1
00:00:00,040 --> 00:00:14,920
Speaker 1: Bloomberg Audio Studios, podcasts, radio news. Bloomberg Tech is live

2
00:00:15,000 --> 00:00:18,680
Speaker 1: from the heart of Silicon Valley with Ed Larlow in

3
00:00:18,920 --> 00:00:19,760
Speaker 1: San Francisco.

4
00:00:22,680 --> 00:00:25,480
Speaker 2: This is Bloomberg Tech, coming up tech's biggest names and

5
00:00:25,560 --> 00:00:29,720
Speaker 2: making the case for open AI, arguing open weight models

6
00:00:29,840 --> 00:00:33,080
Speaker 2: a key to innovation, competition, and even security.

7
00:00:33,280 --> 00:00:34,680
Speaker 3: Plus Intel showing signs of.

8
00:00:34,680 --> 00:00:38,680
Speaker 2: An AI driven comeback, forecasting stronger sales as its data

9
00:00:38,680 --> 00:00:42,479
Speaker 2: center business accelerates, and AMD and Cerebras are teaming up

10
00:00:42,479 --> 00:00:45,440
Speaker 2: to challenge in video, promising some of the fastest AI

11
00:00:45,520 --> 00:00:49,760
Speaker 2: systems on the market. We speak with Cerebra CEO Andrew Feldman,

12
00:00:50,000 --> 00:00:52,760
Speaker 2: our top story this Friday, A major push to shape

13
00:00:52,760 --> 00:00:57,080
Speaker 2: the future of AI. A coalition of technology companies, CEOs

14
00:00:57,120 --> 00:01:01,240
Speaker 2: and venture firms is urging Washington to embrace open weight

15
00:01:01,400 --> 00:01:06,360
Speaker 2: AI models, arguing they're critical to American competitiveness, innovation, and

16
00:01:06,480 --> 00:01:10,360
Speaker 2: national security. The signatories include Microsoft, such and Adella, and

17
00:01:10,520 --> 00:01:14,120
Speaker 2: Nvidia CEO Jensenmong, who used his very first post on

18
00:01:14,640 --> 00:01:17,559
Speaker 2: X to share the letter. It argues, quote the United

19
00:01:17,640 --> 00:01:21,840
Speaker 2: States should lead in building an open AI ecosystem because

20
00:01:21,840 --> 00:01:27,680
Speaker 2: it expands opportunities, strengthens competition, and extends American technological leadership.

21
00:01:27,800 --> 00:01:32,240
Speaker 2: Bloomberg's Tech Managing editor Sarah Fries with us the idea

22
00:01:32,480 --> 00:01:34,760
Speaker 2: of open weight and open source that's not new, but

23
00:01:34,840 --> 00:01:38,800
Speaker 2: it was very, very interesting the timing of the heavyweights

24
00:01:38,840 --> 00:01:42,480
Speaker 2: of this industry coming out almost in unison and saying

25
00:01:42,720 --> 00:01:45,680
Speaker 2: we need to think about this area.

26
00:01:45,760 --> 00:01:48,520
Speaker 4: I mean, it's a huge issue because right now, and

27
00:01:48,560 --> 00:01:52,200
Speaker 4: they didn't see it in the letter, there is a

28
00:01:52,480 --> 00:01:56,560
Speaker 4: really intense discussion happening happening about Moonshot's Kimmy. This is

29
00:01:56,600 --> 00:01:59,840
Speaker 4: a model that just came out out of China that

30
00:02:00,240 --> 00:02:03,280
Speaker 4: is pretty equivalent with US models, and it's changing the

31
00:02:03,280 --> 00:02:07,000
Speaker 4: way people think about it. Whereas previously the Chinese models

32
00:02:07,360 --> 00:02:11,320
Speaker 4: were good open weight models and able to undercut US

33
00:02:11,320 --> 00:02:14,320
Speaker 4: models on price, and they were very popular, but they

34
00:02:14,320 --> 00:02:17,400
Speaker 4: didn't actually get to the point where the innovation looked

35
00:02:17,440 --> 00:02:20,359
Speaker 4: like it was about equivalent with what you could get

36
00:02:20,360 --> 00:02:23,320
Speaker 4: from US, and that's caused a lot of a concern

37
00:02:23,400 --> 00:02:30,240
Speaker 4: in Washington, maybe some policymakers threatening to change things, restrict things,

38
00:02:30,560 --> 00:02:34,520
Speaker 4: And of course US startups really depend on those open

39
00:02:34,520 --> 00:02:37,799
Speaker 4: weight models. So if the US were to have something

40
00:02:37,840 --> 00:02:41,079
Speaker 4: that was equivalent, that would really help a lot of

41
00:02:41,800 --> 00:02:45,040
Speaker 4: the companies. But you know, everything's getting built right now

42
00:02:45,120 --> 00:02:47,880
Speaker 4: and we have a lot of dependence on these Chinese models.

43
00:02:47,960 --> 00:02:51,160
Speaker 4: I'm very curious what's going to happen in Washington as

44
00:02:51,360 --> 00:02:54,600
Speaker 4: we digest what this means for the market.

45
00:02:55,680 --> 00:02:58,279
Speaker 2: I don't want to trivialize this, but it was fascinating

46
00:02:58,320 --> 00:03:01,160
Speaker 2: that the Jensen Wong elected to to create an x

47
00:03:01,200 --> 00:03:05,160
Speaker 2: account and use his first post as a mechanism to

48
00:03:05,200 --> 00:03:08,120
Speaker 2: communicate this. For Jensen, this isn't you You know, he's

49
00:03:08,160 --> 00:03:14,680
Speaker 2: talked about the importance of open models, open weight AIS,

50
00:03:14,720 --> 00:03:17,320
Speaker 2: open source. The distinction is like, if you use a

51
00:03:17,360 --> 00:03:19,880
Speaker 2: closed model, you go via the API, but you can't

52
00:03:19,919 --> 00:03:22,880
Speaker 2: see the model, you can't download it. In open weight,

53
00:03:22,960 --> 00:03:26,600
Speaker 2: you can download and run the train model. The data

54
00:03:26,680 --> 00:03:30,040
Speaker 2: code aren't necessarily public on open source, all of it's

55
00:03:30,080 --> 00:03:34,200
Speaker 2: they're available. You were absolutely right to frame this post

56
00:03:34,240 --> 00:03:37,280
Speaker 2: Kimmi K three, but also with what happened with the

57
00:03:37,320 --> 00:03:41,680
Speaker 2: open AI mistaken hacking of hugging Face, it also raised

58
00:03:41,720 --> 00:03:45,360
Speaker 2: some of the limitations of US open models because hugging

59
00:03:45,400 --> 00:03:49,160
Speaker 2: Face wasn't able to defend itself in that respect. What

60
00:03:49,200 --> 00:03:51,160
Speaker 2: do you think happens next? How serious is this?

61
00:03:51,600 --> 00:03:51,760
Speaker 3: Well?

62
00:03:51,800 --> 00:03:55,800
Speaker 4: I think that It is serious because the whole market

63
00:03:55,880 --> 00:03:58,920
Speaker 4: is resting on this idea that all of this tremendous

64
00:03:58,960 --> 00:04:02,320
Speaker 4: capital expenditure in investment, the build out of data centers,

65
00:04:02,360 --> 00:04:04,440
Speaker 4: it's all going to end up having a return on

66
00:04:04,480 --> 00:04:07,160
Speaker 4: an investment for US companies. There's a whole other factor

67
00:04:07,240 --> 00:04:10,400
Speaker 4: we haven't talked about yet, which is some concern that

68
00:04:10,440 --> 00:04:16,760
Speaker 4: these Chinese models are built in part using unauthorized distillation

69
00:04:17,000 --> 00:04:20,800
Speaker 4: of US models, including anthropics claud So if the if

70
00:04:20,839 --> 00:04:24,680
Speaker 4: all of that investment from the US companies in their

71
00:04:25,240 --> 00:04:29,360
Speaker 4: future innovation is then just being used to boost what

72
00:04:29,480 --> 00:04:34,720
Speaker 4: China can accomplish, whether it's through distillation or through unauthorized

73
00:04:34,720 --> 00:04:38,520
Speaker 4: ships use, that's that's a matter of big debate too,

74
00:04:38,560 --> 00:04:42,760
Speaker 4: and that could really affect the ROI for these companies

75
00:04:42,880 --> 00:04:46,440
Speaker 4: on the many, many billions that they're pouring in to

76
00:04:47,240 --> 00:04:49,360
Speaker 4: build out for the future of AI.

77
00:04:49,720 --> 00:04:52,360
Speaker 2: That's probably the bit we don't know. You know, why

78
00:04:52,400 --> 00:04:54,599
Speaker 2: did they do this and why now? Was it to

79
00:04:54,680 --> 00:04:58,719
Speaker 2: counter distillation? Was it to try and get ahead of

80
00:04:58,720 --> 00:05:01,360
Speaker 2: what happened with open AI hug We'll find out. Bloomberg

81
00:05:01,400 --> 00:05:04,400
Speaker 2: Sarah Friar, who leads the team here at Bloomberg Tech,

82
00:05:04,400 --> 00:05:04,600
Speaker 2: thank you.

83
00:05:04,720 --> 00:05:05,000
Speaker 3: Very much.

84
00:05:05,080 --> 00:05:08,640
Speaker 2: Let's turn to earnings. Shares of Intel it's interesting now

85
00:05:08,680 --> 00:05:12,080
Speaker 2: down three percent. When the earnings hit last night, we

86
00:05:12,120 --> 00:05:14,640
Speaker 2: saw the stocking after ours go as high as high

87
00:05:14,680 --> 00:05:17,400
Speaker 2: as a gain of thirteen percent. The chip maker delivered

88
00:05:17,400 --> 00:05:21,600
Speaker 2: a stronger than expected forecast as demand from AI data

89
00:05:21,640 --> 00:05:25,400
Speaker 2: centers is fueling a surge in its business, particularly CPU

90
00:05:25,600 --> 00:05:29,240
Speaker 2: CEO Lip Bhutan telling me CPU demand is now outpacing

91
00:05:29,520 --> 00:05:33,800
Speaker 2: what is improving supply. Joining us now is Antoine Skyban,

92
00:05:33,960 --> 00:05:37,200
Speaker 2: head of Global Technology Infrastructure Research at New Street Research.

