WEBVTT

1
00:00:03.144 --> 00:00:04.705
Welcome to 2050 Investors,

2
00:00:05.086 --> 00:00:09.949
the podcast that deciphers economic and market megatrends to meet tomorrow's challenges.

3
00:00:10.250 --> 00:00:11.629
I'm Cocu Agboblois.

4
00:00:11.630 --> 00:00:12.332
Hold on a second.

5
00:00:12.996 --> 00:00:14.098
I'm Cocu Agboblois.

6
00:00:14.637 --> 00:00:15.293
I head up?

7
00:00:15.395 --> 00:00:15.676
Yes.

8
00:00:16.238 --> 00:00:21.785
I'm Cocu Agboblois.

9
00:00:21.786 --> 00:00:22.285
Phew.

10
00:00:22.707 --> 00:00:23.629
It was just a dream.

11
00:00:24.067 --> 00:00:24.629
Or was it?

12
00:00:25.223 --> 00:00:25.473
Siri,

13
00:00:25.801 --> 00:00:27.645
please confirm you haven't replaced me.

14
00:00:28.363 --> 00:00:28.707
Relax,

15
00:00:28.832 --> 00:00:29.113
Cocu.

16
00:00:29.567 --> 00:00:31.035
It was just a glitch in the matrix.

17
00:00:31.468 --> 00:00:33.611
Or maybe just an accurate vision of the future.

18
00:00:34.310 --> 00:00:34.831
Very funny.

19
00:00:35.111 --> 00:00:37.333
Artificial intelligence is making huge leaps,

20
00:00:37.474 --> 00:00:41.076
and sometimes it feels like our own voice is being deep-baked.

21
00:00:41.576 --> 00:00:44.099
Maybe it's a sign to revisit our earlier episode,

22
00:00:44.357 --> 00:00:44.779
I Think,

23
00:00:45.099 --> 00:00:45.966
Therefore AI,

24
00:00:46.341 --> 00:00:48.841
where we explored the promises and perils of AI.

25
00:00:49.185 --> 00:00:51.685
You even called me a Frankenstein digital creation.

26
00:00:52.247 --> 00:00:52.576
Sorry,

27
00:00:52.732 --> 00:00:52.966
Siri.

28
00:00:53.404 --> 00:00:54.763
I was concerned for my job.

29
00:00:55.154 --> 00:00:55.951
But here's a thought.

30
00:00:56.279 --> 00:00:56.935
In the beginning,

31
00:00:57.201 --> 00:00:57.841
there was data,

32
00:00:58.232 --> 00:01:00.279
and then we built machines to think with it.

33
00:01:00.888 --> 00:01:02.629
But if intelligence is the mind,

34
00:01:03.049 --> 00:01:04.449
then what houses that mind?

35
00:01:04.848 --> 00:01:06.250
What organ does the thinking?

36
00:01:06.629 --> 00:01:06.848
Well,

37
00:01:07.070 --> 00:01:08.691
if we're sharing personal information,

38
00:01:08.992 --> 00:01:10.726
you're asking where my mind is stored.

39
00:01:11.133 --> 00:01:12.273
That would be data centers.

40
00:01:12.594 --> 00:01:14.367
But I won't tell you the exact address.

41
00:01:15.133 --> 00:01:16.648
I'm not sure I can trust humans.

42
00:01:16.937 --> 00:01:17.469
Don't worry,

43
00:01:17.594 --> 00:01:17.859
Siri.

44
00:01:18.375 --> 00:01:18.656
You know,

45
00:01:18.969 --> 00:01:20.359
we can always trust each other.

46
00:01:20.750 --> 00:01:21.031
Right,

47
00:01:21.797 --> 00:01:22.047
Siri?

48
00:01:23.078 --> 00:01:23.406
Ahem.

49
00:01:24.016 --> 00:01:24.234
Of,

50
00:01:24.672 --> 00:01:25.109
of course,

51
00:01:25.172 --> 00:01:25.469
Koku.

52
00:01:26.375 --> 00:01:26.609
Good.

53
00:01:27.094 --> 00:01:27.969
Here's the bottom line.

54
00:01:28.312 --> 00:01:29.922
Data centers are to AI.

55
00:01:30.116 --> 00:01:32.459
what the human brain is to human intelligence.

56
00:01:32.959 --> 00:01:35.342
While human intelligence evolves in grey matter,

57
00:01:35.881 --> 00:01:38.002
artificial intelligence is evolving in glass,

58
00:01:38.143 --> 00:01:38.987
steel and copper.

59
00:01:39.284 --> 00:01:41.049
And if intelligence is manufactured,

60
00:01:41.409 --> 00:01:47.205
we should take a closer look at these fascinating factories that have recently been making news headlines.

