1
00:00:00,280 --> 00:00:03,040
Speaker 1: This is Dana Perkins and you're listening to Switched on

2
00:00:03,279 --> 00:00:06,960
Speaker 1: the BNAF podcast. Today we bring you a recording from

3
00:00:06,960 --> 00:00:10,040
Speaker 1: our BNAF summit in San Francisco, which took place on

4
00:00:10,080 --> 00:00:13,440
Speaker 1: the fourth and fifth of February. The panel was titled

5
00:00:13,680 --> 00:00:17,800
Speaker 1: Data Dynamics. Now certainly a hot topic in energy circles

6
00:00:17,880 --> 00:00:20,560
Speaker 1: has been the growth in data centers and how AI

7
00:00:20,880 --> 00:00:23,680
Speaker 1: has led to a rise in demand for power. In

8
00:00:23,800 --> 00:00:26,360
Speaker 1: order to meet this, big tech companies are looking for

9
00:00:26,440 --> 00:00:30,240
Speaker 1: solutions and firms like Google are signing power purchase agreements

10
00:00:30,320 --> 00:00:33,920
Speaker 1: in an array of technologies like nuclear, long duration energy storage,

11
00:00:33,960 --> 00:00:38,520
Speaker 1: and even geothermal. While these data centers are undeniably energy intensive,

12
00:00:38,600 --> 00:00:42,479
Speaker 1: efficiencies can be found when it comes to reducing load requirements.

13
00:00:42,760 --> 00:00:45,480
Speaker 1: This is seen with the recent release of Deepseek's AI

14
00:00:45,560 --> 00:00:48,680
Speaker 1: model in January twenty twenty five. On today's show, the

15
00:00:48,720 --> 00:00:52,839
Speaker 1: panelists discuss the increasing capital expenditure on data centers, as

16
00:00:52,880 --> 00:00:55,400
Speaker 1: well as what has been driving the decision making of

17
00:00:55,440 --> 00:00:58,120
Speaker 1: big tech companies when it comes to this spend and

18
00:00:58,240 --> 00:01:01,720
Speaker 1: how potential bottlenecks might slow them down. The panelists include

19
00:01:01,760 --> 00:01:06,080
Speaker 1: Stephen Carlini, Chief Advocate Data Centers and AI at Schneider Electric,

20
00:01:06,400 --> 00:01:09,600
Speaker 1: Will Conkling, head of Data Center Energy for the America's

21
00:01:09,600 --> 00:01:13,280
Speaker 1: in Amia at Google, Kleiber Costa, chief commercial officer for

22
00:01:13,480 --> 00:01:17,160
Speaker 1: AS Corporation, and Darwesh Singh, the founder and CEO of

23
00:01:17,200 --> 00:01:21,160
Speaker 1: Bold Graphics. The panel was moderated by Mark Daily, BNF's

24
00:01:21,200 --> 00:01:25,319
Speaker 1: head of Technology and Innovation. For more information about BNF

25
00:01:25,400 --> 00:01:27,920
Speaker 1: Summit's taking place around the world, as well as our

26
00:01:28,000 --> 00:01:30,120
Speaker 1: upcoming event in New York on the twenty ninth and

27
00:01:30,160 --> 00:01:32,760
Speaker 1: thirtieth of April, an to view recordings from this and

28
00:01:32,840 --> 00:01:36,399
Speaker 1: other previous events, head to about dot BNF dot com

29
00:01:36,400 --> 00:01:39,320
Speaker 1: Forward Slash Summit. Right now, let's hear from our panel

30
00:01:39,440 --> 00:01:48,440
Speaker 1: regarding power demand and data centers.

31
00:01:52,800 --> 00:01:55,160
Speaker 2: Thank you everyone very much for joining us here today.

32
00:01:56,880 --> 00:02:00,480
Speaker 2: We've heard an awful lot about the energy transition, like

33
00:02:00,600 --> 00:02:02,880
Speaker 2: at all be enough events, and I imagine a lot

34
00:02:02,880 --> 00:02:05,240
Speaker 2: of events that everyone here goes to. It's very energy

35
00:02:05,240 --> 00:02:08,600
Speaker 2: focused group, transport focused group, but there's lots of big

36
00:02:08,600 --> 00:02:11,720
Speaker 2: transitions happening in the global economy. We're here today to

37
00:02:11,760 --> 00:02:14,799
Speaker 2: talk about the intersection of the energy transition with one

38
00:02:14,840 --> 00:02:18,480
Speaker 2: of the other really big transformations in the economy, maybe

39
00:02:18,480 --> 00:02:21,000
Speaker 2: the biggest one of our lifetime. If you believe some people.

40
00:02:21,680 --> 00:02:24,560
Speaker 2: So we're here to talk about data centers, energy demand,

41
00:02:24,720 --> 00:02:29,360
Speaker 2: artificial intelligence. I'm joined here today by Stephen Carlini, Chief

42
00:02:29,400 --> 00:02:33,080
Speaker 2: Advocate Data Centers and AI at Schneider Electric, Will Conkling,

43
00:02:33,240 --> 00:02:37,400
Speaker 2: head of Data Center Energy America's Animea at Google, Clever Costa,

44
00:02:38,639 --> 00:02:42,080
Speaker 2: chief commercial Officer at the AEES Corporation, and Darbush Sing,

45
00:02:42,080 --> 00:02:46,960
Speaker 2: Founder and CEO of Bolt Graphics. So the first thing

46
00:02:47,000 --> 00:02:49,560
Speaker 2: that I want to talk about is really why has

47
00:02:49,560 --> 00:02:54,240
Speaker 2: there been so much interest in data centers this year? Like,

48
00:02:54,320 --> 00:02:56,240
Speaker 2: let's just start from the beginning. What's causing out go

49
00:02:56,240 --> 00:02:58,320
Speaker 2: to you first, Star Resh, and if everyone can actually

50
00:02:58,360 --> 00:03:01,120
Speaker 2: just introduce themselves and kind of their view on the

51
00:03:01,160 --> 00:03:03,280
Speaker 2: industry in their first answer, that'd be great.

52
00:03:03,560 --> 00:03:04,440
Speaker 3: Yeah. Thanks Mark.

53
00:03:05,240 --> 00:03:08,240
Speaker 4: I'm dar Wesh Funer Ceobo Graphics. We are a semi

54
00:03:08,240 --> 00:03:11,800
Speaker 4: conductory startup focusing on GPUs. My background is in building

55
00:03:11,840 --> 00:03:14,800
Speaker 4: data centers, so I share a lot of the pains

56
00:03:15,400 --> 00:03:18,560
Speaker 4: at the panel and in the industry in general. Mark,

57
00:03:18,600 --> 00:03:23,760
Speaker 4: I think to your question, compute requirements have always been increasing.

58
00:03:23,800 --> 00:03:25,760
Speaker 4: This is I think it's not new to anyone. We

59
00:03:25,800 --> 00:03:29,799
Speaker 4: have phones now that are more powerful than PlayStation fours

60
00:03:29,880 --> 00:03:33,160
Speaker 4: from twelve years ago. What is new, though, is the

61
00:03:33,280 --> 00:03:38,400
Speaker 4: demand for ten years in the future computing power right now.

62
00:03:38,680 --> 00:03:41,680
Speaker 4: And a lot of these AI companies, whether they're training models,

63
00:03:42,080 --> 00:03:45,880
Speaker 4: whether they're building data centers to host these models, or

64
00:03:45,880 --> 00:03:48,520
Speaker 4: whether they're making phones that can run inference on these things,

65
00:03:48,520 --> 00:03:51,160
Speaker 4: they want it right now. And so I think that

66
00:03:51,200 --> 00:03:53,040
Speaker 4: creates a lot of interest, a lot of hype. It

67
00:03:53,080 --> 00:03:57,680
Speaker 4: also creates an opportunity for new players in the market

68
00:03:57,680 --> 00:03:59,720
Speaker 4: to come in, whether they're DC builders that can build

69
00:03:59,760 --> 00:04:03,480
Speaker 4: data centers in six months instead of four years, small

70
00:04:04,160 --> 00:04:07,240
Speaker 4: modular nuclear reactor companies that are building these and seven

71
00:04:07,320 --> 00:04:09,760
Speaker 4: years instead of twenty five years. So I think it's

72
00:04:09,800 --> 00:04:11,640
Speaker 4: just like a timeline. Let's shift all this left. I

73
00:04:11,720 --> 00:04:13,720
Speaker 4: want to do right now and what does that really enable?

74
00:04:14,680 --> 00:04:17,840
Speaker 2: Okay, and so Stephen, could you give us a primer

75
00:04:17,880 --> 00:04:21,120
Speaker 2: on how Schneider Electric relates to this conversation and when

76
00:04:21,200 --> 00:04:24,920
Speaker 2: did your job change in this kind of this new

77
00:04:25,360 --> 00:04:26,680
Speaker 2: hype cycle that we're going through.

78
00:04:27,520 --> 00:04:30,640
Speaker 5: Yeah, Kinder Electric, if you're not familiar with this largest

79
00:04:31,040 --> 00:04:34,960
Speaker 5: power and cooling solution provider for data centers in the world,

80
00:04:35,279 --> 00:04:37,960
Speaker 5: and as you said in an earlier panel, you know

81
00:04:38,000 --> 00:04:40,719
Speaker 5: AI is not new AI has been around for a while.