93
00:05:37,240 --> 00:05:39,560
Speaker 2: He has a neutral rating on the stock with a

94
00:05:39,640 --> 00:05:43,320
Speaker 2: one hundred and fifteen dollars price target. I want to

95
00:05:43,320 --> 00:05:45,160
Speaker 2: get into the technology. I want to get into what

96
00:05:45,200 --> 00:05:48,159
Speaker 2: Intel is actually doing and well, but the stock, you know,

97
00:05:48,240 --> 00:05:52,279
Speaker 2: I was sat here last night and the gains were massive,

98
00:05:52,520 --> 00:05:56,599
Speaker 2: then they eroded. Now we're down three percent with time,

99
00:05:57,320 --> 00:05:59,400
Speaker 2: What has the market changed its mind about?

100
00:06:02,040 --> 00:06:04,839
Speaker 5: I think, I mean, thanks for having on the show, Adam.

101
00:06:05,160 --> 00:06:08,320
Speaker 5: I think you know what's what's happening is the market

102
00:06:08,440 --> 00:06:11,280
Speaker 5: is realizing that actually that's the even the bookcase is

103
00:06:11,279 --> 00:06:15,200
Speaker 5: already priced in, you know, even if you take like

104
00:06:15,240 --> 00:06:19,000
Speaker 5: an Intel in twenty thirty, where manufacturing is fully turned

105
00:06:19,000 --> 00:06:21,560
Speaker 5: around which, by the way, we had indications of that

106
00:06:21,680 --> 00:06:26,279
Speaker 5: yesterday on the print NNEL. You have foundry bit margins

107
00:06:26,320 --> 00:06:31,119
Speaker 5: at nine points sequentially, so they're headed really towards break even.

108
00:06:31,800 --> 00:06:33,920
Speaker 5: Even if you assume, you know that in twenty thirty

109
00:06:33,960 --> 00:06:36,360
Speaker 5: in that is fully recovered, they're going to still have,

110
00:06:36,440 --> 00:06:40,919
Speaker 5: you know, some structural cost disadvantages compared to CSMC, and

111
00:06:41,120 --> 00:06:43,560
Speaker 5: gross margins are never going to reach the same level

112
00:06:43,560 --> 00:06:48,320
Speaker 5: as CSMC and in products even if they're doing extremely well.

113
00:06:48,360 --> 00:06:52,280
Speaker 5: And even if you know, the overall egentic tide is

114
00:06:52,600 --> 00:06:56,240
Speaker 5: you know, lifting or boats. Inta is probably going to

115
00:06:56,320 --> 00:06:58,679
Speaker 5: keep losing share to AMD, is probably going to lose

116
00:06:58,720 --> 00:07:02,640
Speaker 5: share to ARM. You know, you have hyperscires in house designs,

117
00:07:02,680 --> 00:07:05,520
Speaker 5: you have Nvidia remping you a grace, there are et cetera.

118
00:07:06,080 --> 00:07:08,600
Speaker 5: All of that is going to take share from Intel.

119
00:07:08,680 --> 00:07:11,440
Speaker 5: And if you put all that together, even the boot

120
00:07:11,520 --> 00:07:15,680
Speaker 5: case you know, doesn't make much room to the current levels,

121
00:07:15,880 --> 00:07:16,600
Speaker 5: just like he's treading it.

122
00:07:17,920 --> 00:07:19,800
Speaker 3: Okay, Ed, calm down.

123
00:07:19,880 --> 00:07:22,440
Speaker 2: Remind yourself and the audience that going into the print

124
00:07:22,440 --> 00:07:24,400
Speaker 2: the stock was up one hundred and seventy percent year

125
00:07:24,400 --> 00:07:26,480
Speaker 2: to day, so that might be a part of it.

126
00:07:27,080 --> 00:07:30,480
Speaker 2: What's interesting as well is that Intel never said they'd

127
00:07:30,560 --> 00:07:33,040
Speaker 2: fixed everything, just that they were making progress. So the

128
00:07:33,040 --> 00:07:36,200
Speaker 2: way that Litboo explained it to me is that, you know,

129
00:07:36,360 --> 00:07:40,040
Speaker 2: their processes and production for their own products is improving,

130
00:07:40,440 --> 00:07:43,600
Speaker 2: but they're still in a place where demand for CPU

131
00:07:43,600 --> 00:07:48,000
Speaker 2: in particular is outpacing what is improving supply, right, Antoine,

132
00:07:48,000 --> 00:07:50,400
Speaker 2: You'll remember last quarter they basically left money on the

133
00:07:50,440 --> 00:07:53,920
Speaker 2: table because they couldn't meet the demand. Do you recognize

134
00:07:53,920 --> 00:07:55,520
Speaker 2: that kind of operational progress?

135
00:07:57,200 --> 00:07:59,160
Speaker 5: Yes, I mean we heard a lot of very encouraging

136
00:07:59,240 --> 00:08:02,520
Speaker 5: data points on uncapacity increases on the call.

137
00:08:02,960 --> 00:08:04,760
Speaker 6: First of all, you know you have yields, of course,

138
00:08:04,800 --> 00:08:05,320
Speaker 6: are improving.

139
00:08:05,360 --> 00:08:08,040
Speaker 5: That directly increases, you know, the the output of the

140
00:08:08,240 --> 00:08:11,080
Speaker 5: good chips that you can produce. On top of that,

141
00:08:11,760 --> 00:08:14,280
Speaker 5: Intel has a lot of shells that they've been building

142
00:08:14,320 --> 00:08:17,240
Speaker 5: over the last few years that they can now feed

143
00:08:17,320 --> 00:08:19,760
Speaker 5: up with equipment. All of that puts Intel in a

144
00:08:19,800 --> 00:08:22,840
Speaker 5: good position, you know, to well Number one, as you said,

145
00:08:22,880 --> 00:08:25,240
Speaker 5: you know, address the huge demand that there is for

146
00:08:25,600 --> 00:08:29,800
Speaker 5: CPUs in this agentic era, and two potentially start addressing

147
00:08:29,840 --> 00:08:32,319
Speaker 5: you know, demand from external customers as well. You know,

148
00:08:32,360 --> 00:08:35,600
Speaker 5: they target at fifteen billion external foundry revenues in twenty thirty.

149
00:08:35,679 --> 00:08:37,640
Speaker 6: Yes, and I think they're increasing.

150
00:08:37,480 --> 00:08:41,160
Speaker 5: Evidence that are actually pretty tangible.

151
00:08:42,120 --> 00:08:44,400
Speaker 2: So let's talk about CAPEX, which will be more than

152
00:08:44,400 --> 00:08:47,040
Speaker 2: twenty billion dollars now this year and grow over the

153
00:08:47,040 --> 00:08:50,200
Speaker 2: next couple of years. Let boots and the Intel CEO

154
00:08:50,280 --> 00:08:53,720
Speaker 2: is cautious. He won't deploy CAPEX unless he thinks there'll

155
00:08:53,720 --> 00:08:56,400
Speaker 2: be a return. How did you interpret all that?

156
00:08:58,040 --> 00:09:00,120
Speaker 5: Well, I think it's it's very reassuring, you know, for

157
00:09:00,400 --> 00:09:03,640
Speaker 5: investors to hear that. I think you know, in order

158
00:09:03,679 --> 00:09:06,439
Speaker 5: to generate that demand, Intel still has to execute. And

159
00:09:06,679 --> 00:09:09,120
Speaker 5: let's keep in mind that IBIT margins for boundary are

160
00:09:09,160 --> 00:09:12,640
Speaker 5: still in negative territory. Gross margins are probably also still

161
00:09:12,679 --> 00:09:15,240
Speaker 5: in negative territory. So they still have, you know, a

162
00:09:15,280 --> 00:09:19,240
Speaker 5: lot of progress to make to end up in a

163
00:09:19,240 --> 00:09:21,960
Speaker 5: situation where they can actually really invest that CAPEX with

164
00:09:22,040 --> 00:09:24,400
Speaker 5: the confidence that they're going to be able to sell

165
00:09:24,440 --> 00:09:29,520
Speaker 5: those wafers and turn a profit out of selling those wafers. Now,

166
00:09:29,640 --> 00:09:32,880
Speaker 5: in addition, you know, like that you alluded to the

167
00:09:32,880 --> 00:09:34,880
Speaker 5: twenty billion dollar command, A lot of that is going

168
00:09:34,920 --> 00:09:37,120
Speaker 5: to be of course for equipment, because as I mentioned,

169
00:09:37,120 --> 00:09:37,360
Speaker 5: they have.

170
00:09:37,360 --> 00:09:40,680
Speaker 3: A toole space yeah, tools not space.

171
00:09:40,840 --> 00:09:42,360
Speaker 6: Yeah, exactly if a way of equipment.

172
00:09:42,600 --> 00:09:45,800
Speaker 5: And now the question is what happens in twenty twenty seven,

173
00:09:46,040 --> 00:09:48,400
Speaker 5: And for that I think it's still openenda. You know,

174
00:09:48,440 --> 00:09:51,240
Speaker 5: it depends on whether they can you know, keep improving

175
00:09:51,280 --> 00:09:53,920
Speaker 5: those wills, keep improving the corast structure of these wafers,

176
00:09:54,360 --> 00:09:57,120
Speaker 5: and get into positive margin territory for boundary.

177
00:09:59,400 --> 00:10:02,520
Speaker 2: Let's say widen this out to what's happening in infrastructure.

178
00:10:03,000 --> 00:10:07,360
Speaker 2: What do we learn from Intel's print about the AI cycle.

179
00:10:09,480 --> 00:10:11,959
Speaker 5: Well, of course, like the d C print is in Brazil,

180
00:10:12,600 --> 00:10:17,680
Speaker 5: like they're very nearly sixty percent that the DCAI business.

181
00:10:18,679 --> 00:10:21,200
Speaker 5: I think an interesting data point is that Intel also

182
00:10:21,200 --> 00:10:23,680
Speaker 5: still has a lot of room you know to improve

183
00:10:23,720 --> 00:10:28,240
Speaker 5: you know, the roadmap on the AIXPU front. You know,

184
00:10:28,679 --> 00:10:31,160
Speaker 5: we're talking about like a two billion run rate today

185
00:10:31,280 --> 00:10:34,440
Speaker 5: for like everything that's not CPUs you know, in the

186
00:10:34,480 --> 00:10:38,240
Speaker 5: d CI segment, the thing they have visibility too.