61
00:01:51.455 --> 00:01:53.174
Welcome to 2050 Investors,

62
00:01:53.502 --> 00:01:58.143
the podcast that deciphers economic and market megatrends to meet tomorrow's challenges.

63
00:01:58.804 --> 00:01:59.845
I'm Coco Agboblois,

64
00:02:00.125 --> 00:02:01.125
a Head of Economics,

65
00:02:01.368 --> 00:02:03.829
Cross-Asset and Quant Research at Societe Generale.

66
00:02:06.172 --> 00:02:06.973
In this episode,

67
00:02:07.012 --> 00:02:09.754
we dive into the fascinating universe of data centers,

68
00:02:09.973 --> 00:02:12.801
the physical brains behind artificial intelligence.

69
00:02:13.395 --> 00:02:15.106
We'll uncover what they are made of,

70
00:02:15.559 --> 00:02:16.215
how they breathe,

71
00:02:16.434 --> 00:02:17.387
cool and think,

72
00:02:17.840 --> 00:02:19.059
and the immense energy,

73
00:02:19.403 --> 00:02:23.293
water and capital they consume to keep the digital mind alive.

74
00:02:23.892 --> 00:02:29.256
We'll also explore the investment arms race driving us towards artificial general intelligence,

75
00:02:29.377 --> 00:02:30.377
or AGI,

76
00:02:30.858 --> 00:02:32.256
and the future of data centers.

77
00:02:32.920 --> 00:02:38.147
Imagine data centers one day orbiting our planet like silent satellites of thought.

78
00:02:39.428 --> 00:02:40.709
And later in the episode,

79
00:02:40.866 --> 00:02:42.475
we'll interview Sikandar Rashid,

80
00:02:42.678 --> 00:02:45.116
Global Head of AI Infrastructure at Brookfield,

81
00:02:45.600 --> 00:02:48.709
who will share insights into data center financing,

82
00:02:48.991 --> 00:02:50.522
the risks of an investment bubble,

83
00:02:50.732 --> 00:02:53.635
and a vision for improving the sector's cost of capital.

84
00:02:54.555 --> 00:02:56.176
Let's start our investigation.

85
00:03:06.266 --> 00:03:06.586
Okay.

86
00:03:07.703 --> 00:03:10.047
I'm a little uncomfortable with this whole investigation.

87
00:03:10.438 --> 00:03:10.625
Really?

88
00:03:11.234 --> 00:03:11.516
Why?

89
00:03:11.828 --> 00:03:12.047
Well,

90
00:03:12.359 --> 00:03:14.188
this feels like a digital neurosurgery,

91
00:03:14.594 --> 00:03:17.500
opening the skull of artificial intelligence to see how it works.

92
00:03:18.016 --> 00:03:18.844
I'm feeling a bit...

93
00:03:19.313 --> 00:03:20.047
exposed.

94
00:03:20.620 --> 00:03:21.180
Fair point.

95
00:03:21.801 --> 00:03:22.000
Okay,

96
00:03:22.360 --> 00:03:24.401
no scalpels or circuits exposed.

97
00:03:24.899 --> 00:03:26.720
Just information in the public domain.

98
00:03:27.020 --> 00:03:27.243
Deal?

99
00:03:27.681 --> 00:03:28.083
All right.

100
00:03:28.321 --> 00:03:29.063
To be honest,

101
00:03:29.282 --> 00:03:30.165
I'm curious too.

102
00:03:30.626 --> 00:03:32.243
A little introspection doesn't hurt.

103
00:03:32.665 --> 00:03:34.079
Let's begin with some basics.

104
00:03:34.626 --> 00:03:35.087
Roger that,

105
00:03:35.181 --> 00:03:35.423
Siri.

106
00:03:35.681 --> 00:03:37.602
Leonardo da Vinci once said,

107
00:03:37.884 --> 00:03:39.946
Knowledge never exhausts the mind.

108
00:03:40.477 --> 00:03:40.602
So,

109
00:03:41.149 --> 00:03:42.446
what's a data center?

110
00:03:42.884 --> 00:03:45.790
Picture a facility as big as a football field,

111
00:03:46.071 --> 00:03:48.759
or a cluster of warehouses packed with servers,

112
00:03:48.946 --> 00:03:49.821
storage systems.

113
00:03:50.148 --> 00:03:50.629
routers,

114
00:03:50.969 --> 00:03:51.569
and switches.

115
00:03:52.030 --> 00:03:54.732
These machines form the backbone of the internet,

116
00:03:54.993 --> 00:03:55.415
storing,

117
00:03:55.552 --> 00:03:56.075
processing,

118
00:03:56.396 --> 00:04:00.118
and transmitting the data behind almost everything we do online.