82
00:04:40,760 --> 00:04:43,440
Speaker 5: We've been talking to a lot of the hyperscalers. We

83
00:04:43,520 --> 00:04:47,400
Speaker 5: have staff of people signed to each hyperscaler, each large COLO,

84
00:04:47,520 --> 00:04:52,760
Speaker 5: each large enterprise accounts. And we you know, we noticed

85
00:04:53,080 --> 00:04:55,800
Speaker 5: five or six years ago, you know, these data centers

86
00:04:55,800 --> 00:04:57,880
Speaker 5: can't be built overnight. And five or six years ago,

87
00:04:57,920 --> 00:04:59,680
Speaker 5: a lot of the hyper scalers were coming to us,

88
00:04:59,800 --> 00:05:02,200
Speaker 5: not only talking to us about higher densities, but higher

89
00:05:02,279 --> 00:05:05,120
Speaker 5: densities and scales that we've never heard of. You know,

90
00:05:05,200 --> 00:05:07,800
Speaker 5: twenty twenty megawat data centers. Back then it was a

91
00:05:07,800 --> 00:05:10,600
Speaker 5: big data center. They're talking one hundred, one hundred and

92
00:05:10,640 --> 00:05:13,880
Speaker 5: fifty three hundred megawat campuses back then, and we're like, wow,

93
00:05:13,920 --> 00:05:17,240
Speaker 5: this is this is a big change. So the densities,

94
00:05:17,720 --> 00:05:22,039
Speaker 5: you know, started to really change when Nvidia came out

95
00:05:22,080 --> 00:05:24,680
Speaker 5: with you know, the A one hundreds and the.

96
00:05:24,600 --> 00:05:25,400
Speaker 3: A one hundreds.

97
00:05:26,360 --> 00:05:29,720
Speaker 5: It was about three years ago and and the A

98
00:05:29,839 --> 00:05:32,760
Speaker 5: one hundreds were kind of the first you know, at

99
00:05:32,760 --> 00:05:35,680
Speaker 5: scale deployments for a lot of the hyperscalers, and a

100
00:05:35,720 --> 00:05:37,280
Speaker 5: lot of those were air cool they were twenty five

101
00:05:37,360 --> 00:05:40,680
Speaker 5: kilowats parac Then you saw the grace hoppers, which were

102
00:05:41,800 --> 00:05:45,040
Speaker 5: last year, which were thirty six kilowats are acting seventy

103
00:05:45,040 --> 00:05:48,080
Speaker 5: two kilowats parak and now we have the Blackwells. The

104
00:05:48,120 --> 00:05:50,640
Speaker 5: black Wells are one hundred and thirty two kilowats per rack,

105
00:05:51,160 --> 00:05:54,240
Speaker 5: which is really pushing the limit of what we can

106
00:05:54,279 --> 00:05:58,000
Speaker 5: get powered to these racks and cooling to these racks. Next,

107
00:05:58,240 --> 00:06:02,000
Speaker 5: they're working on Ruben, which is the next generation. They're

108
00:06:02,000 --> 00:06:04,680
Speaker 5: already we're already working on the designs for that at

109
00:06:04,680 --> 00:06:08,240
Speaker 5: two hundred and forty kilowats per rack. So it's just exponentially.

110
00:06:08,320 --> 00:06:11,440
Speaker 5: For years, it was just two socket x eighty six

111
00:06:11,680 --> 00:06:14,760
Speaker 5: pizza box servers in these data centers, and we were

112
00:06:14,760 --> 00:06:17,320
Speaker 5: doing ten kilo wats per rack. And in the last

113
00:06:17,400 --> 00:06:20,599
Speaker 5: four or five years, it's just exponentially, you know, increased,

114
00:06:20,920 --> 00:06:23,240
Speaker 5: and not just increased in small scale, but it's a

115
00:06:23,320 --> 00:06:24,599
Speaker 5: large scale that we talked about.

116
00:06:25,040 --> 00:06:28,359
Speaker 2: Okay, so will you're you're coming from Google, the company

117
00:06:28,360 --> 00:06:30,320
Speaker 2: that started the all it was Google's R and D

118
00:06:30,400 --> 00:06:34,080
Speaker 2: Live a transformer and Google's obviously been a huge energy

119
00:06:34,080 --> 00:06:38,520
Speaker 2: procurement company for years and years though and when did

120
00:06:38,560 --> 00:06:40,840
Speaker 2: when did this translate into changes in your job? The

121
00:06:41,000 --> 00:06:42,880
Speaker 2: recent technology advances.

122
00:06:43,680 --> 00:06:47,080
Speaker 6: Yeah, so the story for me is like I've been

123
00:06:47,080 --> 00:06:49,480
Speaker 6: at Google just over ten years in some way to

124
00:06:49,560 --> 00:06:53,400
Speaker 6: performed buying energy for data centers and buying renewal energy

125
00:06:53,400 --> 00:06:56,080
Speaker 6: for data centers, and working with utilities and working energy supply.

126
00:06:56,600 --> 00:06:59,320
Speaker 6: And you know, Google has been an AI first company

127
00:06:59,440 --> 00:07:01,680
Speaker 6: since about like twenty sixteen, twenty seventeen. Our CEO has

128
00:07:01,720 --> 00:07:03,360
Speaker 6: been saying that for a long time, right, And we've

129
00:07:03,400 --> 00:07:06,400
Speaker 6: been infusing AI and machine learning into our products since

130
00:07:06,400 --> 00:07:09,679
Speaker 6: then in various various ways that you know, we probably

131
00:07:09,720 --> 00:07:12,240
Speaker 6: don't notice as users, but it's been there. And then

132
00:07:12,560 --> 00:07:14,320
Speaker 6: in the last you know, eighteen months or so, it

133
00:07:14,360 --> 00:07:16,160
Speaker 6: see the emergence of you know, sort of more consumer

134
00:07:16,240 --> 00:07:20,400
Speaker 6: facing pure mL and AI products you know before so,

135
00:07:20,480 --> 00:07:23,520
Speaker 6: like i'd say, mid twenty twenty three was kind of

136
00:07:23,840 --> 00:07:27,920
Speaker 6: a watershed moment for us, but we saw the green

137
00:07:28,000 --> 00:07:30,640
Speaker 6: suits of this before that, when you know, the grid

138
00:07:30,720 --> 00:07:36,760
Speaker 6: started to see demand growth, you know, generally across a

139
00:07:36,840 --> 00:07:39,400
Speaker 6: number of sectors and industries, right from more manufacturing and

140
00:07:39,440 --> 00:07:43,360
Speaker 6: more batteries and manufacturing and more car manufacturing and more

141
00:07:43,400 --> 00:07:45,280
Speaker 6: s reshoring of a lot of stuff back to the US,

142
00:07:46,120 --> 00:07:49,080
Speaker 6: and that started to manifest itself, right, and utilities coming

143
00:07:49,120 --> 00:07:50,840
Speaker 6: to us and saying, you know, they have to buil

144
00:07:50,840 --> 00:07:53,400
Speaker 6: out their transmission system in ways that they hadn't anticipated

145
00:07:53,440 --> 00:07:55,920
Speaker 6: in order to continue to serve load. And that's sort

146
00:07:55,920 --> 00:07:57,600
Speaker 6: of you know, power for the course as business as

147
00:07:57,680 --> 00:08:02,320
Speaker 6: usual for us. You couple that sort of like growth

148
00:08:02,400 --> 00:08:05,120
Speaker 6: of going from twenty years ago or sorry for the

149
00:08:05,200 --> 00:08:07,920
Speaker 6: last twenty years twenty twenty three of zero point five

150
00:08:07,960 --> 00:08:11,200
Speaker 6: percent annual growth on the grid to to three four

151
00:08:11,360 --> 00:08:13,200
Speaker 6: now four and a half five percent annual growth from

152
00:08:13,200 --> 00:08:15,960
Speaker 6: the grid, and then you have an emergence of a

153
00:08:16,000 --> 00:08:18,320
Speaker 6: new sort of large you know, data center demand with

154
00:08:18,400 --> 00:08:20,800
Speaker 6: machine learning chips to what Steve has talked about with

155
00:08:20,800 --> 00:08:23,920
Speaker 6: with nvidious chips and then our own internal tens of

156
00:08:23,960 --> 00:08:27,760
Speaker 6: processing units. We saw in mid twenty twenty three the

157
00:08:27,880 --> 00:08:32,280
Speaker 6: sort of crossover point from you know, being able to

158
00:08:32,360 --> 00:08:34,640
Speaker 6: sort of generally source power where when we needed it

159
00:08:34,760 --> 00:08:37,040
Speaker 6: on a timescal that made sense to us, to an

160
00:08:37,080 --> 00:08:40,880
Speaker 6: acceleration of demand to what Darbush just said around you know,

161
00:08:41,080 --> 00:08:43,680
Speaker 6: needing it sooner, and then a push out of lead

162
00:08:43,679 --> 00:08:45,439
Speaker 6: times on some of the grid buildout and some of

163
00:08:45,720 --> 00:08:48,679
Speaker 6: and serving that demand by utilities, and it's a it

164
00:08:48,800 --> 00:08:52,439
Speaker 6: was kind of a collision of technology sort of product

165
00:08:52,920 --> 00:08:59,040
Speaker 6: development curves and infrastructure timelines that don't always mesh well.

166
00:08:59,240 --> 00:09:01,200
Speaker 6: And we've been kind of been in that that soup

167
00:09:01,280 --> 00:09:03,200
Speaker 6: ever since. And it's I like to say, it's been

168
00:09:03,240 --> 00:09:04,800
Speaker 6: one of the more dynamic times in my career. And

169
00:09:04,800 --> 00:09:06,959
Speaker 6: I don't thpect I'll see anything like I would again, So.

170
00:09:08,080 --> 00:09:10,880
Speaker 2: And clever, let's get the energy company perspective on this.

171
00:09:11,559 --> 00:09:13,680
Speaker 2: When did you start noticing a big change from what

172
00:09:13,800 --> 00:09:14,280
Speaker 2: was happening.