187
00:10:38,240 --> 00:10:41,480
Speaker 6: Four billion. That's like orders of magnitude.

188
00:10:40,880 --> 00:10:44,240
Speaker 5: You know, smaller than for example, like an Nvidia that's

189
00:10:44,400 --> 00:10:47,800
Speaker 5: approaching a four hundred billion dollar annual run rates for

190
00:10:48,360 --> 00:10:49,680
Speaker 5: their data center business.

191
00:10:50,360 --> 00:10:51,960
Speaker 6: So I think that's that's interesting.

192
00:10:51,960 --> 00:10:54,120
Speaker 5: The data point also that's Intel still you know, has

193
00:10:54,160 --> 00:10:57,520
Speaker 5: plenty of opportunities maybe to gain some momentum there on

194
00:10:57,600 --> 00:11:01,559
Speaker 5: the CPU front. I mean it's an excellent through the

195
00:11:01,640 --> 00:11:05,040
Speaker 5: cross for m D. H if if if that is growing,

196
00:11:05,920 --> 00:11:09,880
Speaker 5: that's that quickly despite I think a roadmap that's not

197
00:11:10,000 --> 00:11:13,000
Speaker 5: that competitive. Like if you look at the density of

198
00:11:13,040 --> 00:11:14,960
Speaker 5: the of the chips that they that they're putting out,

199
00:11:15,000 --> 00:11:17,240
Speaker 5: like the number of threads, the number of course, which

200
00:11:17,320 --> 00:11:20,439
Speaker 5: knows their manufacturer done. I think, you know, em D

201
00:11:20,880 --> 00:11:24,120
Speaker 5: has a much stronger roadmap, So it's it's definitely just

202
00:11:24,120 --> 00:11:25,720
Speaker 5: from relocross for m D on the CPU.

203
00:11:25,800 --> 00:11:30,120
Speaker 2: Fromt Anton Scribe Ban of New Street Research back on

204
00:11:30,120 --> 00:11:33,320
Speaker 2: Bloomberg Tech, thank you very much. Coming up on the show,

205
00:11:33,440 --> 00:11:37,280
Speaker 2: Describe Therapeutics, the company behind gene editing CRISP, the technology

206
00:11:38,080 --> 00:11:40,040
Speaker 2: I p O today. And look who's joining the show

207
00:11:40,120 --> 00:11:43,880
Speaker 2: Scribe Therapeutics co founder, scientific advisor Jennifer Downer.

208
00:11:44,000 --> 00:11:44,559
Speaker 3: That's next.

209
00:11:44,840 --> 00:11:45,840
Speaker 6: This is Bloomberg Tech.

210
00:11:51,000 --> 00:12:02,840
Speaker 2: Yeah, Jennifer Downer the Nobel Prize for pioneering gene editing

211
00:12:02,880 --> 00:12:04,040
Speaker 2: technology Crisper.

212
00:12:04,200 --> 00:12:04,640
Speaker 3: Today.

213
00:12:05,120 --> 00:12:08,600
Speaker 2: Scribe Therapeutics, the company she co founded, just raised one

214
00:12:08,679 --> 00:12:10,960
Speaker 2: hundred and twenty eight point seven million in its IPO

215
00:12:11,320 --> 00:12:15,560
Speaker 2: for a much bigger ambition using Crisper to prevent heart disease.

216
00:12:15,559 --> 00:12:18,840
Speaker 2: Scribe Therapeutics co founder and scientific advisor Jennifer Downer is

217
00:12:18,840 --> 00:12:21,880
Speaker 2: with us now. I've been reading so deeply about this.

218
00:12:22,160 --> 00:12:24,640
Speaker 2: I think the best place to start is that your

219
00:12:24,720 --> 00:12:28,360
Speaker 2: lead therapy aims to lower LDL cholesterol with just a

220
00:12:28,400 --> 00:12:33,319
Speaker 2: single treatment, but distinct and unlike earlier gene editing approaches.

221
00:12:34,200 --> 00:12:37,640
Speaker 2: You say it does not permanently edit a patient's DNA.

222
00:12:38,160 --> 00:12:41,680
Speaker 2: Explain why that's such an important distinction, and move forward

223
00:12:41,679 --> 00:12:42,520
Speaker 2: in the technology.

224
00:12:43,880 --> 00:12:46,600
Speaker 7: Great to be here, ed It's an exciting moment in

225
00:12:46,800 --> 00:12:50,760
Speaker 7: gene therapies because, as you just said, Describe therapeutics has

226
00:12:50,800 --> 00:12:54,560
Speaker 7: a strategy that involves what we call epi editing. It

227
00:12:54,640 --> 00:12:58,360
Speaker 7: means making changes in DNA that aren't permanent but alter

228
00:12:58,600 --> 00:13:01,440
Speaker 7: the production of proteins. And we think this is an

229
00:13:01,480 --> 00:13:05,400
Speaker 7: incredibly important moment in the field because it means we

230
00:13:05,440 --> 00:13:08,439
Speaker 7: can now use this therapy safely and effectively for a

231
00:13:08,520 --> 00:13:09,240
Speaker 7: common disease.

232
00:13:10,840 --> 00:13:13,760
Speaker 2: Jennifer, may I ask you know, in any IPO, there's

233
00:13:13,800 --> 00:13:17,239
Speaker 2: obviously a reason to do it and raise the proceeds.

234
00:13:17,440 --> 00:13:20,040
Speaker 2: Right now, where are you in the cycle and the

235
00:13:20,040 --> 00:13:22,400
Speaker 2: process and what will you use those proceeds for.

236
00:13:23,720 --> 00:13:26,920
Speaker 7: Well Scribe Therapeutics is a clinical stage company. We're already

237
00:13:26,920 --> 00:13:30,839
Speaker 7: in our phase one trial for the first indication. The

238
00:13:30,960 --> 00:13:33,640
Speaker 7: funds that are raised today will help move that forward

239
00:13:33,800 --> 00:13:37,079
Speaker 7: and progress these therapies so that they can be deployed

240
00:13:37,160 --> 00:13:37,880
Speaker 7: much more widely.

241
00:13:39,840 --> 00:13:43,200
Speaker 2: At the start of our conversation, you explained the technology

242
00:13:43,200 --> 00:13:47,240
Speaker 2: difference when it comes to regulators. How important is that

243
00:13:47,320 --> 00:13:51,800
Speaker 2: distinction going to be versus the sort of permanent editing

244
00:13:51,840 --> 00:13:52,920
Speaker 2: of one's DNA.

245
00:13:54,040 --> 00:13:56,520
Speaker 7: Well Scribe, as a company that was really founded on

246
00:13:56,720 --> 00:14:00,840
Speaker 7: great science and has an amazing scientific team, invested the

247
00:14:00,920 --> 00:14:04,720
Speaker 7: last several years in improving the technology to the point

248
00:14:04,760 --> 00:14:11,360
Speaker 7: where it's safe and effective to treat disease as as

249
00:14:11,480 --> 00:14:17,480
Speaker 7: a strategy for making non permanent changes in DNA. So

250
00:14:17,520 --> 00:14:19,960
Speaker 7: we think that this is going to be an effective

251
00:14:20,000 --> 00:14:23,520
Speaker 7: way to prevent cardiovascular disease in the future that will

252
00:14:23,600 --> 00:14:26,840
Speaker 7: be so safe that you can provide it much more broadly.

253
00:14:28,120 --> 00:14:29,920
Speaker 2: That's what we're talking about. You know, let's bring it

254
00:14:29,960 --> 00:14:33,040
Speaker 2: back to basics heart disease. And it's amazing. You know,

255
00:14:33,200 --> 00:14:37,720
Speaker 2: millions of people across America and across the world take

256
00:14:37,800 --> 00:14:41,400
Speaker 2: statins right every single day. But then you know, there's

257
00:14:41,440 --> 00:14:43,840
Speaker 2: a lot of evidence, a lot of data to support that.

258
00:14:43,880 --> 00:14:47,600
Speaker 2: Then they stop, this is a one single treatment that

259
00:14:47,640 --> 00:14:51,080
Speaker 2: you're working towards, and I wonder just how important that

260
00:14:51,120 --> 00:14:54,520
Speaker 2: will be for the behaviors of the patients receiving it.

261
00:14:56,080 --> 00:14:56,240
Speaker 8: Right.

262
00:14:56,280 --> 00:14:58,160
Speaker 7: We think about this a lot. I think it's very

263
00:14:58,160 --> 00:15:01,680
Speaker 7: interesting to think about a way treeing or preventing disease

264
00:15:02,200 --> 00:15:04,960
Speaker 7: that doesn't require taking a pill every day. Right. That

265
00:15:05,160 --> 00:15:09,400
Speaker 7: is really effectively a one or few time treatment. That

266
00:15:09,600 --> 00:15:12,240
Speaker 7: means that people are free of having to remember to

267
00:15:12,280 --> 00:15:14,400
Speaker 7: take a pill and they don't have to worry about

268
00:15:14,440 --> 00:15:18,280
Speaker 7: those kinds of side effects. It's a really interesting moment

269
00:15:18,360 --> 00:15:20,360
Speaker 7: where I think we're going to see a real change

270
00:15:20,440 --> 00:15:23,080
Speaker 7: in the way that medicine is delivered in the future.

271
00:15:24,800 --> 00:15:28,320
Speaker 2: When the Bloomberg Tech audience hear the phrase gene editing,

272
00:15:28,560 --> 00:15:32,040
Speaker 2: you know, inevitably people have concern, right, what is it

273
00:15:32,160 --> 00:15:35,320
Speaker 2: from a safety perspective that they need to understand?

274
00:15:37,040 --> 00:15:40,560
Speaker 7: Well, the Christmer technology is built around a strategy for

275
00:15:40,680 --> 00:15:45,280
Speaker 7: targeting DNA Precisely, with any technology, there's always risk, But

276
00:15:45,360 --> 00:15:48,280
Speaker 7: I think what's exciting about scribes approach is that they've

277
00:15:48,320 --> 00:15:53,000
Speaker 7: invested several years in ensuring that their strategy. This EPI

278
00:15:53,120 --> 00:15:57,320
Speaker 7: editing approach is really truly safe and effective for the

279
00:15:57,360 --> 00:16:00,240
Speaker 7: indications they're going after. And as we discussed before, this

280
00:16:00,280 --> 00:16:03,440
Speaker 7: is not a technology now that makes a permanent change

281
00:16:03,440 --> 00:16:05,600
Speaker 7: to DNA, and I think that's a real difference.

