853
00:56:44,810 --> 00:56:50,500
Speaker 2: Yeah, absolutely. Well, yeah, I mentioned Niantic Spatial, um, so

854
00:56:50,500 --> 00:56:53,689
Speaker 2: we're very excited about the real world model that they're

855
00:56:53,689 --> 00:56:59,459
Speaker 2: building and what that can accomplish. We have, uh, we

856
00:56:59,459 --> 00:57:04,649
Speaker 2: also invested in the neurosymbolic company that I mentioned, um,

857
00:57:04,860 --> 00:57:09,138
Speaker 2: and we're excited about how that could be potentially a

858
00:57:09,139 --> 00:57:14,080
Speaker 2: new paradigm for general purpose AI systems.

859
00:57:15,229 --> 00:57:21,530
Speaker 2: We're also big believers in in small models and vertical

860
00:57:21,739 --> 00:57:26,010
Speaker 2: uses of AI. We recently backed a company called Fuse

861
00:57:26,310 --> 00:57:32,889
Speaker 2: AI that has created a marketing automation solution that allows

862
00:57:32,949 --> 00:57:37,229
Speaker 2: a small to medium sized business to execute an end

863
00:57:37,229 --> 00:57:44,090
Speaker 2: to end process using AI and smart agents to identify customers.

864
00:57:44,370 --> 00:57:48,810
Speaker 2: Initiate outreach to them, engage with them, and then track

865
00:57:48,810 --> 00:57:54,290
Speaker 2: their progress to replace 5 common systems used for the

866
00:57:54,290 --> 00:57:58,969
Speaker 2: sales management cycle from the CRM from data gathering to

867
00:57:58,969 --> 00:58:00,729
Speaker 2: the CRM to the email.

868
00:58:01,159 --> 00:58:07,590
Speaker 2: Outreach to the the the sales process oversight and uh

869
00:58:08,010 --> 00:58:11,639
Speaker 2: we're we're very excited about that um it's a great,

870
00:58:11,830 --> 00:58:14,879
Speaker 2: it's a great ROI and it also um really brings

871
00:58:14,879 --> 00:58:18,879
Speaker 2: the benefit of AI to a smaller customer set that

872
00:58:18,879 --> 00:58:21,439
Speaker 2: can use it in a turnkey way, um.

873
00:58:22,399 --> 00:58:26,379
Speaker 2: We, we also have backed a um text to speech

874
00:58:26,800 --> 00:58:32,479
Speaker 2: company called Resemble AI which just released an open-source version

875
00:58:32,479 --> 00:58:36,469
Speaker 2: of their platform which has really taken off and uh we, we're,

876
00:58:36,600 --> 00:58:41,320
Speaker 2: we're seeing a lot of um that um uh benefit

877
00:58:41,320 --> 00:58:45,300
Speaker 2: of open-source um ecosystem development.

878
00:58:45,770 --> 00:58:50,000
Speaker 2: Um, to that enables developers to test and try something

879
00:58:50,000 --> 00:58:52,600
Speaker 2: and get familiar with it and then adopt it to

880
00:58:52,600 --> 00:58:58,010
Speaker 2: be really important in this world. Resemble also took their

881
00:58:58,010 --> 00:59:03,810
Speaker 2: experience in enabling voice, um, AI to be, you know, so,

882
00:59:03,969 --> 00:59:09,510
Speaker 2: so much more useful and has essentially applied the inverse

883
00:59:09,510 --> 00:59:13,209
Speaker 2: of it to create a deep fake detection platform which

884
00:59:13,209 --> 00:59:14,629
Speaker 2: is a cybersecurity tool.

885
00:59:15,199 --> 00:59:18,510
Speaker 2: Which they now market on an, on an enterprise B2B

886
00:59:18,510 --> 00:59:23,459
Speaker 2: basis and recently raised a round of funding with some

887
00:59:23,719 --> 00:59:26,919
Speaker 2: great investors to to take that to market. So we're

888
00:59:26,919 --> 00:59:29,959
Speaker 2: we're very excited about resemble AI as well.

889
00:59:31,060 --> 00:59:31,750
Speaker 1: All right, that

890
00:59:31,750 --> 00:59:34,469
Speaker 1: sounds really, really good. Actually, I'm really glad that we're

891
00:59:34,469 --> 00:59:38,489
Speaker 1: ending on an optimistic note. Uh, Noah, uh, thank you

892
00:59:38,489 --> 00:59:40,510
Speaker 1: so much for your time and insights and thank you

893
00:59:40,510 --> 00:59:42,429
Speaker 1: again for taking the time out of your holiday for us.

894
00:59:43,659 --> 00:59:45,639
Speaker 2: Absolutely, it's my pleasure. Great to be with you.

895
00:59:46,070 --> 00:59:48,129
Speaker 1: Great to be with you. Thanks to our listeners as well.

896
00:59:48,469 --> 00:59:51,379
Speaker 1: This podcast was produced by Ken Delbridge at Spice Studios.

897
00:59:51,780 --> 00:59:54,770
Speaker 1: Valet Lee and Daisy S Sherma provided additional assistance. Copy

898
00:59:54,770 --> 00:59:57,629
Speaker 1: time is for information only and does not constitute any

899
00:59:57,629 --> 01:00:01,310
Speaker 1: investment advice. All 167 episodes of the series are available

900
01:00:01,310 --> 01:00:05,169
Speaker 1: on YouTube and on all major platforms including Apple and Spotify.

901
01:00:05,510 --> 01:00:07,590
Speaker 1: For our research content and webinars, you can find them

902
01:00:07,590 --> 01:00:11,010
Speaker 1: all by Googling DBS Research Library. Have a great day.