173
00:09:14,640 --> 00:09:18,080
Speaker 7: Look, I think if you hear my my my colleagues

174
00:09:18,120 --> 00:09:21,280
Speaker 7: here on the panel, it has been about a lot

175
00:09:21,280 --> 00:09:24,280
Speaker 7: of changes over a very short period of time, a

176
00:09:24,320 --> 00:09:27,440
Speaker 7: lot of volatility. I would say, you know, I think,

177
00:09:28,000 --> 00:09:32,440
Speaker 7: I think you'll see, uh, these these massive growth over

178
00:09:32,440 --> 00:09:37,880
Speaker 7: the past few years projections and then one day opened

179
00:09:38,040 --> 00:09:41,600
Speaker 7: the papers and there's deep, deep seek there changing everything

180
00:09:41,640 --> 00:09:43,280
Speaker 7: and you don't know what is true what is not.

181
00:09:43,520 --> 00:09:47,240
Speaker 7: So everybody's trying to figure figure that out. Look the industry,

182
00:09:47,320 --> 00:09:51,480
Speaker 7: I'm with AES corporation. I've been at AS for for

183
00:09:51,520 --> 00:09:53,679
Speaker 7: about seven years there, but I've been in the energy

184
00:09:53,800 --> 00:09:56,880
Speaker 7: business for about twenty five years. I started my career

185
00:09:56,960 --> 00:10:00,439
Speaker 7: right when air Ron was the greatest thing on earth.

186
00:10:00,960 --> 00:10:04,280
Speaker 7: Everybody wanted to work for and run not not after

187
00:10:04,320 --> 00:10:08,640
Speaker 7: that and Ron, we all know what happened. And then UH,

188
00:10:08,679 --> 00:10:11,400
Speaker 7: there's the boom and bust on the gas cycle. Hainesville

189
00:10:11,440 --> 00:10:17,760
Speaker 7: Economics showed up with Shell Gas UH. A lot of

190
00:10:17,800 --> 00:10:21,680
Speaker 7: companies also went bankrupt, and volatility in the in the

191
00:10:21,800 --> 00:10:24,640
Speaker 7: energy sector disappeared for a long period of time, with

192
00:10:24,880 --> 00:10:29,840
Speaker 7: very small growth over the past five years or maybe

193
00:10:29,840 --> 00:10:33,720
Speaker 7: a little a little more, as Will said, is explosive demand.

194
00:10:33,880 --> 00:10:37,520
Speaker 7: I think it's been one of the most dynamics periods

195
00:10:37,559 --> 00:10:39,360
Speaker 7: of times of my career as well.

196
00:10:39,920 --> 00:10:41,040
Speaker 3: But I guess the point here.

197
00:10:40,960 --> 00:10:45,280
Speaker 7: Is that the energy the industry is not UH is

198
00:10:45,440 --> 00:10:49,400
Speaker 7: very familiar with challenges and how to overcome those challenges.

199
00:10:49,720 --> 00:10:53,640
Speaker 7: So I think my job really changed when when we

200
00:10:53,720 --> 00:10:59,559
Speaker 7: moved from being providers of projects of or or or

201
00:10:59,679 --> 00:11:04,840
Speaker 7: or or technology to partnering with some of these hyperscalers,

202
00:11:05,120 --> 00:11:11,440
Speaker 7: some of these large data center companies and start putting

203
00:11:11,440 --> 00:11:15,200
Speaker 7: together solutions with them. Not only at AS, not only

204
00:11:15,240 --> 00:11:20,120
Speaker 7: we're one of the largest according to bn EF. Actually

205
00:11:20,120 --> 00:11:25,440
Speaker 7: we're the largest provider of renewable energy to corporates over

206
00:11:25,480 --> 00:11:28,480
Speaker 7: the past three years, so we know how that works.

207
00:11:28,520 --> 00:11:31,800
Speaker 7: But we also own utility companies. We own utilities in

208
00:11:31,800 --> 00:11:36,400
Speaker 7: Ohio and Indiana, and we work very closely with those

209
00:11:36,800 --> 00:11:39,720
Speaker 7: hyper scalers and data center companies to meet their needs

210
00:11:39,760 --> 00:11:41,120
Speaker 7: at the utility level as well.

211
00:11:41,480 --> 00:11:43,440
Speaker 3: But again, I think the biggest takeaway.

212
00:11:44,559 --> 00:11:49,880
Speaker 7: It feels very very uncertain in terms of demand projections.

213
00:11:49,880 --> 00:11:53,559
Speaker 7: We're going to talk about that later. But whatever projections

214
00:11:53,600 --> 00:11:58,280
Speaker 7: you look, however you slice and dice there. The growth

215
00:11:58,360 --> 00:12:02,719
Speaker 7: is here, is real, and I think the industry is

216
00:12:04,320 --> 00:12:08,719
Speaker 7: everybody said to meet those challenges and meet solutions for

217
00:12:09,600 --> 00:12:10,120
Speaker 7: that growth.

218
00:12:10,440 --> 00:12:13,000
Speaker 2: So this growth is real, we need to deal with it.

219
00:12:13,520 --> 00:12:16,320
Speaker 2: What's been a big question that I've been asked actually

220
00:12:16,360 --> 00:12:20,320
Speaker 2: a lot in my role is does this look different

221
00:12:20,400 --> 00:12:23,120
Speaker 2: than the data center growth that came before? With the

222
00:12:23,160 --> 00:12:27,000
Speaker 2: idea being data centers have been cited close to populations

223
00:12:27,120 --> 00:12:29,440
Speaker 2: so that latency is low and your Netflix loads really quick.

224
00:12:30,160 --> 00:12:31,960
Speaker 2: But that might not be the case for a new

225
00:12:32,000 --> 00:12:35,360
Speaker 2: AI applications where actually training doesn't need to take place

226
00:12:35,520 --> 00:12:37,600
Speaker 2: close to people. Maybe it'll take place in a far

227
00:12:37,640 --> 00:12:42,840
Speaker 2: away place. Starsh what's your sense of this kind of

228
00:12:42,920 --> 00:12:44,120
Speaker 2: regional dynamic.

229
00:12:45,240 --> 00:12:50,720
Speaker 4: Yeah, definitely, AI is changing the requirements for where you

230
00:12:50,720 --> 00:12:53,360
Speaker 4: build data centers, where you can source power and also

231
00:12:53,400 --> 00:12:56,160
Speaker 4: the conversation. Like I think four years ago, if I

232
00:12:56,200 --> 00:12:58,959
Speaker 4: had a conversation with someone about sourcing like hydrogen power

233
00:12:58,960 --> 00:13:00,640
Speaker 4: for a data center, be laughed at.

234
00:13:01,080 --> 00:13:03,400
Speaker 3: And now I'd be laughed at, but like a lot less.

235
00:13:03,600 --> 00:13:07,200
Speaker 4: I think I'd be pointed towards other areas maybe in

236
00:13:07,240 --> 00:13:07,720
Speaker 4: this direction.

237
00:13:08,720 --> 00:13:10,920
Speaker 3: But definitely.

238
00:13:12,760 --> 00:13:15,240
Speaker 4: The workload that that's running in the data center does

239
00:13:15,360 --> 00:13:18,120
Speaker 4: impact really heavily what those requirements are where I can

240
00:13:18,160 --> 00:13:20,480
Speaker 4: put that data center. And now with training and with

241
00:13:20,920 --> 00:13:25,160
Speaker 4: let's say training is a batch workload, all right, HPEC

242
00:13:25,320 --> 00:13:28,000
Speaker 4: supercomputing is also a batch workload that can also be

243
00:13:28,080 --> 00:13:32,440
Speaker 4: an maybe choose Jensen's common It can be an Antarctica somewhere, right,

244
00:13:32,480 --> 00:13:34,839
Speaker 4: it doesn't be very close to me. So yeah, that

245
00:13:35,240 --> 00:13:38,040
Speaker 4: definitely does change, But I don't think it changes the

246
00:13:38,320 --> 00:13:41,480
Speaker 4: economics that much because you still have to deliver a

247
00:13:41,520 --> 00:13:43,040
Speaker 4: lot of power to it. You still need a really

248
00:13:43,080 --> 00:13:46,840
Speaker 4: fat gigabit multi hunter gigabit network link, and these requirements

249
00:13:46,880 --> 00:13:50,400
Speaker 4: like haven't really changed, I think, to be honest, Like,

250
00:13:50,400 --> 00:13:51,920
Speaker 4: if you're going to build a lot of data centers,

251
00:13:51,960 --> 00:13:54,520
Speaker 4: you're already going in the direction of, hey, I need

252
00:13:54,520 --> 00:13:57,400
Speaker 4: to build these further and further away from large metropolitan

253
00:13:57,440 --> 00:13:58,920
Speaker 4: areas because I can't get power anyway.

254
00:13:59,000 --> 00:14:00,520
Speaker 3: So this is a problem like three four years ago,

255
00:14:00,840 --> 00:14:01,640
Speaker 3: it's just worse now.

256
00:14:02,240 --> 00:14:05,079
Speaker 2: Well, actually, interested in you, you're the energy part of

257
00:14:05,120 --> 00:14:08,719
Speaker 2: the equation at Google. Can you give us a bit

258
00:14:08,760 --> 00:14:12,040
Speaker 2: of information? How does that factor into decision making rom

259
00:14:12,080 --> 00:14:15,120
Speaker 2: where to put data centers? Is it it's decided where

260
00:14:15,160 --> 00:14:16,960
Speaker 2: it goes and then we need to find energy or

261
00:14:17,080 --> 00:14:18,600
Speaker 2: is it part of the decision making process?

262
00:14:18,720 --> 00:14:20,000
Speaker 3: Is that the first thing you decide on?