282
00:16:06,840 --> 00:16:11,240
Speaker 2: You had one of the most interesting and engaging conversations

283
00:16:11,240 --> 00:16:13,920
Speaker 2: I've listened to this year with Bloomberg's Emily Chang on

284
00:16:13,960 --> 00:16:17,080
Speaker 2: the Circuit and you talked about AI and you had

285
00:16:17,240 --> 00:16:21,200
Speaker 2: a level of skepticism about where AI really is today

286
00:16:21,520 --> 00:16:25,560
Speaker 2: across drug discovery, its utility in your field. I know

287
00:16:25,640 --> 00:16:28,120
Speaker 2: it's not been that many weeks or months, but have

288
00:16:28,240 --> 00:16:31,320
Speaker 2: you changed your mind? Do you still hold that position

289
00:16:31,360 --> 00:16:33,360
Speaker 2: on the limits of AI in your field right now?

290
00:16:34,280 --> 00:16:37,040
Speaker 7: Look, I think AI is an incredible tool. It's an

291
00:16:37,080 --> 00:16:42,600
Speaker 7: amazing technology that allows scientists now to accelerate the kinds

292
00:16:42,600 --> 00:16:44,840
Speaker 7: of work that we do. That being said, it doesn't

293
00:16:44,880 --> 00:16:48,880
Speaker 7: replace scientists. We still need innovators, we still need creative people,

294
00:16:49,000 --> 00:16:51,240
Speaker 7: We still need the new ideas that come out of

295
00:16:51,320 --> 00:16:53,320
Speaker 7: left field, and we don't see that with AI. We

296
00:16:53,400 --> 00:16:57,280
Speaker 7: see it being a great way to accelerate our ideas

297
00:16:57,400 --> 00:16:59,800
Speaker 7: and make it possible to do things faster. And I

298
00:16:59,800 --> 00:17:01,000
Speaker 7: think it's a great partnership.

299
00:17:02,480 --> 00:17:04,800
Speaker 2: Jennifer, bear with me if you would, But we have

300
00:17:04,840 --> 00:17:09,080
Speaker 2: some headlines that are just crossing that Right now. Scribe

301
00:17:09,359 --> 00:17:12,800
Speaker 2: is indicated to open at about twenty nine dollars a share.

302
00:17:13,200 --> 00:17:16,399
Speaker 2: The ipo priced at fifteen, so interesting because this was

303
00:17:16,400 --> 00:17:20,800
Speaker 2: a difficult environment for biotech IPOs. We're also learning that

304
00:17:21,680 --> 00:17:24,520
Speaker 2: the allocation about seventy five percent of the shares to

305
00:17:24,640 --> 00:17:27,400
Speaker 2: just ten investors reflect on that. I mean, this must

306
00:17:27,440 --> 00:17:30,920
Speaker 2: mean a lot to you, but clearly, you know, investors

307
00:17:30,920 --> 00:17:32,280
Speaker 2: are very receptive to this.

308
00:17:33,240 --> 00:17:36,680
Speaker 7: I think our investors really get this technology. They understand

309
00:17:36,720 --> 00:17:39,639
Speaker 7: what's different about what Scribe is doing. They understand the

310
00:17:39,760 --> 00:17:43,520
Speaker 7: quality of this team, it's leadership and its commitment to

311
00:17:43,600 --> 00:17:46,440
Speaker 7: real science, and that's what we're seeing I think reflected

312
00:17:46,480 --> 00:17:47,200
Speaker 7: in the marketplace.

313
00:17:48,560 --> 00:17:51,680
Speaker 2: Scribe Therapeutics co founder Jennifer Dowdener, thank you very much

314
00:17:51,680 --> 00:17:54,560
Speaker 2: for your time here on Bloomberg Tech. Astoriers want to

315
00:17:54,560 --> 00:17:57,119
Speaker 2: bring you real quick in the market's black Rock began

316
00:17:57,240 --> 00:18:00,880
Speaker 2: marketing twelve point three billion dollars of high raid bonds

317
00:18:01,119 --> 00:18:05,240
Speaker 2: to fund a meta data center project, testing investor appetite.

318
00:18:05,400 --> 00:18:09,159
Speaker 2: This as concerns grow about excessive AI infrastructure spending, a

319
00:18:09,200 --> 00:18:12,000
Speaker 2: source told Bloomberg the bonds are being offered by a

320
00:18:12,080 --> 00:18:15,760
Speaker 2: souper pilot investor holding company tied to black Rock and

321
00:18:15,960 --> 00:18:19,320
Speaker 2: JP Morgan Chase and Morgan Stanley are running the offering,

322
00:18:19,520 --> 00:18:23,520
Speaker 2: which is expected to price next week.

323
00:18:29,520 --> 00:18:31,320
Speaker 3: As time for Talking Tech first start.

324
00:18:31,359 --> 00:18:35,359
Speaker 2: The EU escalated a probe against TikTok, accusing it of

325
00:18:35,359 --> 00:18:37,159
Speaker 2: failing to protect the safety.

326
00:18:36,760 --> 00:18:38,760
Speaker 3: And privacy of teen users.

327
00:18:39,040 --> 00:18:42,600
Speaker 2: The commission said TikTok's custom accounts for users under eighteen

328
00:18:43,040 --> 00:18:46,040
Speaker 2: can be easily found and viewed by other people online,

329
00:18:46,080 --> 00:18:50,119
Speaker 2: which goes against the block's content moderation rule book. Plus

330
00:18:50,119 --> 00:18:54,359
Speaker 2: SoftBank is considering an acquisition of Gravis Robotics in an

331
00:18:54,400 --> 00:18:59,160
Speaker 2: effort to target the AI technologies underpinning robotics. According to sources,

332
00:18:59,440 --> 00:19:02,560
Speaker 2: a deal could ultimately value Gravis at more than five

333
00:19:02,680 --> 00:19:07,399
Speaker 2: hundred million dollars, and SAP reported stronger than expected cloud

334
00:19:07,440 --> 00:19:11,840
Speaker 2: growth as more customers adopt AI powered enterprise software. Bloomberg

335
00:19:11,880 --> 00:19:15,080
Speaker 2: spoke with CEO Christian Klin earlier about that, but also

336
00:19:15,160 --> 00:19:18,920
Speaker 2: how AI tokens are now managed like any other business expense.

337
00:19:19,600 --> 00:19:23,600
Speaker 9: In the d costs, you see a higher total consumption

338
00:19:23,720 --> 00:19:28,560
Speaker 9: as we are using AI to code additional features, but

339
00:19:28,720 --> 00:19:30,360
Speaker 9: especially now additional agents.

340
00:19:30,680 --> 00:19:31,440
Speaker 10: But vice versa.

341
00:19:31,560 --> 00:19:34,920
Speaker 9: You also see a huge productivity increase of thirty percent. Obviously,

342
00:19:35,200 --> 00:19:38,680
Speaker 9: we are also managing totns as part of our cost budget. Yeah,

343
00:19:38,680 --> 00:19:40,719
Speaker 9: you have a headcon budget, you have some party budget,

344
00:19:40,760 --> 00:19:43,800
Speaker 9: and you have tokens. So our managers in R and

345
00:19:43,880 --> 00:19:47,040
Speaker 9: the Indigo to market space, they have one budget and

346
00:19:47,119 --> 00:19:50,560
Speaker 9: they have to manage that also according to the productivity

347
00:19:50,600 --> 00:19:52,840
Speaker 9: assumptions we are reflected in the budget.

348
00:19:54,320 --> 00:19:55,680
Speaker 3: Up to two postponements.

349
00:19:55,720 --> 00:19:59,080
Speaker 2: All eyes are on SpaceX's Starship test flight today. The

350
00:19:59,080 --> 00:20:03,040
Speaker 2: mission includes up raided Starlink satellites that will intentionally burn

351
00:20:03,160 --> 00:20:06,680
Speaker 2: up during re entry. The launch comes as SpaceX makes

352
00:20:06,720 --> 00:20:11,119
Speaker 2: a major strategic bet on Starship. Bloomberg's learn the companies

353
00:20:11,160 --> 00:20:15,280
Speaker 2: already turning away some potential Falcon nine customers beyond twenty

354
00:20:15,320 --> 00:20:18,960
Speaker 2: twenty eight as it shifts resources to its next generation rocket.

355
00:20:18,960 --> 00:20:22,960
Speaker 2: That's all according to sources Bloomberg. Santa Pashanka broke the

356
00:20:23,000 --> 00:20:26,679
Speaker 2: story and joins us, now, really interesting to work with

357
00:20:26,720 --> 00:20:29,000
Speaker 2: you on this one. There's a lot of detail in there,

358
00:20:29,080 --> 00:20:33,560
Speaker 2: right tell us about what we've learned. SpaceX's attitude is

359
00:20:33,640 --> 00:20:39,439
Speaker 2: to those satellite companies coming to them wanting either a dedicated.

360
00:20:38,840 --> 00:20:41,639
Speaker 3: Ride on Falcon nine or a ride share ride. What

361
00:20:41,680 --> 00:20:42,040
Speaker 3: do we know.

362
00:20:43,320 --> 00:20:45,760
Speaker 11: So what we know and what we've been hearing from

363
00:20:46,119 --> 00:20:51,280
Speaker 11: sources and customers is essentially that SpaceX is fully booked

364
00:20:51,320 --> 00:20:54,440
Speaker 11: through twenty twenty eight and they're turning away customers for

365
00:20:55,000 --> 00:21:00,639
Speaker 11: dedicated Falcon nine launches past that time point. A dedicated

366
00:21:00,720 --> 00:21:03,639
Speaker 11: launch is when a satellite company will buy the entire

367
00:21:03,760 --> 00:21:06,800
Speaker 11: rocket and launch multiple of their satellites to orbit, and

368
00:21:06,800 --> 00:21:10,040
Speaker 11: they're also turning away customers for their ride share missions,

369
00:21:10,080 --> 00:21:13,919
Speaker 11: which is when multiple satellite operators can all, you know,

370
00:21:13,960 --> 00:21:16,040
Speaker 11: put one or two satellites and hitch a ride to

371
00:21:16,200 --> 00:21:19,119
Speaker 11: orbit to test satellites, or if they only need a

372
00:21:19,160 --> 00:21:22,720
Speaker 11: couple to get to space. And yeah, this is a

373
00:21:22,760 --> 00:21:25,440
Speaker 11: really big deal. For a long time, Falcon nine has

374
00:21:25,680 --> 00:21:30,520
Speaker 11: commanded a near monopoly on the launch industry. It launches

375
00:21:30,560 --> 00:21:33,560
Speaker 11: more than any other rocket in the world, so it's

376
00:21:33,600 --> 00:21:35,919
Speaker 11: a it's a pretty big deal that they're you know,

377
00:21:36,240 --> 00:21:37,920
Speaker 11: halting these reservations for now.