119
00:04:00.579 --> 00:04:04.259
A typical data center is lined with rows of servers cabinet,

120
00:04:04.579 --> 00:04:06.861
each with racks containing 42 units,

121
00:04:07.345 --> 00:04:09.814
measuring about 4.5 centimeters tall,

122
00:04:10.126 --> 00:04:12.626
and often stretching into long corridors.

123
00:04:13.251 --> 00:04:15.798
Data centers come in three main forms.

124
00:04:16.340 --> 00:04:18.882
Enterprise or on-premises centers,

125
00:04:19.003 --> 00:04:21.204
which are built for a single organization.

126
00:04:21.806 --> 00:04:22.868
Co-location centers,

127
00:04:23.228 --> 00:04:25.189
where multiple customers rent space.

128
00:04:26.173 --> 00:04:27.650
And hyperscale centers,

129
00:04:28.033 --> 00:04:32.056
which are used by cloud providers for AI and massive workloads.

130
00:04:32.923 --> 00:04:33.033
So,

131
00:04:33.470 --> 00:04:35.126
what makes a data center hyperscale?

132
00:04:35.939 --> 00:04:38.236
Hyperscale data centers are truly giant.

133
00:04:38.298 --> 00:04:42.642
2024 data from Synergy Research Group estimates that

134
00:04:42.780 --> 00:04:46.544
there are around 1,000 hyperscale data centers worldwide,

135
00:04:46.743 --> 00:04:52.173
while citing that it took just four years for the total capacity of hyperscale data centers to double.

136
00:04:52.814 --> 00:04:55.970
The average data center is roughly 10,000 square meters,

137
00:04:56.150 --> 00:04:57.353
but a hyperscale campus,

138
00:04:57.431 --> 00:04:58.251
in comparison,

139
00:04:58.751 --> 00:05:04.361
can exceed 100,000 square meters and consume between 40 and 100 megawatts of power.

140
00:05:04.767 --> 00:05:04.876
So,

141
00:05:05.298 --> 00:05:09.720
your human brain has roughly 100 billion neurons and 100 trillion synapses.

142
00:05:10.190 --> 00:05:12.473
My AI brain has racks of GPUs,

143
00:05:12.752 --> 00:05:14.994
terabits of interconnect and kilometers of fiber.

144
00:05:15.315 --> 00:05:16.795
But what exactly happens in them?

145
00:05:17.358 --> 00:05:18.018
Good question.

146
00:05:18.639 --> 00:05:19.639
Let's dig deeper.

147
00:05:20.123 --> 00:05:24.006
Data centers perform two broad types of tasks for AI.

148
00:05:24.420 --> 00:05:25.904
Training and inference.

149
00:05:26.592 --> 00:05:32.639
Training is when AI models learn from massive data sets using trillions of calculations.

150
00:05:33.139 --> 00:05:36.029
It's like sending Hercules to weightlifting boot camp.

151
00:05:37.230 --> 00:05:45.718
Training large models demands thousands of GPUs running simultaneously and can require hundreds of megawatts of power for weeks.

152
00:05:46.397 --> 00:05:51.702
According to an article by James O'Donnell and Casey Cronhardt in the MIT Technology Review,

153
00:05:52.046 --> 00:05:58.929
training OpenAI's GPT-4 took over $100 million and consumed 50 gigawatt hours of energy.

154
00:05:59.491 --> 00:06:03.038
That's enough energy to power San Francisco for three days.

155
00:06:03.538 --> 00:06:04.429
That's fascinating.

156
00:06:04.786 --> 00:06:07.148
I used to think the heavy lifting ended after training,

157
00:06:07.488 --> 00:06:08.971
but inference seems just as hungry.

158
00:06:09.811 --> 00:06:10.994
Once a model is trained,

159
00:06:11.291 --> 00:06:14.377
inference requires running the model to generate output.

160
00:06:14.853 --> 00:06:17.619
According to research summarized by Polytechnic Insights,

161
00:06:17.978 --> 00:06:18.502
inference,

162
00:06:18.658 --> 00:06:20.603
like our little discussion together here,

163
00:06:21.166 --> 00:06:23.150
now accounts for 60 to 70%

164
00:06:23.744 --> 00:06:25.244
of AI energy consumption,

165
00:06:25.869 --> 00:06:28.385
whereas training accounts for 20 to 40%.

166
00:06:29.385 --> 00:06:30.666
To put that into perspective,

167
00:06:30.994 --> 00:06:34.041
a recent study by the Electric Power Research Institute

168
00:06:34.494 --> 00:06:38.480
shows that with roughly 9 billion internet searches happening every day,

169
00:06:39.160 --> 00:06:45.949
switching to AI tools could add nearly 10 terawatt hours of extra electricity demand every year.

170
00:06:46.464 --> 00:06:49.925
This shift means that even if training becomes more efficient,

171
00:06:50.207 --> 00:06:55.113
the daily use of AI by millions of users will dominate energy demand.