263
00:14:21,440 --> 00:14:26,840
Speaker 6: Yes? And yes it you know historically so before you know,

264
00:14:26,960 --> 00:14:28,840
Speaker 6: before the growth and the grid that I talked about

265
00:14:29,040 --> 00:14:31,520
Speaker 6: kind of started. You know, it's not quite true, but

266
00:14:31,520 --> 00:14:33,000
Speaker 6: you could almost like throw a dart on a map

267
00:14:33,040 --> 00:14:35,520
Speaker 6: and you know, if you were somewhere within a reasonable

268
00:14:35,560 --> 00:14:38,240
Speaker 6: distance of a metropolitan area, you could probably find power

269
00:14:38,240 --> 00:14:39,840
Speaker 6: and a utility would at least build something for you

270
00:14:39,880 --> 00:14:41,680
Speaker 6: in a couple of years, right, And and that was

271
00:14:41,720 --> 00:14:46,680
Speaker 6: generally okay. You know, today, available grid and available generation

272
00:14:46,880 --> 00:14:49,360
Speaker 6: on that grid are are more scarce, at least for

273
00:14:49,440 --> 00:14:51,520
Speaker 6: the you know, short to medium term, right, And so

274
00:14:51,560 --> 00:14:53,920
Speaker 6: you have to be sort of smarter and better at

275
00:14:53,920 --> 00:14:57,080
Speaker 6: picking locations that have available power in the time scale

276
00:14:57,120 --> 00:15:00,000
Speaker 6: you're looking for and and and sort of act quick

277
00:15:00,080 --> 00:15:04,720
Speaker 6: to go and and reserve it and use it, right. So,

278
00:15:04,720 --> 00:15:08,640
Speaker 6: so yet, yes, it's a it's a more important factor

279
00:15:09,320 --> 00:15:12,320
Speaker 6: for us than maybe it was in the past. But

280
00:15:12,440 --> 00:15:15,200
Speaker 6: it doesn't I think divers to your last point there.

281
00:15:15,200 --> 00:15:17,120
Speaker 6: It's like if you go in the middle of nowhere,

282
00:15:17,360 --> 00:15:19,320
Speaker 6: like there isn't power, there isn't fiber, there isn't there

283
00:15:19,320 --> 00:15:21,880
Speaker 6: aren't people to build things, right, There's just these things

284
00:15:21,880 --> 00:15:26,040
Speaker 6: require a certain amount of infrastructure and civilization around them

285
00:15:26,080 --> 00:15:28,040
Speaker 6: to support them, right, and no one wants to live

286
00:15:28,040 --> 00:15:29,240
Speaker 6: next to it. You have to have people that work

287
00:15:29,240 --> 00:15:31,960
Speaker 6: there all the time, right like so so so there's

288
00:15:32,000 --> 00:15:34,960
Speaker 6: there's that factor. And and then our products also like

289
00:15:35,520 --> 00:15:36,960
Speaker 6: don't want to be in the middle of nowhere if

290
00:15:37,000 --> 00:15:39,320
Speaker 6: they can help it. Because if you build a building

291
00:15:39,480 --> 00:15:42,600
Speaker 6: for a data center, and and and so I'll back up.

292
00:15:42,640 --> 00:15:44,840
Speaker 6: There's there's a few different ways that or its fuveent

293
00:15:44,840 --> 00:15:46,640
Speaker 6: things that happen in an mL data center, right or

294
00:15:46,800 --> 00:15:49,080
Speaker 6: a data center. One, you could be training at Google

295
00:15:49,080 --> 00:15:51,480
Speaker 6: at least an internal model too. You could have a

296
00:15:51,520 --> 00:15:54,040
Speaker 6: customer paying you to be able to train their model, right,

297
00:15:54,360 --> 00:15:56,480
Speaker 6: or Three, you could be serving a model or serve

298
00:15:56,560 --> 00:15:59,720
Speaker 6: doing inference and serving AI to customers, right. Only that

299
00:15:59,800 --> 00:16:02,480
Speaker 6: first thing really is that flexible because our customers still

300
00:16:02,480 --> 00:16:04,480
Speaker 6: want to be like within spitting distance of their of

301
00:16:04,480 --> 00:16:07,080
Speaker 6: their footprint on the cloud. And then for inference, we

302
00:16:07,120 --> 00:16:09,120
Speaker 6: still want to be having little latency service to our

303
00:16:09,160 --> 00:16:12,440
Speaker 6: to our consumers, right. And so the internal training, yes,

304
00:16:12,480 --> 00:16:14,320
Speaker 6: it's more flexible. But if you build a data in

305
00:16:14,360 --> 00:16:15,880
Speaker 6: the middle of nowhere just for that and then you

306
00:16:15,920 --> 00:16:18,120
Speaker 6: finish that job, you actually have a Strandard asset, right.

307
00:16:18,120 --> 00:16:19,720
Speaker 6: And so if you think about efficiency of capital and

308
00:16:19,760 --> 00:16:21,520
Speaker 6: how you want to be able to reuse your capital

309
00:16:21,560 --> 00:16:24,720
Speaker 6: and recycle it, you actually don't necessarily always want to

310
00:16:24,760 --> 00:16:27,520
Speaker 6: be just going far afield to to you know, the

311
00:16:28,080 --> 00:16:30,520
Speaker 6: antarcticas or the deserts. We're might to be power, but

312
00:16:30,640 --> 00:16:33,240
Speaker 6: no people or no no users or no customers. So

313
00:16:33,240 --> 00:16:35,160
Speaker 6: so we tend to still have to follow where our

314
00:16:35,200 --> 00:16:38,080
Speaker 6: products want to be. And and then within those you

315
00:16:38,120 --> 00:16:40,920
Speaker 6: know regions or those uber regions we have to we

316
00:16:40,960 --> 00:16:42,680
Speaker 6: have to go then you know, find power availability.

317
00:16:42,920 --> 00:16:46,960
Speaker 2: So okay, great, and Stephen, you you have great insight

318
00:16:47,040 --> 00:16:50,520
Speaker 2: into the entire kind of data center supply chain in

319
00:16:50,560 --> 00:16:53,920
Speaker 2: your ola Schneider, I'm interested, do you have any kind

320
00:16:53,920 --> 00:16:56,480
Speaker 2: of sense of how much of this new data build,

321
00:16:56,560 --> 00:17:00,680
Speaker 2: data center build that we're seeing is specifically I related.

322
00:17:00,800 --> 00:17:02,680
Speaker 2: Is that even something that makes sense? Is there an

323
00:17:02,720 --> 00:17:05,280
Speaker 2: AI data center versus something else or is it just bits?

324
00:17:05,440 --> 00:17:08,639
Speaker 5: Absolutely, the you know, the servers are completely different, and

325
00:17:08,880 --> 00:17:10,600
Speaker 5: you know, the power and the cooling is different. And

326
00:17:10,800 --> 00:17:12,440
Speaker 5: if you look at a data center, it's an AI

327
00:17:12,560 --> 00:17:14,920
Speaker 5: data center today, say it's a ten megawaut data center.

328
00:17:15,359 --> 00:17:17,040
Speaker 5: You know, five or six years ago, there'll be a

329
00:17:17,080 --> 00:17:19,600
Speaker 5: thousand IT racks and the data hall would be huge.

330
00:17:19,760 --> 00:17:21,920
Speaker 5: Now the data hall has seven d I t racks

331
00:17:21,960 --> 00:17:24,639
Speaker 5: and you have all these chillers outside, you know, to

332
00:17:24,680 --> 00:17:27,679
Speaker 5: support the cooling that so it's much different. It's not

333
00:17:27,760 --> 00:17:29,600
Speaker 5: these Amazon warehouse type data centers.

334
00:17:29,720 --> 00:17:29,880
Speaker 6: More.

335
00:17:29,920 --> 00:17:33,280
Speaker 5: There's smaller, more confined in the IT rooms, a lot

336
00:17:33,320 --> 00:17:35,159
Speaker 5: of power and cooling going through it. It's not a

337
00:17:35,200 --> 00:17:37,560
Speaker 5: place that you know, it used to be walk around

338
00:17:37,560 --> 00:17:39,439
Speaker 5: and you know your place, servers and everyone had a

339
00:17:39,440 --> 00:17:42,520
Speaker 5: good time, but not anymore. It's a it's a business.

340
00:17:42,840 --> 00:17:45,560
Speaker 5: But you know, as we were saying, you know, the

341
00:17:45,880 --> 00:17:49,919
Speaker 5: you know, putting an asset where there's where there's power,

342
00:17:49,920 --> 00:17:52,639
Speaker 5: and everybody in the world right now is saying that

343
00:17:52,640 --> 00:17:54,200
Speaker 5: the data centers are going to go where.

344
00:17:54,040 --> 00:17:54,639
Speaker 3: The power is.

345
00:17:54,680 --> 00:17:57,600
Speaker 5: But but what we're what we're seeing and the hyper

346
00:17:57,600 --> 00:18:00,439
Speaker 5: scales are all doing this as you just said, is

347
00:18:00,560 --> 00:18:05,119
Speaker 5: they're they're building these training clusters that are close to

348
00:18:05,320 --> 00:18:08,480
Speaker 5: where people are with the intention of using those to

349
00:18:08,520 --> 00:18:10,840
Speaker 5: make money. There's not a lot of money being made,

350
00:18:11,320 --> 00:18:13,760
Speaker 5: you know, training a model, and the question is how

351
00:18:13,760 --> 00:18:15,360
Speaker 5: many more of these models are we going to need,

352
00:18:15,840 --> 00:18:17,800
Speaker 5: so the money is going to be made and deploying

353
00:18:17,840 --> 00:18:20,560
Speaker 5: them in the field. And we're seeing a big shift

354
00:18:20,640 --> 00:18:22,960
Speaker 5: and this, in my opinion, is kind of the year

355
00:18:22,960 --> 00:18:26,040
Speaker 5: of infernts and and we're starting to see you know

356
00:18:26,520 --> 00:18:29,640
Speaker 5: a lot of these training clusters that were originally deployed

357
00:18:29,720 --> 00:18:32,640
Speaker 5: just to train are now doing infrints, either full time

358
00:18:32,920 --> 00:18:35,480
Speaker 5: or part time. And the other thing that we're seeing

359
00:18:35,560 --> 00:18:38,760
Speaker 5: is you know, with infrints close to the users, optimized

360
00:18:38,760 --> 00:18:41,840
Speaker 5: for different applications. We're not seeing those yet because AI

361
00:18:41,960 --> 00:18:44,120
Speaker 5: is still developing. We're still at the beginning of this.