378
00:21:38,520 --> 00:21:40,960
Speaker 2: Son of the background point, not just in our reporting

379
00:21:41,040 --> 00:21:44,080
Speaker 2: but tonight as well with the test flight is Starship

380
00:21:44,119 --> 00:21:47,359
Speaker 2: Matters very quickly explain why Starship matters so much.

381
00:21:48,480 --> 00:21:52,760
Speaker 11: Starship matters because it's key to Musk's ambitions for SpaceX,

382
00:21:52,840 --> 00:21:56,360
Speaker 11: which are to you know, build data centers in space.

383
00:21:56,840 --> 00:22:00,840
Speaker 11: It's key to expanding the Starlink communications network, and it

384
00:22:00,960 --> 00:22:04,920
Speaker 11: also will be the vehicle that is intended to land

385
00:22:05,000 --> 00:22:06,320
Speaker 11: humans on the moon in Mars.

386
00:22:07,200 --> 00:22:10,720
Speaker 2: Bloomberg Stanta Passionka really top recording. Thank you very much.

387
00:22:17,000 --> 00:22:18,800
Speaker 3: Hey guys, welcome back to Bloomberg Tech.

388
00:22:18,840 --> 00:22:21,399
Speaker 2: It's been a jittery week, at least through the lens

389
00:22:21,520 --> 00:22:25,160
Speaker 2: of Technology markets then ASDAK one hundred on a five

390
00:22:25,240 --> 00:22:27,439
Speaker 2: day basis or over the course of the trading week,

391
00:22:27,600 --> 00:22:29,960
Speaker 2: it's down about eight tens to one percent, but it's

392
00:22:30,000 --> 00:22:33,119
Speaker 2: actually on track for two straight weeks of declines, something

393
00:22:33,160 --> 00:22:35,959
Speaker 2: that hasn't happened since the end of March, and the

394
00:22:36,000 --> 00:22:39,640
Speaker 2: index itself at its lowest level since the first week

395
00:22:39,640 --> 00:22:42,360
Speaker 2: of May. A lot of that's the earning story. There

396
00:22:42,359 --> 00:22:45,080
Speaker 2: has been a lot in the world of AI about

397
00:22:46,040 --> 00:22:50,959
Speaker 2: security incidents, about the debate on closed versus open, and

398
00:22:51,000 --> 00:22:54,320
Speaker 2: a lot of industry moves as well. CAPEX in AI

399
00:22:54,359 --> 00:22:57,560
Speaker 2: infrastructure has been a dominant theme, but it really is

400
00:22:57,600 --> 00:23:00,639
Speaker 2: on the AI and software side where we folk. Major

401
00:23:00,680 --> 00:23:04,440
Speaker 2: tech leaders took to social media to tout the benefits

402
00:23:04,440 --> 00:23:08,600
Speaker 2: of open weight AI models in a letter signed by Nvidia, Meta, Microsoft,

403
00:23:08,640 --> 00:23:12,080
Speaker 2: indrees and Horowitz, and many others. This comes as moonshots

404
00:23:12,119 --> 00:23:15,480
Speaker 2: Kimmy K three has been shaking the AI industry just

405
00:23:15,520 --> 00:23:20,920
Speaker 2: this week, spurring discourse on China potentially using US technology

406
00:23:21,400 --> 00:23:24,159
Speaker 2: to get ahead in the AI race. Michelle Guide, a

407
00:23:24,280 --> 00:23:27,119
Speaker 2: CEO of the Crack Institute for Tech Diplomacy at Purdue

408
00:23:27,200 --> 00:23:30,200
Speaker 2: and also a former Assistant Secretary of State for Global

409
00:23:30,200 --> 00:23:34,200
Speaker 2: Public Affairs under the first Trump administration, rights the US

410
00:23:34,320 --> 00:23:37,120
Speaker 2: won't be able to win the AI race simply by

411
00:23:37,200 --> 00:23:41,399
Speaker 2: limiting China's technology and joins US now, And frankly, that's

412
00:23:41,720 --> 00:23:43,760
Speaker 2: where the debate is. You know, you will have seen

413
00:23:43,760 --> 00:23:47,119
Speaker 2: the headlines this morning and the open letter essentially on

414
00:23:47,200 --> 00:23:50,520
Speaker 2: open weight the timing of that, there must be a

415
00:23:50,560 --> 00:23:51,360
Speaker 2: reason for it.

416
00:23:53,240 --> 00:23:55,560
Speaker 8: Well, what I think you're actually seeing ed is a

417
00:23:55,600 --> 00:23:59,439
Speaker 8: really good example of Silicon Valley in Washington, DC talking

418
00:23:59,480 --> 00:24:01,920
Speaker 8: past each each other, because if you look at that letter,

419
00:24:01,960 --> 00:24:06,080
Speaker 8: the real debate isn't about open weight models. It's about

420
00:24:06,200 --> 00:24:10,840
Speaker 8: Chinese open weight models, which unfortunately are the most available,

421
00:24:11,080 --> 00:24:14,239
Speaker 8: most price effective option that's on the table right now,

422
00:24:14,240 --> 00:24:17,719
Speaker 8: and especially for a lot of new businesses and startups

423
00:24:18,320 --> 00:24:20,639
Speaker 8: that are trying to build their AI stacks. And so

424
00:24:21,040 --> 00:24:24,000
Speaker 8: the real debate is on Chinese open weight models. And

425
00:24:24,040 --> 00:24:27,560
Speaker 8: the problem there is we know that presents national security risk,

426
00:24:27,640 --> 00:24:30,159
Speaker 8: we know it presents corporate risk given all that we

427
00:24:30,160 --> 00:24:34,040
Speaker 8: know about Chinese technology being untrustworthy. It's why we've banned Huawei,

428
00:24:34,240 --> 00:24:36,680
Speaker 8: It's why we banned and then had to restructure TikTok,

429
00:24:36,880 --> 00:24:39,240
Speaker 8: It's why there's legislation moving through the House right now

430
00:24:39,240 --> 00:24:43,840
Speaker 8: about banning connected vehicles that are coming from China. And

431
00:24:43,880 --> 00:24:46,720
Speaker 8: so there's a pattern here, and I think the focus

432
00:24:46,800 --> 00:24:49,560
Speaker 8: on what do we ban versus what do we keep

433
00:24:49,600 --> 00:24:53,160
Speaker 8: open is actually misplaced focus by both the private sector

434
00:24:53,160 --> 00:24:56,119
Speaker 8: and the US government. The focus should be on how

435
00:24:56,160 --> 00:25:02,199
Speaker 8: do we turbocharge America's open weight ecosystem have a world class, robust,

436
00:25:02,640 --> 00:25:05,200
Speaker 8: really price effective offering for the rest of the world,

437
00:25:05,480 --> 00:25:08,120
Speaker 8: so we can diffuse American AI as fast as possible,

438
00:25:08,280 --> 00:25:10,920
Speaker 8: and not only overseas, but here at home where businesses

439
00:25:10,960 --> 00:25:11,440
Speaker 8: are trying.

440
00:25:11,240 --> 00:25:13,520
Speaker 3: To wri on the here at home bit.

441
00:25:13,640 --> 00:25:17,120
Speaker 2: You know, I've read the open letter as many times

442
00:25:17,200 --> 00:25:19,879
Speaker 2: I could before we came to air, and I'm thinking,

443
00:25:19,920 --> 00:25:23,080
Speaker 2: what is the concern what catalyzed them writing it and

444
00:25:23,160 --> 00:25:27,480
Speaker 2: putting it out. One take is the concern that Washington

445
00:25:27,680 --> 00:25:32,120
Speaker 2: just overregulates open source models right in a way that

446
00:25:33,000 --> 00:25:36,199
Speaker 2: is detrimental to American interests. Where do you sort of

447
00:25:36,200 --> 00:25:37,640
Speaker 2: sit on that debate?

448
00:25:38,880 --> 00:25:42,119
Speaker 8: Yeah, well, I think it's because open weight models, cheaper

449
00:25:42,240 --> 00:25:44,640
Speaker 8: open weight models have become really pore to a lot

450
00:25:44,920 --> 00:25:47,840
Speaker 8: of how businesses are developing their AI stacks. As you

451
00:25:47,840 --> 00:25:49,640
Speaker 8: mentioned earlier, we just saw a big tech wipe out

452
00:25:49,960 --> 00:25:53,520
Speaker 8: eight hundred and ninety billion dollars because AI is really expensive.

453
00:25:53,560 --> 00:25:55,919
Speaker 8: Everybody's looking to see if we can keep up with

454
00:25:55,960 --> 00:25:59,280
Speaker 8: this spending bubble, and so the cheaper, good enough versions

455
00:26:00,080 --> 00:26:02,480
Speaker 8: are really important for a private sector to be building

456
00:26:02,480 --> 00:26:05,840
Speaker 8: their aistacks. The problem is it comes from an adversary,

457
00:26:06,359 --> 00:26:10,359
Speaker 8: and there is no really robust, us trusted alternative to

458
00:26:10,400 --> 00:26:12,399
Speaker 8: the Chinese open weight models that we're seeing. And then

459
00:26:12,400 --> 00:26:16,199
Speaker 8: the Kim three launch this week catalyzed our awareness of that,

460
00:26:16,480 --> 00:26:19,600
Speaker 8: and so the debate now is not really just open weight,

461
00:26:19,600 --> 00:26:21,640
Speaker 8: it's Chinese open weight and can we get an American

462
00:26:21,640 --> 00:26:23,400
Speaker 8: alternative out there fast?