362
00:18:44,359 --> 00:18:47,320
Speaker 5: We don't know, you know, what it's going to take

363
00:18:47,640 --> 00:18:50,320
Speaker 5: to do an AI agent and a genta AI. You

364
00:18:50,320 --> 00:18:51,639
Speaker 5: know what's that going to take. It's going to be

365
00:18:51,720 --> 00:18:54,440
Speaker 5: multi modal how much processing, how much how much of

366
00:18:54,480 --> 00:18:56,520
Speaker 5: the IT stack is going to be needed and where,

367
00:18:56,920 --> 00:19:00,040
Speaker 5: and as we start inputting more and more video. You

368
00:19:00,640 --> 00:19:03,400
Speaker 5: right now everything's text to text, you know, multimodal, it's

369
00:19:03,400 --> 00:19:05,280
Speaker 5: going to be video, tech, text, image, it's going to

370
00:19:05,359 --> 00:19:08,680
Speaker 5: be all these different things. So we can't optimize the

371
00:19:09,440 --> 00:19:12,360
Speaker 5: data centers close to the users yet for inference, we're

372
00:19:12,400 --> 00:19:14,520
Speaker 5: still going to have to do those in data centers.

373
00:19:14,800 --> 00:19:17,119
Speaker 5: I think that's going to be the case for a

374
00:19:17,119 --> 00:19:17,960
Speaker 5: few years.

375
00:19:18,480 --> 00:19:21,840
Speaker 2: Okay, so things not really changing too dramatically. It's just

376
00:19:21,880 --> 00:19:23,800
Speaker 2: build close to the users, like always.

377
00:19:25,480 --> 00:19:25,920
Speaker 3: So clever.

378
00:19:25,960 --> 00:19:28,400
Speaker 2: Actually, something I want to ask you about. I'm from

379
00:19:28,400 --> 00:19:30,680
Speaker 2: Ireland and so when I hear data center, I think,

380
00:19:30,720 --> 00:19:34,040
Speaker 2: oh my god, it's destroying the power system. There's like

381
00:19:34,119 --> 00:19:36,399
Speaker 2: moratorium and new data centers there because it's such a

382
00:19:36,440 --> 00:19:38,840
Speaker 2: large share of the power system. And there's a couple

383
00:19:38,880 --> 00:19:40,720
Speaker 2: of reasons in Europe where something like this has happened,

384
00:19:40,920 --> 00:19:43,840
Speaker 2: and now there's been conversations about this. This is going to

385
00:19:43,840 --> 00:19:46,160
Speaker 2: happen in the United States. Kind of seems like that's

386
00:19:46,200 --> 00:19:48,199
Speaker 2: calm down in the last few months. But interested to

387
00:19:48,240 --> 00:19:51,440
Speaker 2: hear your thoughts on how big a challenge this will

388
00:19:51,480 --> 00:19:53,200
Speaker 2: be for your business.

389
00:19:54,080 --> 00:19:56,800
Speaker 7: It's hard to talk about that without politicizing things. But

390
00:19:58,400 --> 00:20:01,879
Speaker 7: I think, look, there will be parts of the country

391
00:20:01,920 --> 00:20:05,359
Speaker 7: where there's there would be some resistance uh to data

392
00:20:05,400 --> 00:20:07,879
Speaker 7: center deployment. We are seeing some of that in the

393
00:20:08,200 --> 00:20:12,800
Speaker 7: Southeast and other and other parts. I think I think

394
00:20:12,840 --> 00:20:15,800
Speaker 7: it's the real answer to that is all going to

395
00:20:15,880 --> 00:20:20,080
Speaker 7: depend on the solutions that we bring to to that

396
00:20:20,240 --> 00:20:21,760
Speaker 7: data center load growth.

397
00:20:21,800 --> 00:20:23,800
Speaker 3: When you say we, do you mean as or do

398
00:20:23,840 --> 00:20:24,480
Speaker 3: you mean we.

399
00:20:24,680 --> 00:20:30,760
Speaker 7: The providers of energy and together with the data center

400
00:20:30,840 --> 00:20:33,840
Speaker 7: operators and the and the hyper skaters. I think a

401
00:20:33,840 --> 00:20:36,080
Speaker 7: lot of all we talked about here is that there

402
00:20:36,160 --> 00:20:39,080
Speaker 7: was a there was a time not long ago, there

403
00:20:39,320 --> 00:20:41,920
Speaker 7: was this idea that not because of the l LAM

404
00:20:42,160 --> 00:20:46,040
Speaker 7: training phase, a lot of data centers, large data centers

405
00:20:46,040 --> 00:20:51,160
Speaker 7: will co locate with generation in places where there's as

406
00:20:51,200 --> 00:20:53,920
Speaker 7: Will was saying, there's no load, that there's no infrastructure,

407
00:20:53,960 --> 00:20:57,320
Speaker 7: there's no fiber offs so you have to make up

408
00:20:57,359 --> 00:21:01,480
Speaker 7: for the lack of all that with low cost of power.

409
00:21:01,520 --> 00:21:04,159
Speaker 7: But when you when you when when you look at

410
00:21:04,359 --> 00:21:07,679
Speaker 7: the amount of capital that in you to deploy in

411
00:21:07,760 --> 00:21:10,280
Speaker 7: those data centers, you don't want to run the risk

412
00:21:10,720 --> 00:21:14,719
Speaker 7: of being stranded after the training phase of the of

413
00:21:14,760 --> 00:21:19,120
Speaker 7: the large language model uh ends, so you want to

414
00:21:19,200 --> 00:21:22,840
Speaker 7: use that for something else. So we talked a lot

415
00:21:22,840 --> 00:21:25,720
Speaker 7: about this that here you end up going back to

416
00:21:26,320 --> 00:21:29,439
Speaker 7: where the traditional data center markets are, and in some

417
00:21:29,520 --> 00:21:32,760
Speaker 7: of those markets we are seeing local resistance.

418
00:21:33,000 --> 00:21:33,359
Speaker 3: Uh.

419
00:21:33,600 --> 00:21:38,760
Speaker 7: We're also seeing local resistance to the development of power

420
00:21:38,800 --> 00:21:43,440
Speaker 7: plants to supply those those data centers. The famous NIM

421
00:21:43,480 --> 00:21:46,199
Speaker 7: business not in my backyard. So I think that is

422
00:21:46,240 --> 00:21:49,680
Speaker 7: a challenge that that the industry, both the energy industry

423
00:21:50,040 --> 00:21:53,960
Speaker 7: and the data center industry, has to overcome. It is

424
00:21:54,000 --> 00:21:57,920
Speaker 7: a real challenge, uh. And I think it's one that

425
00:21:58,240 --> 00:22:02,200
Speaker 7: will be will be met with deployment of transmission, because

426
00:22:03,119 --> 00:22:05,919
Speaker 7: I think that there's more flexibility where you can deploy

427
00:22:06,040 --> 00:22:08,480
Speaker 7: data centers than there will be on where you deploy

428
00:22:09,160 --> 00:22:14,040
Speaker 7: the generation, whether it's gas or or renewable. I hopefully

429
00:22:14,119 --> 00:22:17,280
Speaker 7: we don't get to new coal to supply to supply

430
00:22:17,400 --> 00:22:22,080
Speaker 7: this demand. And the bottomneck here is actually transmission to

431
00:22:22,160 --> 00:22:25,679
Speaker 7: get to from from from the generational sources to the

432
00:22:26,080 --> 00:22:27,240
Speaker 7: to the data centers.

433
00:22:27,720 --> 00:22:30,920
Speaker 2: Okay, and we'll actually going to ask you about Google

434
00:22:31,000 --> 00:22:34,560
Speaker 2: has been quite active on trying to develop new sources

435
00:22:34,600 --> 00:22:39,560
Speaker 2: of clean energy before actually this whole AI drive became

436
00:22:39,600 --> 00:22:42,560
Speaker 2: a thing, but you signed a couple of pretty first

437
00:22:42,560 --> 00:22:44,440
Speaker 2: of a kind PPAs in the last couple of years.

438
00:22:44,480 --> 00:22:47,680
Speaker 2: It'd be interesting to hear how progress in that is developing.

439
00:22:47,840 --> 00:22:48,040
Speaker 3: Yeah.

440
00:22:48,119 --> 00:22:52,760
Speaker 6: Sure, So the history of you know, energy consumption and

441
00:22:52,840 --> 00:22:55,760
Speaker 6: energy generation Google is a long one. But the short

442
00:22:55,760 --> 00:22:58,119
Speaker 6: story is this, since I've been there last you know,

443
00:22:58,200 --> 00:23:00,360
Speaker 6: ten and a half yars or so, are our energy

444
00:23:00,359 --> 00:23:03,240
Speaker 6: consumption globally has grown between twenty twenty five percent a year.

445
00:23:04,080 --> 00:23:07,840
Speaker 6: We're now approaching thirty plus maybe more tearrawad hours of

446
00:23:08,000 --> 00:23:11,879
Speaker 6: energy you know, consumed every year. That's doubling every you know,

447
00:23:12,000 --> 00:23:15,480
Speaker 6: five years or so. And we've also signed you know,

448
00:23:17,000 --> 00:23:20,600
Speaker 6: you know, twenty plus gigawatts of renewable energy generation you know,

449
00:23:20,680 --> 00:23:24,720
Speaker 6: contracts to with folks like A Yes and others to

450
00:23:24,720 --> 00:23:29,280
Speaker 6: to help supply energy to our facilities. We we maintain

451
00:23:29,400 --> 00:23:32,160
Speaker 6: a strong climent to our to our clean energy goals,

452
00:23:32,160 --> 00:23:35,960
Speaker 6: and we have an hourly carbon free energy goal a

453
00:23:35,960 --> 00:23:38,879
Speaker 6: BYT twenty thirty that we continue to chase and uh

454
00:23:39,080 --> 00:23:41,439
Speaker 6: chase very vigorously and and to that end, you know,

455
00:23:41,480 --> 00:23:44,520
Speaker 6: we The story there is you can get to about

456
00:23:44,520 --> 00:23:47,560
Speaker 6: seventy to eighty percent carbon free and energy supplies you know,

457
00:23:47,800 --> 00:23:51,719
Speaker 6: through wind and solar and batteries and sort of like

458
00:23:51,800 --> 00:23:55,159
Speaker 6: the mix of things there. But that last twenty to

459
00:23:55,160 --> 00:23:58,640
Speaker 6: twenty five percent, that last mile is is harder because

460
00:23:58,640 --> 00:24:00,919
Speaker 6: you have to start thinking about capacity and baseload and

461
00:24:00,920 --> 00:24:03,640
Speaker 6: reliability and this sort of stuff. And so we spent

462
00:24:03,680 --> 00:24:06,040
Speaker 6: the last few years thinking about and working on, you know,

463
00:24:06,119 --> 00:24:09,400
Speaker 6: sort of what are those next gen technologies after wind

464
00:24:09,400 --> 00:24:11,119
Speaker 6: and solar that are going to be carbon free and

465
00:24:11,680 --> 00:24:14,840
Speaker 6: start to supply you know, up to that that nearly

466
00:24:14,840 --> 00:24:17,359
Speaker 6: one hundred percent carbon free energy. And for US, it's

467
00:24:17,400 --> 00:24:22,200
Speaker 6: things like nuclear power, it's things like launderation storage, it's

468
00:24:22,400 --> 00:24:26,600
Speaker 6: potentially hydrogen, don't look too hard. It's potentially like carbon

469
00:24:26,640 --> 00:24:30,800
Speaker 6: capture and storage and geothermal and we've done a couple

470
00:24:30,880 --> 00:24:32,360
Speaker 6: of these in the last couple of years. We did

471
00:24:32,520 --> 00:24:35,639
Speaker 6: a deal in Nevada between US and a company called Ferbo,

472
00:24:36,080 --> 00:24:39,560
Speaker 6: who's a geothermal developer. They're using our cloud technology to

473
00:24:39,840 --> 00:24:42,000
Speaker 6: optimize how they drill wells and operate their wells and

474
00:24:42,000 --> 00:24:45,800
Speaker 6: operate their plants to then build advanced geothermal plants to

475
00:24:45,960 --> 00:24:49,800
Speaker 6: sell that power to Novada Energy, the utility with whom

476
00:24:49,880 --> 00:24:52,280
Speaker 6: We designed a tariff for a new rate that the

477
00:24:52,359 --> 00:24:56,080
Speaker 6: regulator is approving that's gonna a sign or a lot

478
00:24:56,240 --> 00:24:58,880
Speaker 6: allocate the costs of that geothermal above and beyond sort

479
00:24:58,880 --> 00:25:01,639
Speaker 6: of business as usual to us, the customer right so

480
00:25:01,680 --> 00:25:03,760
Speaker 6: that we don't so that we get the product we want,

481
00:25:03,840 --> 00:25:07,240
Speaker 6: the grid gets a clean baseload generation source, and other

482
00:25:07,280 --> 00:25:10,240
Speaker 6: right pairers don't don't pay the cost. So that's what

483
00:25:10,320 --> 00:25:12,119
Speaker 6: we're really proud of. And that's a model, you know,

484
00:25:12,240 --> 00:25:14,840
Speaker 6: that sort of rate structure and that you can slot

485
00:25:14,880 --> 00:25:17,560
Speaker 6: different technologies into is a model we're working on with

486
00:25:17,600 --> 00:25:20,679
Speaker 6: a lot of other utilities across the US. And then

487
00:25:20,720 --> 00:25:23,639
Speaker 6: the other is a partnership we recently as signed with

488
00:25:23,680 --> 00:25:28,640
Speaker 6: a small modular reactor technology provider called Chiros Power. Chiros

489
00:25:28,720 --> 00:25:32,320
Speaker 6: is developing Gen four small monulor reactors. They have a

490
00:25:32,320 --> 00:25:35,119
Speaker 6: pilot planned for twenty twenty nine for which we're going

491
00:25:35,160 --> 00:25:38,479
Speaker 6: to be a customer in the Tennessee Valley. And in

492
00:25:38,520 --> 00:25:40,400
Speaker 6: addition to being a customer for the pilot to help

493
00:25:40,400 --> 00:25:43,080
Speaker 6: get that commercialized and off the ground, we committed to

494
00:25:43,119 --> 00:25:46,320
Speaker 6: being a customer for their next five reactors. And what

495
00:25:46,320 --> 00:25:49,880
Speaker 6: we get out of that is, you know, some confidence

496
00:25:49,920 --> 00:25:52,680
Speaker 6: of having access to power, assuming that everything goes well

497
00:25:52,680 --> 00:25:54,639
Speaker 6: with our technology, but also the ability to sort of

498
00:25:54,960 --> 00:25:58,199
Speaker 6: partner with them to cite those next reactors in places

499
00:25:58,200 --> 00:26:00,280
Speaker 6: that you know, hopefully make make sense for us and

500
00:26:00,280 --> 00:26:02,560
Speaker 6: our loads and our data centers to then meet that

501
00:26:02,600 --> 00:26:05,200
Speaker 6: hourly carbon free energy goal in places we have growing load.

502
00:26:05,280 --> 00:26:08,639
Speaker 6: And so we expect in the twenty thirties to be

503
00:26:08,800 --> 00:26:12,199
Speaker 6: deploying small modulor reactors onto the grid. It's not going

504
00:26:12,240 --> 00:26:14,800
Speaker 6: to be like a you know, Antarctica small modul the

505
00:26:14,840 --> 00:26:17,040
Speaker 6: actor of data center behind the meter microgrid thing. That's

506
00:26:17,080 --> 00:26:19,880
Speaker 6: not the plan. It's really meant to be a grid

507
00:26:19,920 --> 00:26:23,360
Speaker 6: participant and putting capacity on the grid to supply our needs.

508
00:26:23,440 --> 00:26:25,400
Speaker 6: And so Bi you said about that as well.

509
00:26:25,560 --> 00:26:26,680
Speaker 3: Yeah, okay, great.

510
00:26:27,160 --> 00:26:30,040
Speaker 2: So something someone alluded to earlier was the topic of

511
00:26:30,200 --> 00:26:34,400
Speaker 2: energy efficiency and the word deep seek, which hads.

512
00:26:34,240 --> 00:26:34,639
Speaker 3: Up in the room.

513
00:26:34,680 --> 00:26:36,399
Speaker 2: Does everyone know like what I'm referring to when I

514
00:26:36,400 --> 00:26:41,680
Speaker 2: talk about deep seek? Yes, okay, great, darsh I'm gonna

515
00:26:41,680 --> 00:26:44,399
Speaker 2: go to you here because we're having a conversation whether

516
00:26:45,560 --> 00:26:48,800
Speaker 2: this this something that everyone saw coming. Not necessarily the

517
00:26:48,880 --> 00:26:50,919
Speaker 2: idea that like deep seek was going to release a

518
00:26:50,920 --> 00:26:53,080
Speaker 2: model and this would be the exact market reaction, but

519
00:26:54,200 --> 00:26:56,640
Speaker 2: that there was going to be big gains in energy

520
00:26:56,640 --> 00:26:59,120
Speaker 2: efficiency improvements. This has obviously been the history of data

521
00:26:59,119 --> 00:27:02,200
Speaker 2: centers for ages that energy efficiency has improved. So like,

522
00:27:02,240 --> 00:27:04,960
Speaker 2: why were people so surprised by this? And what's the

523
00:27:04,960 --> 00:27:09,480
Speaker 2: future of energy efficiency in artificial intelligence? Simple question if.

524
00:27:09,400 --> 00:27:13,560
Speaker 3: You could just do great question.

525
00:27:15,119 --> 00:27:17,720
Speaker 4: I think technology comes in waves where like you make

526
00:27:17,800 --> 00:27:20,520
Speaker 4: really good hardware and then you try to extract performance

527
00:27:20,520 --> 00:27:22,359
Speaker 4: out of the hardware as much you can. When you

528
00:27:22,400 --> 00:27:25,000
Speaker 4: reach the limit of what you can do in software space,

529
00:27:25,040 --> 00:27:27,560
Speaker 4: you go back and make better hardware. And ideally that's

530
00:27:27,600 --> 00:27:29,960
Speaker 4: like a very quick trend of like, hey, I spent

531
00:27:30,040 --> 00:27:32,480
Speaker 4: six months, let's say one to two years making hardware,

532
00:27:33,040 --> 00:27:35,160
Speaker 4: one to two years making software, and then I find

533
00:27:35,160 --> 00:27:36,920
Speaker 4: the holes in the hardware and I make it better. Right,

534
00:27:37,600 --> 00:27:39,600
Speaker 4: So I think, what I think, it's just more mostly

535
00:27:39,600 --> 00:27:41,840
Speaker 4: timing like this happened now, I think, yeah, this is

536
00:27:41,880 --> 00:27:44,320
Speaker 4: the expectation is that, hey, I can only buy a

537
00:27:44,359 --> 00:27:46,720
Speaker 4: certain number of Vida GPUs. What can I do with this?