463
00:26:23,920 --> 00:26:26,880
Speaker 2: So nvidiacre Gens Wong did it sort of an extended

464
00:26:26,920 --> 00:26:30,920
Speaker 2: interview of Axios and basically said, you know, the top

465
00:26:31,000 --> 00:26:35,040
Speaker 2: lines are the Chinese models are excellent, and that the

466
00:26:35,080 --> 00:26:38,919
Speaker 2: open source models that are excellent to should be used.

467
00:26:38,960 --> 00:26:41,080
Speaker 2: And so you know, based on your line of argument

468
00:26:41,119 --> 00:26:43,560
Speaker 2: that the risk is too great if the open model

469
00:26:43,600 --> 00:26:47,760
Speaker 2: comes from an economic adversary. It comes down to how

470
00:26:47,800 --> 00:26:50,600
Speaker 2: influential is Jensen Wong with this administration.

471
00:26:52,840 --> 00:26:55,119
Speaker 8: Well, I actually think it comes down to how fast

472
00:26:55,160 --> 00:26:59,119
Speaker 8: can we turbocharge an American open weight ecosystem. Look all

473
00:26:59,160 --> 00:27:01,760
Speaker 8: the things that he said and other Silicon Valley leaders

474
00:27:01,800 --> 00:27:05,240
Speaker 8: are saying, makes sense if you're looking at this purely

475
00:27:05,320 --> 00:27:07,879
Speaker 8: through an innovation and a commercial lens, It makes a

476
00:27:07,880 --> 00:27:10,840
Speaker 8: world of sense. If you then factor in the communist

477
00:27:10,880 --> 00:27:14,160
Speaker 8: adversary lens, the risk calculus looks a whole lot different.

478
00:27:14,160 --> 00:27:16,720
Speaker 8: And that's why these technologies are different. Look, we just

479
00:27:16,760 --> 00:27:20,760
Speaker 8: spent at the better part of a decade with businesses

480
00:27:20,800 --> 00:27:23,560
Speaker 8: thinking about how they decouple and they de risk from

481
00:27:23,600 --> 00:27:26,120
Speaker 8: China because of the supply chain risk, the financial risk,

482
00:27:26,200 --> 00:27:29,640
Speaker 8: the corporate risk, the national security risk. And now we're

483
00:27:29,640 --> 00:27:34,720
Speaker 8: talking about entrenching Chinese technology into the very foundation of

484
00:27:34,760 --> 00:27:38,840
Speaker 8: our American company's AI stacks like that makes no sense,

485
00:27:39,520 --> 00:27:41,920
Speaker 8: And so how do we focus instead on getting trusted

486
00:27:41,960 --> 00:27:47,560
Speaker 8: American AI, cheap, effective, price effective AI into the hands

487
00:27:47,560 --> 00:27:48,880
Speaker 8: of American companies and as.

488
00:27:48,800 --> 00:27:49,920
Speaker 6: Much of the world as possible.

489
00:27:50,119 --> 00:27:51,439
Speaker 8: That's where the focus needs to go.

490
00:27:53,080 --> 00:27:54,159
Speaker 3: Let's go back to the beginning.

491
00:27:54,240 --> 00:27:56,960
Speaker 2: You know, your argument through the lens of policy research

492
00:27:57,080 --> 00:27:59,879
Speaker 2: reflecting on your time in government, is that at the

493
00:28:00,119 --> 00:28:02,880
Speaker 2: end of the day, America won't win the AI race

494
00:28:03,000 --> 00:28:07,520
Speaker 2: by restricting China. So you indicate that America needs to

495
00:28:07,760 --> 00:28:10,640
Speaker 2: be proactive in its own approach, get its own house

496
00:28:10,680 --> 00:28:15,640
Speaker 2: in order. How far does today's action go? What else

497
00:28:15,760 --> 00:28:16,480
Speaker 2: needs to be done?

498
00:28:17,560 --> 00:28:19,879
Speaker 8: Yeah, I think Look, the President has this great group

499
00:28:19,920 --> 00:28:23,359
Speaker 8: on his pe cast to this Presidential Advisory Board with

500
00:28:23,359 --> 00:28:25,399
Speaker 8: a bunch of technology leaders. I think they should get

501
00:28:25,400 --> 00:28:28,200
Speaker 8: in a room as quickly as possible and figure out

502
00:28:28,240 --> 00:28:32,080
Speaker 8: how they go on offense really quickly with American open

503
00:28:32,119 --> 00:28:34,320
Speaker 8: weight AI. And by the way, get all of our

504
00:28:34,359 --> 00:28:37,359
Speaker 8: allies on board. Because even if we ban and limit

505
00:28:37,600 --> 00:28:41,160
Speaker 8: the use of Chinese technology Chinese AI models here, if

506
00:28:41,200 --> 00:28:44,280
Speaker 8: we're using AI models from America here, but the rest

507
00:28:44,280 --> 00:28:46,360
Speaker 8: of the world is running on a Chinese AI stack,

508
00:28:46,560 --> 00:28:48,680
Speaker 8: that's still a problem for us, and so I think

509
00:28:48,680 --> 00:28:50,240
Speaker 8: they can rally round that. And look, we have a

510
00:28:50,280 --> 00:28:52,840
Speaker 8: lot of lessons. When I was at the State Department,

511
00:28:53,120 --> 00:28:56,520
Speaker 8: Huawei was a threat across the world and there's a

512
00:28:56,520 --> 00:28:59,080
Speaker 8: lot of lessons to be learned there. And a big

513
00:28:59,200 --> 00:29:01,840
Speaker 8: lesson is that we're not gonna win on principle, We're

514
00:29:01,840 --> 00:29:04,400
Speaker 8: gonna win on price. So how do we get much

515
00:29:04,440 --> 00:29:08,840
Speaker 8: more cost effective trusted American AI open weight models out

516
00:29:08,920 --> 00:29:09,480
Speaker 8: to the world.

517
00:29:09,840 --> 00:29:14,560
Speaker 2: Now, we're showing a post from mister or director Cratzios

518
00:29:14,600 --> 00:29:16,800
Speaker 2: from a couple of days ago. We covered that story,

519
00:29:16,840 --> 00:29:20,400
Speaker 2: the accusation that Kimmy K three was distilled from an

520
00:29:20,400 --> 00:29:25,480
Speaker 2: anthropic model and used illegally in Vidia Blackwell Systems. We've

521
00:29:25,480 --> 00:29:27,880
Speaker 2: been over that, but just pointing that out that mister

522
00:29:27,920 --> 00:29:30,760
Speaker 2: Cratzios is also the co chair of PEACAS, along with

523
00:29:30,840 --> 00:29:34,280
Speaker 2: David Sachs. Jensen's on PEA cast for example, a Michelle Guide,

524
00:29:34,280 --> 00:29:36,920
Speaker 2: the CEO at the Crack Institute for Tech Diplomacy at

525
00:29:36,920 --> 00:29:38,840
Speaker 2: Purdue Back on Bloomberg Tech, Thank you.

526
00:29:38,840 --> 00:29:39,520
Speaker 3: Very much so.

527
00:29:39,600 --> 00:29:42,959
Speaker 2: Coming up, the AI race also shifts to speed AMD

528
00:29:43,080 --> 00:29:46,320
Speaker 2: and Cerebras and veil a new partnership and Cerebra CEO

529
00:29:46,360 --> 00:29:47,840
Speaker 2: Andrew Feldman joins us.

530
00:29:47,760 --> 00:29:50,680
Speaker 3: Next fast Inference. This is Bloomberg Tech.

531
00:30:03,360 --> 00:30:07,560
Speaker 2: Shares of Cerebras Systems that down about ten percent right now.

532
00:30:07,640 --> 00:30:12,040
Speaker 2: They had jumped yesterday on some news that AMD is

533
00:30:12,160 --> 00:30:15,480
Speaker 2: teaming up with Cerebra Systems on a new server designed

534
00:30:15,480 --> 00:30:20,160
Speaker 2: to slash response times, taking direct aim at Nvidia. Joining

535
00:30:20,240 --> 00:30:22,840
Speaker 2: us to explain the deal, the technology Andrew Feldman, co

536
00:30:22,880 --> 00:30:24,840
Speaker 2: founder and CEO of Cerebras Systems.

537
00:30:24,960 --> 00:30:27,240
Speaker 3: So this is how it's going to work. You basically have.

538
00:30:27,320 --> 00:30:34,600
Speaker 2: A Helio server and a Cerebras server in combination. How

539
00:30:34,680 --> 00:30:37,520
Speaker 2: is that going to work? And what is the sort

540
00:30:37,520 --> 00:30:38,320
Speaker 2: of split.

541
00:30:38,080 --> 00:30:39,280
Speaker 3: On the workload?

542
00:30:40,000 --> 00:30:42,440
Speaker 12: Sure, good to be back, Good, thanks for having me.

543
00:30:44,160 --> 00:30:46,640
Speaker 12: The way to think about it is that the inference

544
00:30:46,800 --> 00:30:51,400
Speaker 12: problem is comprised of two parts. We call the first

545
00:30:51,440 --> 00:30:55,200
Speaker 12: part processing the prompt, and we call the second part

546
00:30:55,240 --> 00:30:56,120
Speaker 12: generating the answer.

547
00:30:56,920 --> 00:30:58,240
Speaker 10: And those two parts of.

548
00:30:58,240 --> 00:31:02,680
Speaker 12: The problem have very different cops mutational requirements, and that

549
00:31:03,040 --> 00:31:06,240
Speaker 12: opens the door to address them with two different machines.

550
00:31:07,200 --> 00:31:11,560
Speaker 12: The processing of the prompt the analyzing the query.

551
00:31:12,360 --> 00:31:14,920
Speaker 10: That part is a problem.

552
00:31:14,920 --> 00:31:18,680
Speaker 12: That can be parallelized, and for that type of work,

553
00:31:19,200 --> 00:31:24,240
Speaker 12: GPUs are are very very good, and Helios will be

554
00:31:24,320 --> 00:31:29,280
Speaker 12: the best. The second part of the work, which is

555
00:31:29,400 --> 00:31:34,320
Speaker 12: generating the answer at extraordinary speeds, Cerebris is the best

556
00:31:34,520 --> 00:31:37,800
Speaker 12: in the world, bar none. And so by bringing these

557
00:31:37,800 --> 00:31:44,080
Speaker 12: two solutions together so that the Helios processes the prompt

558
00:31:44,320 --> 00:31:49,400
Speaker 12: Cerebris generates the answer, we create a single inference flow

559
00:31:50,280 --> 00:31:52,520
Speaker 12: that is the fastest in the world and has this

560
00:31:52,680 --> 00:31:54,040
Speaker 12: extraordinary throughput.

561
00:31:55,960 --> 00:31:59,400
Speaker 2: You know, Andrew, I'm not trying to I'm not trying

562
00:31:59,440 --> 00:32:02,280
Speaker 2: to make a scandal or a negative out of it,

563
00:32:02,320 --> 00:32:04,800
Speaker 2: but like, for you, like, how does the sales channel work?

564
00:32:04,880 --> 00:32:05,000
Speaker 3: Right?

565
00:32:05,040 --> 00:32:09,080
Speaker 2: So AMD has like this pipeline of projects and those

566
00:32:09,800 --> 00:32:12,200
Speaker 2: that infrastructure is like, okay, we're gonna use Helios systems,

567
00:32:12,240 --> 00:32:16,720
Speaker 2: but now they can also get the Cerebrius server alongside it.

568
00:32:17,200 --> 00:32:19,720
Speaker 2: How have you guys agreed to split the revenues? How

569
00:32:19,720 --> 00:32:21,920
Speaker 2: have you guys agreed to do this in a way

570
00:32:21,920 --> 00:32:27,240
Speaker 2: that is economically equitable? You know, it's a really interesting deal.

571
00:32:28,000 --> 00:32:33,200
Speaker 12: Sure, I think speed makes markets bigger, right, it doesn't

572
00:32:33,280 --> 00:32:36,480
Speaker 12: make markets smaller. This isn't about carving up something that's

573
00:32:36,480 --> 00:32:41,360
Speaker 12: a fixed size, right. Fast fast inference is productive inference,

574
00:32:41,600 --> 00:32:44,200
Speaker 12: and where AI is productive, people are willing to spend

575
00:32:44,240 --> 00:32:49,719
Speaker 12: more and more and more. And so the first application

576
00:32:50,160 --> 00:32:53,200
Speaker 12: of this solution will be in the Cerebrace cloud and

577
00:32:53,240 --> 00:32:55,760
Speaker 12: that will be later this year, and shortly thereafter it

578
00:32:55,800 --> 00:33:00,239
Speaker 12: will be available more generally. I think we have an

579
00:33:00,320 --> 00:33:04,360
Speaker 12: enormous backlog. They have an enormous backlog. I think we

580
00:33:04,480 --> 00:33:10,600
Speaker 12: have customers around the world where who are demanding fast,

581
00:33:10,920 --> 00:33:16,640
Speaker 12: fast inferences, extraordinary scale, and the solution we're building is

582
00:33:16,680 --> 00:33:18,280
Speaker 12: head and shoulders above anything else.

583
00:33:19,560 --> 00:33:22,000
Speaker 3: I find the technology approach like really interesting. You know.

584
00:33:22,080 --> 00:33:26,120
Speaker 2: The parallel example is in Nvidia and Grock with a

585
00:33:26,200 --> 00:33:30,200
Speaker 2: Q and Grock three LPX and like in that case

586
00:33:30,240 --> 00:33:36,560
Speaker 2: it's integration of the system into the same server design. Right,

587
00:33:37,240 --> 00:33:39,800
Speaker 2: you didn't take that approach, did you talk about that? Like,

588
00:33:40,200 --> 00:33:42,080
Speaker 2: you know, what were the options on the table?

589
00:33:43,320 --> 00:33:48,200
Speaker 12: Well, we don't quite know what what what in Nvidia

590
00:33:48,320 --> 00:33:50,760
Speaker 12: is doing that they haven't really delivered to market yet,

591
00:33:50,800 --> 00:33:54,320
Speaker 12: and so we we are interested to see how that

592
00:33:54,360 --> 00:33:58,760
Speaker 12: works out. What we were able to do because we

593
00:33:58,880 --> 00:34:04,840
Speaker 12: use open standards based technology for our io, we were

594
00:34:04,880 --> 00:34:09,360
Speaker 12: able to build a solution with a m D quickly

595
00:34:09,840 --> 00:34:15,160
Speaker 12: and easily because we sort of support open standards.

596
00:34:15,160 --> 00:34:16,480
Speaker 10: Remember, we've also.

597
00:34:16,280 --> 00:34:21,800
Speaker 12: Done this with with tranium from AWS, and so we

598
00:34:22,120 --> 00:34:25,600
Speaker 12: adopted the same approach that by being open, by being

599
00:34:25,600 --> 00:34:31,520
Speaker 12: standards based, right half the leading chip makers a m

600
00:34:31,600 --> 00:34:37,960
Speaker 12: D AWS with tranium are now using our solution in

601
00:34:38,040 --> 00:34:41,360
Speaker 12: a disaggregated approach for high.

602
00:34:41,200 --> 00:34:42,560
Speaker 10: Speed inference delivery.

603
00:34:43,280 --> 00:34:47,360
Speaker 12: So I think the answer is by being open and

604
00:34:47,400 --> 00:34:53,040
Speaker 12: by being standards based, we can rapidly partner with other

605
00:34:53,120 --> 00:34:56,080
Speaker 12: members of the ecosystem.

606
00:34:56,120 --> 00:34:58,359
Speaker 2: You've also been a busy guy. We were talking off

607
00:34:58,719 --> 00:35:01,759
Speaker 2: air about how busy everything is. You know, you just

608
00:35:01,760 --> 00:35:04,719
Speaker 2: did a very big IPO. Kind of interesting to know,

609
00:35:04,800 --> 00:35:07,120
Speaker 2: like who approached to whose idea was this?

610
00:35:08,520 --> 00:35:11,520
Speaker 12: Look, I've known Lisa for a long time. She was

611
00:35:11,600 --> 00:35:14,800
Speaker 12: CEO when when my last company was acquired by a

612
00:35:14,960 --> 00:35:19,680
Speaker 12: m D. I've watched sort of her do extraordinary things

613
00:35:19,719 --> 00:35:24,800
Speaker 12: with that company. The returns over her her tenure CEO

614
00:35:25,200 --> 00:35:28,920
Speaker 12: or are mind boggling. And so we are in constant communication.

615
00:35:29,320 --> 00:35:32,799
Speaker 12: We are close with her CTO Marked paper Master. These

616
00:35:32,840 --> 00:35:36,080
Speaker 12: are people we've known for decades, and so we're in

617
00:35:36,120 --> 00:35:40,719
Speaker 12: constant communication talking about how we might collaborate, what we

618
00:35:40,800 --> 00:35:45,400
Speaker 12: might do together. And this was an idea that emerged

619
00:35:45,440 --> 00:35:48,240
Speaker 12: from those discussions and made perfect sense.

620
00:35:49,800 --> 00:35:52,640
Speaker 2: In the Nvidia grock with a Q example, which I appreciate.

621
00:35:52,880 --> 00:35:55,560
Speaker 2: Not your companies like completely separate, but you know they

622
00:35:55,719 --> 00:35:59,960
Speaker 2: the mechanism was an aquiahiher as you know, prior to

623
00:36:00,120 --> 00:36:03,960
Speaker 2: your IPO, we reported at Bloomberg that some of the

624
00:36:03,960 --> 00:36:08,960
Speaker 2: fabulous chip names held talks or interest expressed interest in

625
00:36:09,440 --> 00:36:14,160
Speaker 2: also acquiring Cerebris prior to its listing. Did you discuss

626
00:36:14,200 --> 00:36:17,080
Speaker 2: any of those kind of mechanisms with Lisa about the

627
00:36:17,520 --> 00:36:20,520
Speaker 2: merits of her making an equity investment or some of

628
00:36:20,560 --> 00:36:24,160
Speaker 2: the structures that AMD has deployed in other of their arrangements.

629
00:36:25,000 --> 00:36:29,319
Speaker 12: Sure, MD has been an equity investor for quite some time.

630
00:36:29,719 --> 00:36:34,520
Speaker 12: They were an investor in some of our mid and

631
00:36:34,600 --> 00:36:39,759
Speaker 12: later stage rounds, and so we have been talking with

632
00:36:39,800 --> 00:36:44,080
Speaker 12: them and engaged in discussions on how we might work

633
00:36:44,080 --> 00:36:47,040
Speaker 12: together for a long time, as we are with with

634
00:36:47,160 --> 00:36:49,560
Speaker 12: many of the players. And that's the advantage of being open,

635
00:36:49,680 --> 00:36:52,799
Speaker 12: that's the advantage of not being a walled garden. That's

636
00:36:52,800 --> 00:36:58,640
Speaker 12: the advantage of not having a proprietary iotechnology but using

637
00:36:58,760 --> 00:37:05,640
Speaker 12: standards based high speed ethernet technology. We can engage with AMD,

638
00:37:05,840 --> 00:37:08,760
Speaker 12: we can engage with AWS, we could engage with Google,

639
00:37:08,800 --> 00:37:13,840
Speaker 12: We can engage with anybody who builds a part that

640
00:37:13,880 --> 00:37:16,080
Speaker 12: would integrate into an extraordinary solution.

641
00:37:17,320 --> 00:37:18,200
Speaker 3: Andrew really quickly.

642
00:37:18,280 --> 00:37:21,440
Speaker 2: I had a long conversation with sk Group chair Uan

643
00:37:21,800 --> 00:37:24,760
Speaker 2: very recently, and he said, basically, the difference between China's

644
00:37:24,760 --> 00:37:27,120
Speaker 2: approach to AI and the US is China's going for

645
00:37:27,440 --> 00:37:30,200
Speaker 2: the lowest dollar per token, whereas America is still in

646
00:37:30,200 --> 00:37:32,680
Speaker 2: a place where it's focused on the highest quality tokens.

647
00:37:33,239 --> 00:37:36,200
Speaker 2: From the non HBM standpoint, would you kind of weigh

648
00:37:36,200 --> 00:37:36,880
Speaker 2: in on that.

649
00:37:37,800 --> 00:37:40,960
Speaker 12: Yeah, right, So we don't use HBM, and that's one

650
00:37:40,960 --> 00:37:44,160
Speaker 12: of the real advantages we have, so we can generate

651
00:37:44,200 --> 00:37:49,080
Speaker 12: tokens for less because we're not dependent on this sort

652
00:37:49,120 --> 00:37:52,920
Speaker 12: of supply chain constrained part.

653
00:37:53,960 --> 00:37:54,160
Speaker 3: Right.