538
00:27:46,800 --> 00:27:47,080
Speaker 3: Now?

539
00:27:48,000 --> 00:27:52,680
Speaker 4: Let me hire highly specialized PTX programmers that don't write

540
00:27:52,720 --> 00:27:55,840
Speaker 4: Kuda code. They right level below that because Kuda doesn't

541
00:27:55,880 --> 00:27:58,320
Speaker 4: solve the problem that I needed to solve. It's too abstracted,

542
00:27:58,359 --> 00:28:01,120
Speaker 4: it too, it's too power hungry, right, so I don't

543
00:28:01,119 --> 00:28:03,159
Speaker 4: get I don't get enough control over the hardware, so

544
00:28:03,200 --> 00:28:05,240
Speaker 4: I need to go to a lower level. You keep

545
00:28:05,240 --> 00:28:07,520
Speaker 4: going down that stack. You get down to hardware, then

546
00:28:07,520 --> 00:28:09,159
Speaker 4: you go down, you go to TSMC, right, then you

547
00:28:09,200 --> 00:28:11,840
Speaker 4: go down you get to like minerals and things like that.

548
00:28:11,920 --> 00:28:14,119
Speaker 3: So you can keep going down that stack.

549
00:28:14,200 --> 00:28:18,040
Speaker 4: But definitely, I think what Deep Sea proved and if

550
00:28:18,040 --> 00:28:20,359
Speaker 4: you guys with the pre the paper, the last page,

551
00:28:20,400 --> 00:28:24,280
Speaker 4: there's suggestions on how to improve in Vida GPUs, which

552
00:28:24,320 --> 00:28:28,439
Speaker 4: is really interesting because this is a redesigning them how

553
00:28:28,480 --> 00:28:30,440
Speaker 4: to use them. Yeah, the micro architecture of the Vida

554
00:28:30,520 --> 00:28:33,960
Speaker 4: GPUs is not optimal for what deep SEEQ wants basically,

555
00:28:34,680 --> 00:28:37,560
Speaker 4: which is interesting, right because that's the that's the cycle

556
00:28:37,600 --> 00:28:40,640
Speaker 4: we're going to go through and so not related. But

557
00:28:40,880 --> 00:28:44,080
Speaker 4: you know, the gp that we're designing solves those problems.

558
00:28:44,120 --> 00:28:46,280
Speaker 4: We did our own benchmarks three years ago and we

559
00:28:46,320 --> 00:28:47,920
Speaker 4: found out some of the problems that Deep Sea found

560
00:28:47,960 --> 00:28:51,760
Speaker 4: out as well. So there are ways to improve hardware honestly,

561
00:28:51,920 --> 00:28:55,280
Speaker 4: like the vendor should do you know, benchmarking and research

562
00:28:55,320 --> 00:28:57,520
Speaker 4: and make the GPUs better themselves. But yeah, it does

563
00:28:57,560 --> 00:29:01,960
Speaker 4: require some interactivity with and end user that's like, hey,

564
00:29:02,000 --> 00:29:04,120
Speaker 4: I want this to be better. Here's like four things

565
00:29:04,160 --> 00:29:05,840
Speaker 4: you can do to make it better. And that will

566
00:29:05,880 --> 00:29:09,000
Speaker 4: continue happening, right. People will make new GPUs, new accelerators,

567
00:29:10,000 --> 00:29:12,080
Speaker 4: people will make co package optics, they'll do all sorts

568
00:29:12,120 --> 00:29:14,760
Speaker 4: of fancy stuff and people will use it. But like,

569
00:29:14,760 --> 00:29:16,600
Speaker 4: I'd actually don't like the way this is running. It's

570
00:29:16,600 --> 00:29:18,480
Speaker 4: actually too slow for my use case. Can you fix

571
00:29:18,520 --> 00:29:20,080
Speaker 4: this thing? So I think this is like normal, but

572
00:29:20,600 --> 00:29:23,480
Speaker 4: I think there's so much focus on the volume of

573
00:29:23,560 --> 00:29:25,880
Speaker 4: GPU ship that I think people forgot that there's still

574
00:29:25,920 --> 00:29:28,760
Speaker 4: optimization room. There's a lot of headroom and optimizing software

575
00:29:28,800 --> 00:29:29,080
Speaker 4: for that.

576
00:29:29,800 --> 00:29:32,400
Speaker 2: Okay, how do you how does that kind of coordination

577
00:29:32,520 --> 00:29:36,160
Speaker 2: work between the software developers and a video like is there?

578
00:29:37,120 --> 00:29:39,040
Speaker 2: Do they have channels to work on this together? Doing

579
00:29:39,120 --> 00:29:42,000
Speaker 2: video have their own internal research teams? Yeah, it's called

580
00:29:42,040 --> 00:29:44,840
Speaker 2: hardware software code design. Some companies do it much better

581
00:29:44,880 --> 00:29:49,680
Speaker 2: than others. We do it the best. Yeah, we do

582
00:29:49,680 --> 00:29:50,120
Speaker 2: a good job.

583
00:29:50,160 --> 00:29:52,880
Speaker 4: But yeah, no, they have teams internally that are building

584
00:29:52,920 --> 00:29:56,720
Speaker 4: fundational models at video train them on supercomputers, AI clusters

585
00:29:57,160 --> 00:29:59,720
Speaker 4: and then they're finding these things. But I think it's

586
00:29:59,760 --> 00:30:02,320
Speaker 4: interest that, Like your next question is, well, why didn't

587
00:30:02,320 --> 00:30:03,960
Speaker 4: they find this out, you know last year when they

588
00:30:04,040 --> 00:30:06,320
Speaker 4: made the hardware, Well, it's the same thing. Why is

589
00:30:06,320 --> 00:30:08,160
Speaker 4: black WU one hundred three to two kilowats per wack

590
00:30:08,160 --> 00:30:09,720
Speaker 4: instead of one hundred and twenty five? So there are

591
00:30:09,760 --> 00:30:13,160
Speaker 4: like there are fuzzy zones where you can miss things,

592
00:30:13,200 --> 00:30:15,320
Speaker 4: and I'm sure I'll miss things and it happens. But

593
00:30:15,840 --> 00:30:20,400
Speaker 4: I think the magnitude of that coming out so aggressively

594
00:30:20,440 --> 00:30:22,600
Speaker 4: saying hey, we don't need this much computing power. And

595
00:30:22,640 --> 00:30:24,880
Speaker 4: also there are things in the GPU at the micro

596
00:30:24,920 --> 00:30:27,240
Speaker 4: architecture level in the silicon that I don't like that

597
00:30:27,320 --> 00:30:28,200
Speaker 4: I want you to change.

598
00:30:28,240 --> 00:30:30,360
Speaker 3: Is is a step shift because.

599
00:30:30,160 --> 00:30:33,120
Speaker 4: Now you're expecting every customer to go down to that level.

600
00:30:33,160 --> 00:30:34,640
Speaker 4: And I think that's the race now as to how

601
00:30:35,240 --> 00:30:38,320
Speaker 4: optimized can you get with one water one hundred watts

602
00:30:38,400 --> 00:30:41,080
Speaker 4: or one giga water or whatever?

603
00:30:41,600 --> 00:30:43,040
Speaker 2: Do you have a kind of benchmark in your own

604
00:30:43,040 --> 00:30:46,640
Speaker 2: mind internally of this is the energy standard that like

605
00:30:47,680 --> 00:30:50,680
Speaker 2: a query and chat GPT was a year ago. How

606
00:30:50,800 --> 00:30:52,840
Speaker 2: much more energy efficient can we get? That? Is it

607
00:30:53,000 --> 00:30:54,040
Speaker 2: orders of magnitude or.

608
00:30:54,080 --> 00:30:56,120
Speaker 3: Orders of magnitude orders of magnitude.

609
00:30:56,120 --> 00:30:59,000
Speaker 4: Absolutely, yeah, I think this conversation is good because everyone

610
00:30:59,040 --> 00:31:01,760
Speaker 4: in the panels like it. Will we're delivering power, that's great,

611
00:31:01,880 --> 00:31:04,600
Speaker 4: that's a problem. But also like I'm a chip guy,

612
00:31:04,680 --> 00:31:07,240
Speaker 4: like we should make better chips that consume less power.

613
00:31:07,640 --> 00:31:10,640
Speaker 4: Perhaps maybe there's like a push and pulled balance there

614
00:31:10,640 --> 00:31:12,200
Speaker 4: of you know, like keep using a good job. Right,

615
00:31:12,200 --> 00:31:15,520
Speaker 4: that's orders of magnitude less, less power consumption, more efficient

616
00:31:15,520 --> 00:31:17,880
Speaker 4: than and envita GPU. It is the main specific in

617
00:31:17,880 --> 00:31:19,760
Speaker 4: that sense, but it does solve the problem and it's

618
00:31:19,760 --> 00:31:23,120
Speaker 4: more efficient. So you will see like a chip startups,

619
00:31:23,160 --> 00:31:26,440
Speaker 4: AI startups, quote package optic startups, all these all these

620
00:31:26,480 --> 00:31:30,120
Speaker 4: companies competing and being able to deliver much orders a

621
00:31:30,160 --> 00:31:31,280
Speaker 4: magnitude better efficiency.

622
00:31:31,320 --> 00:31:32,000
Speaker 3: That's what we're doing.

623
00:31:32,600 --> 00:31:35,000
Speaker 2: Okay, So for the three other panelists, I'm got to

624
00:31:35,000 --> 00:31:39,080
Speaker 2: ask you the same question is did you see this coming? Well? Actually, sorry,

625
00:31:39,120 --> 00:31:40,959
Speaker 2: The first question is how do you operate and like

626
00:31:41,160 --> 00:31:43,160
Speaker 2: you need to make these big decisions about what to

627
00:31:43,160 --> 00:31:46,640
Speaker 2: build and what to allocate resource to under this level

628
00:31:46,680 --> 00:31:49,600
Speaker 2: of uncertainty around energy efficiency improvements. But then I guess

629
00:31:49,640 --> 00:31:52,000
Speaker 2: the second part of the question is does everyone who

630
00:31:52,000 --> 00:31:54,480
Speaker 2: works in the industry kind of assume there's gonna be

631
00:31:54,520 --> 00:31:56,920
Speaker 2: these energy efficiency imrovements you're kind of making decisions around that.

632
00:31:57,360 --> 00:32:02,000
Speaker 5: So ho to you first student, I think everyone expected

633
00:32:02,160 --> 00:32:05,080
Speaker 5: more efficiencies to be to be gained in the transformers

634
00:32:05,080 --> 00:32:07,960
Speaker 5: and the algorithms and and but you know, I think

635
00:32:07,960 --> 00:32:09,960
Speaker 5: he had a panelist earlier that said, you know, all

636
00:32:10,080 --> 00:32:12,080
Speaker 5: the all the models that have been trained have been

637
00:32:12,080 --> 00:32:14,720
Speaker 5: trained on the public data, and there's all this other,

638
00:32:15,240 --> 00:32:17,920
Speaker 5: you know, private data that's actually going to be you know,

639
00:32:18,040 --> 00:32:21,240
Speaker 5: more beneficial. But all I can say about about that

640
00:32:21,480 --> 00:32:25,880
Speaker 5: is that it was this is probably the most confusing topic,

641
00:32:26,000 --> 00:32:28,680
Speaker 5: all the experts that are weighing in and saying completely

642
00:32:28,680 --> 00:32:33,920
Speaker 5: different things. But we haven't seen any reaction, negative reaction

643
00:32:34,080 --> 00:32:36,760
Speaker 5: from of our from our customers as far as you know.

644
00:32:36,840 --> 00:32:39,840
Speaker 3: Orders or or forecasts. It's all. It's all, you know,

645
00:32:39,920 --> 00:32:42,760
Speaker 3: business as usual. There's been like no effect at all.

646
00:32:43,080 --> 00:32:49,720
Speaker 6: Yeah, well, yeah, I think you know, Google and look,

647
00:32:50,040 --> 00:32:53,280
Speaker 6: training models and and and uh, software and hardware are

648
00:32:53,400 --> 00:32:55,960
Speaker 6: by no means my own expertise, and so so i'm

649
00:32:56,040 --> 00:32:58,280
Speaker 6: I'm I don't have a lot to offer here except

650
00:32:58,320 --> 00:33:01,120
Speaker 6: for I think we always expect deficiency gains. I think

651
00:33:01,120 --> 00:33:04,280
Speaker 6: we welcome them. I think we're seeing the same sort

652
00:33:04,320 --> 00:33:07,320
Speaker 6: of efficiency gains and our own you know, model building

653
00:33:07,360 --> 00:33:11,520
Speaker 6: and model training and uh and uh, you know that

654
00:33:11,720 --> 00:33:14,960
Speaker 6: it hasn't to See's point, it hasn't really changed our

655
00:33:15,120 --> 00:33:19,800
Speaker 6: look on on our on our business planning because training

656
00:33:19,840 --> 00:33:22,600
Speaker 6: is only part of you know, the AI story and

657
00:33:22,640 --> 00:33:26,640
Speaker 6: the machine learning story. The serving and the inferences is

658
00:33:26,360 --> 00:33:28,640
Speaker 6: A is also a big part of it. And and

659
00:33:28,760 --> 00:33:31,040
Speaker 6: you know, don't forget that Google does myriad other things

660
00:33:31,040 --> 00:33:33,040
Speaker 6: and data centers right, and so data center, so mL

661
00:33:33,080 --> 00:33:34,840
Speaker 6: and A I are are but a portion of our

662
00:33:35,280 --> 00:33:37,200
Speaker 6: of our forecast and our outlook, and so so we

663
00:33:37,320 --> 00:33:39,120
Speaker 6: remain sort of on a stalle path.

664
00:33:41,000 --> 00:33:42,800
Speaker 7: Yeah, I think the same thing here. I haven't seen

665
00:33:42,840 --> 00:33:47,200
Speaker 7: any real change perhaps with the with the small exception

666
00:33:47,280 --> 00:33:52,640
Speaker 7: of investors over reaction to to headlines, but but the

667
00:33:52,680 --> 00:33:55,480
Speaker 7: fundamentals of the business remain very strong. Let's not forget

668
00:33:55,520 --> 00:33:57,880
Speaker 7: here that a lot of what we're trying to solve

669
00:33:57,880 --> 00:34:03,560
Speaker 7: for in the power markets is also replacement of aging

670
00:34:03,760 --> 00:34:08,680
Speaker 7: infrastructure in the US, especially no cold generation gas generation.

671
00:34:09,360 --> 00:34:13,759
Speaker 7: So efficiency gains are definitely welcome because what that's going

672
00:34:13,840 --> 00:34:16,319
Speaker 7: to do is to make sure that we don't overbuild.

673
00:34:16,800 --> 00:34:19,000
Speaker 7: The worst thing that could happen here is if the

674
00:34:19,040 --> 00:34:26,040
Speaker 7: market starts overbuild, overbuilding generation, transmission, distribution, and rate payers

675
00:34:26,080 --> 00:34:30,120
Speaker 7: get left with massive bills and the demand doesn't show up, right,

676
00:34:30,160 --> 00:34:32,640
Speaker 7: that would be a cycle of boom and bus.

677
00:34:32,719 --> 00:34:32,839
Speaker 2: Right.

678
00:34:32,880 --> 00:34:36,040
Speaker 7: So we believe in competitive markets. We believe in market efficiency.

679
00:34:36,640 --> 00:34:41,400
Speaker 7: And I think if if the efficiency is on the on,

680
00:34:41,440 --> 00:34:44,080
Speaker 7: the on, the on the computing power, also if that

681
00:34:44,120 --> 00:34:47,839
Speaker 7: brings demand back to a more reasonable level. But no

682
00:34:47,880 --> 00:34:51,520
Speaker 7: matter what the projections are, the numbers are staggering. The

683
00:34:51,960 --> 00:34:55,480
Speaker 7: projections for demand for new generation and new and new

684
00:34:55,520 --> 00:34:59,720
Speaker 7: infrastructure for transmission are staggering. So we just hope that

685
00:34:59,719 --> 00:35:02,200
Speaker 7: that that we don't overbuild. I guess that's right.

686
00:35:02,239 --> 00:35:02,520
Speaker 3: Okay.

687
00:35:02,560 --> 00:35:04,719
Speaker 2: So I have one quick fire question which you're only

688
00:35:04,719 --> 00:35:08,000
Speaker 2: allowed to answer one number two by twenty thirty. What

689
00:35:08,160 --> 00:35:10,480
Speaker 2: percentage of US power demand do you think will come

690
00:35:10,520 --> 00:35:11,480
Speaker 2: from data centers?

691
00:35:12,320 --> 00:35:14,400
Speaker 3: Not no explanation, just a number.

692
00:35:15,040 --> 00:35:21,840
Speaker 2: There's a networking session after seven and a half, well SIXI.

693
00:35:21,440 --> 00:35:26,360
Speaker 7: Ish, yeah, I think I'm a little perhaps a little

694
00:35:26,400 --> 00:35:29,200
Speaker 7: more more bullish than that. I think it would be

695
00:35:29,200 --> 00:35:31,320
Speaker 7: like somewhere between eight and ten percent.

696
00:35:32,160 --> 00:35:36,239
Speaker 2: Okay, especially pretty clustered around a certain range. But thank

697
00:35:36,239 --> 00:35:38,600
Speaker 2: you very much. This is really informantive panel. I think

698
00:35:38,600 --> 00:35:41,680
Speaker 2: we learned a lot about the idea that the data

699
00:35:41,719 --> 00:35:44,359
Speaker 2: center industry is going to grow. We're very all, very

700
00:35:44,400 --> 00:35:47,279
Speaker 2: confident on that, but actually maybe won't change quite as

701
00:35:47,320 --> 00:35:51,440
Speaker 2: much in terms of its geographic structure or capital investment cycled.

702
00:35:51,920 --> 00:35:54,319
Speaker 2: So thank you very much for joining me. Please join

703
00:35:54,400 --> 00:35:55,920
Speaker 2: me and giving my Palels a round of applause.

704
00:36:06,760 --> 00:36:09,879
Speaker 1: Today's episode of Switched On was produced by Cam Gray

705
00:36:10,080 --> 00:36:13,759
Speaker 1: with production assistance from Kamala Shelling. Bloomberg NIF is a

706
00:36:13,800 --> 00:36:16,880
Speaker 1: service provided by Bloomberg Finance LP and its affiliates. This

707
00:36:17,000 --> 00:36:19,680
Speaker 1: recording does not constitute, nor should it be construed as

708
00:36:19,719 --> 00:36:23,480
Speaker 1: investment in vice, investment recommendations, or a recommendation as to

709
00:36:23,520 --> 00:36:26,360
Speaker 1: an investment or other strategy. Bloomberg ANIF should not be

710
00:36:26,440 --> 00:36:30,200
Speaker 1: considered as information sufficient upon which to base an investment decision.

711
00:36:30,320 --> 00:36:33,279
Speaker 1: Neither Bloomberg Finance LP nor any of its affiliates makes

712
00:36:33,320 --> 00:36:37,040
Speaker 1: any representation or warranty as to the accuracy or completeness

713
00:36:37,040 --> 00:36:40,040
Speaker 1: of the information contained in this recording, and any liability

714
00:36:40,080 --> 00:36:42,759
Speaker 1: as a result of this recording is expressly disclaimed,