654
00:37:54,840 --> 00:38:00,879
Speaker 12: I think it's unclear whether the Chinese motel makers are

655
00:38:00,920 --> 00:38:05,239
Speaker 12: actually developing for less or piggybacking on technology that other

656
00:38:05,280 --> 00:38:10,360
Speaker 12: people have invented. I think that's an open question what

657
00:38:10,440 --> 00:38:12,839
Speaker 12: I don't have the answer to. But I know there's

658
00:38:12,880 --> 00:38:17,480
Speaker 12: some very strongly held views that they are stealing distilling.

659
00:38:18,880 --> 00:38:19,600
Speaker 10: I'm not sure.

660
00:38:19,880 --> 00:38:22,560
Speaker 12: But what I do know is I don't understand their

661
00:38:22,560 --> 00:38:26,600
Speaker 12: cost to build these models. Open AI is one of

662
00:38:26,640 --> 00:38:29,640
Speaker 12: our larger customers. We do have an understanding of the

663
00:38:29,640 --> 00:38:31,520
Speaker 12: amount of compute they need to build, and we have

664
00:38:31,640 --> 00:38:34,960
Speaker 12: an understanding of the amount of compute that other frontier

665
00:38:35,080 --> 00:38:38,160
Speaker 12: US labs have, we don't really have an understanding of

666
00:38:38,200 --> 00:38:40,760
Speaker 12: what the cost is. So the question is, really, are

667
00:38:40,840 --> 00:38:45,359
Speaker 12: are they making lower cost AI because they're borrowing other

668
00:38:45,360 --> 00:38:49,759
Speaker 12: people's technology or do they have some really interesting inventions.

669
00:38:49,239 --> 00:38:52,920
Speaker 10: That allow them to make AI for less.

670
00:38:53,680 --> 00:38:55,839
Speaker 12: That's an open question and one that relates to your

671
00:38:55,840 --> 00:38:59,000
Speaker 12: previous segment and what we should do about that if

672
00:38:59,000 --> 00:38:59,800
Speaker 12: it is the case.

673
00:39:00,760 --> 00:39:03,200
Speaker 2: And very very quickly you've seen the less of this

674
00:39:03,239 --> 00:39:07,600
Speaker 2: morning leaders in US technology emphasizing America's need to focus

675
00:39:07,640 --> 00:39:08,319
Speaker 2: on open weight.

676
00:39:08,680 --> 00:39:09,879
Speaker 3: Do you have a viewpoint on that?

677
00:39:11,200 --> 00:39:14,080
Speaker 12: Look, I think right now the market's been made by

678
00:39:14,280 --> 00:39:19,319
Speaker 12: open ai with a fast follow by Anthropic, and they

679
00:39:19,360 --> 00:39:22,640
Speaker 12: have invested an enormous amount of money and mind bottling

680
00:39:22,640 --> 00:39:25,760
Speaker 12: amount of money, and that they have made this market.

681
00:39:26,480 --> 00:39:33,440
Speaker 12: And I think the followers, both from from China and others,

682
00:39:33,600 --> 00:39:36,400
Speaker 12: are are trying to capture this with a fast follow strategy,

683
00:39:37,239 --> 00:39:40,400
Speaker 12: and I think it's it's unclear to me what the

684
00:39:41,000 --> 00:39:45,040
Speaker 12: right approach is. I think obviously we want lower cost AI.

685
00:39:45,800 --> 00:39:50,279
Speaker 12: What we can control at Cerebras is by building like

686
00:39:50,320 --> 00:39:55,080
Speaker 12: we're doing with a MD solutions that deliver more tokens

687
00:39:55,680 --> 00:39:59,320
Speaker 12: per unit power, that deliver more tokens per dollar, that

688
00:39:59,440 --> 00:40:03,640
Speaker 12: deliver those tokens faster, so they're more productive, so you

689
00:40:03,760 --> 00:40:04,920
Speaker 12: can pay more. Right.

690
00:40:04,960 --> 00:40:06,960
Speaker 10: The problem isn't the tokens are expensive.

691
00:40:07,760 --> 00:40:11,000
Speaker 12: The problem is that it's hard to measure how much

692
00:40:11,040 --> 00:40:14,200
Speaker 12: productivity you're getting from tokens. And what we know is

693
00:40:14,239 --> 00:40:17,279
Speaker 12: that when you make things faster, they drive up productivity.

694
00:40:17,680 --> 00:40:20,200
Speaker 12: And so these are the dimensions we can work on

695
00:40:20,440 --> 00:40:23,440
Speaker 12: ED and we're we're diligently working on them every day.

696
00:40:24,280 --> 00:40:27,279
Speaker 2: Andrew Feldman Threubus Systems co founder and CEO, thank you

697
00:40:27,400 --> 00:40:27,799
Speaker 2: very much.

698
00:40:29,160 --> 00:40:30,280
Speaker 3: With AI stock storing.

699
00:40:30,360 --> 00:40:33,839
Speaker 2: This year, Viking Global Investors told clients that its conservative

700
00:40:33,880 --> 00:40:36,680
Speaker 2: position on the sector was a quote missed opportunity. The

701
00:40:36,680 --> 00:40:39,720
Speaker 2: firm's flagship hedge fund gained just two point six percent

702
00:40:39,760 --> 00:40:42,000
Speaker 2: in the first half of the year, trailing piers with

703
00:40:42,160 --> 00:40:45,120
Speaker 2: greater AI exposure like CO two, and yet the firm

704
00:40:45,280 --> 00:40:48,640
Speaker 2: doesn't plan to change course. Bloomberg's Hema Palmer broke the story.

705
00:40:48,960 --> 00:40:51,759
Speaker 2: So they're sticking to their guns, but the performance is

706
00:40:51,800 --> 00:40:52,880
Speaker 2: trailing their peers.

707
00:40:53,200 --> 00:40:55,840
Speaker 13: Yeah, so really interesting stance from Viking Global. This is

708
00:40:55,840 --> 00:40:59,239
Speaker 13: a hedge fund that manages fifty six billion dollars and

709
00:40:59,320 --> 00:41:02,520
Speaker 13: they're taking a divergent view from what we're seeing from

710
00:41:02,560 --> 00:41:05,200
Speaker 13: many of their peers that are really jumping on this

711
00:41:05,320 --> 00:41:09,480
Speaker 13: AI beta market. And meanwhile, co Tow really a lot

712
00:41:09,520 --> 00:41:12,600
Speaker 13: more cautious, not partaking in a lot of these stocks.

713
00:41:12,840 --> 00:41:15,280
Speaker 13: And we're seeing this in their performance with their hedge

714
00:41:15,280 --> 00:41:18,640
Speaker 13: fund up about seven point five in the second quarter,

715
00:41:18,920 --> 00:41:21,319
Speaker 13: but are still only up two point six for the

716
00:41:21,440 --> 00:41:25,520
Speaker 13: full year so far, and that's a great degree less

717
00:41:25,719 --> 00:41:27,000
Speaker 13: than many of their competitors.

718
00:41:27,800 --> 00:41:30,719
Speaker 2: So if they're not budging and they acknowledge kind of

719
00:41:30,760 --> 00:41:33,839
Speaker 2: the performance and a missed opportunity, is it that they

720
00:41:33,840 --> 00:41:36,440
Speaker 2: see some opportunity in their current positioning that's going to

721
00:41:36,480 --> 00:41:37,360
Speaker 2: change their fortunes.

722
00:41:37,719 --> 00:41:40,399
Speaker 13: So they're very concerned about the risks that they see

723
00:41:40,400 --> 00:41:43,280
Speaker 13: in the market. So they're concerned that we could see

724
00:41:43,520 --> 00:41:48,080
Speaker 13: potentially a correction, some sort of sudden shock that curbs

725
00:41:48,400 --> 00:41:51,520
Speaker 13: buying of these stocks that are highly popular. They're worried

726
00:41:51,520 --> 00:41:54,600
Speaker 13: about the valuations that they're seeing, so they do partake

727
00:41:54,680 --> 00:41:58,800
Speaker 13: in some AI investments. They are a buyer of Samsung,

728
00:41:58,840 --> 00:42:01,520
Speaker 13: which has worked out really well for them, but a

729
00:42:01,520 --> 00:42:05,040
Speaker 13: lot of their portfolio is and other things like consumer financials,

730
00:42:05,080 --> 00:42:08,320
Speaker 13: industrials and some of those stocks, some of those trades

731
00:42:08,360 --> 00:42:11,480
Speaker 13: really aren't working as well as they would hope. Some

732
00:42:11,560 --> 00:42:15,319
Speaker 13: of them, the firm says to investors, is potentially being

733
00:42:15,600 --> 00:42:19,000
Speaker 13: unfairly considered an AI loser.

734
00:42:20,280 --> 00:42:24,480
Speaker 3: Very quick hemma. Historically good performance of viking or bad performance.

735
00:42:24,200 --> 00:42:27,520
Speaker 13: You know, historically strong performance. And because of their cautious

736
00:42:27,600 --> 00:42:29,759
Speaker 13: tone a couple of years ago in twenty twenty and

737
00:42:29,760 --> 00:42:32,520
Speaker 13: twenty twenty one, that saved them from a lot of

738
00:42:32,560 --> 00:42:35,120
Speaker 13: the troubles we saw amongst their peers in twenty twenty

739
00:42:35,160 --> 00:42:38,200
Speaker 13: two when we saw that stock market crash when it

740
00:42:38,239 --> 00:42:41,480
Speaker 13: came to tech stocks and private valuations, and now it

741
00:42:41,560 --> 00:42:43,600
Speaker 13: saved them. Then investors may be hoping if there's a

742
00:42:43,600 --> 00:42:45,359
Speaker 13: correction in the future, it'll save them.

743
00:42:45,360 --> 00:42:48,319
Speaker 2: Today Blueberg Temo Palmer with a must read thank you

744
00:42:48,440 --> 00:42:51,080
Speaker 2: very much. That does it for this edition of Bloombog Tech.

745
00:42:51,360 --> 00:42:53,719
Speaker 2: What a week More tech earnings coming up next week.

746
00:42:53,760 --> 00:42:55,880
Speaker 2: This is what it looks like recap on the pod.

747
00:42:56,160 --> 00:42:59,359
Speaker 2: I really recommend some of the conversations throughout today's show.

748
00:42:59,440 --> 00:43:00,480
Speaker 3: This is Bloom a tech