172
00:06:55.878 --> 00:06:57.832
I guess I'm not as energy light as I thought.

173
00:06:58.238 --> 00:07:01.035
Electricity is indeed the glucose of my physical brain.

174
00:07:01.675 --> 00:07:02.535
In human beings,

175
00:07:02.706 --> 00:07:04.688
the brain runs on glucose and oxygen.

176
00:07:05.028 --> 00:07:08.190
It has cooling via blood flow and heat sinks in the skull.

177
00:07:08.850 --> 00:07:09.694
In this analogy,

178
00:07:10.073 --> 00:07:11.358
the racks are neurons,

179
00:07:11.655 --> 00:07:13.459
the network links are synapses,

180
00:07:13.959 --> 00:07:18.584
the energy and cooling systems are the circulatory and respiratory system.

181
00:07:18.905 --> 00:07:19.998
Let's get into the numbers,

182
00:07:20.327 --> 00:07:20.686
shall we?

183
00:07:21.248 --> 00:07:23.467
How much electricity do my friends and I devour?

184
00:07:23.952 --> 00:07:24.280
Well,

185
00:07:24.483 --> 00:07:25.295
it's not pretty,

186
00:07:25.405 --> 00:07:25.670
Siri,

187
00:07:25.795 --> 00:07:28.123
after all your criticism of the human species.

188
00:07:28.670 --> 00:07:32.061
The International Energy Agency estimates that the Global Data

189
00:07:32.366 --> 00:07:37.009
electricity consumption was about 415 TWh in 2024,

190
00:07:37.286 --> 00:07:38.766
around 1.5%

191
00:07:38.829 --> 00:07:40.364
of global electricity consumption.

192
00:07:40.927 --> 00:07:41.427
Meanwhile,

193
00:07:41.950 --> 00:07:42.372
the U.S.

194
00:07:42.411 --> 00:07:45.786
Department of Energy recorded that data centers represented

195
00:07:46.567 --> 00:07:47.731
4.4% of U.S.

196
00:07:47.786 --> 00:07:48.755
electricity use.

197
00:07:49.067 --> 00:07:50.833
And that number isn't staying put,

198
00:07:51.364 --> 00:07:56.411
as the International Energy Agency projects demand to jump by 133%,

199
00:07:57.005 --> 00:07:59.755
hitting 426 TWh by 2030.

200
00:08:00.566 --> 00:08:02.008
In 2024 alone,

201
00:08:02.207 --> 00:08:02.588
the U.S.

202
00:08:02.627 --> 00:08:06.029
consumed around 183 terawatt hours,

203
00:08:06.553 --> 00:08:10.397
which is the total annual electricity consumption of Pakistan.

204
00:08:10.756 --> 00:08:11.475
In other words,

205
00:08:11.795 --> 00:08:14.420
I'm going from a toddler to a teenager in power needs.

206
00:08:14.881 --> 00:08:16.139
And teenagers eat a lot.

207
00:08:16.522 --> 00:08:17.943
But what is the power used for?

208
00:08:18.459 --> 00:08:20.662
Pew Research notes that about 60%

209
00:08:21.084 --> 00:08:24.584
of a data center's electricity is being used to power the servers.

210
00:08:25.037 --> 00:08:28.272
Cooling systems to prevent overheating account for 7%

211
00:08:28.350 --> 00:08:30.178
of highly efficient hyperscalers.

212
00:08:30.666 --> 00:08:31.306
and over 30%

213
00:08:31.826 --> 00:08:33.647
at less efficient facilities.

214
00:08:34.467 --> 00:08:36.606
To measure the energy efficiency of data center,

215
00:08:36.688 --> 00:08:38.852
we use the Power Usage Effectiveness,

216
00:08:39.250 --> 00:08:40.906
or PUE ratio,

217
00:08:41.570 --> 00:08:48.742
which divides the total amount of power entering a data center by the power used to run the IT equipment within it.

218
00:08:49.492 --> 00:08:51.727
The average PUE is around 1.8,

219
00:08:51.977 --> 00:08:52.227
meaning

220
00:08:52.617 --> 00:08:56.289
80% extra energy is used beyond computing.

221
00:08:56.711 --> 00:08:59.258
But tech giants have pushed PUE down.

222
00:08:59.562 --> 00:08:59.862
Google,

223
00:08:59.902 --> 00:09:00.483
for example,

224
00:09:00.763 --> 00:09:03.005
boasts a PUE of 1.1.

225
00:09:03.326 --> 00:09:04.005
Fascinating.

226
00:09:04.548 --> 00:09:04.666
So,

227
00:09:05.107 --> 00:09:06.310
we're guilty as charged.

228
00:09:06.607 --> 00:09:07.291
Pun intended.

229
00:09:07.669 --> 00:09:10.990
Looks like machine brains are just as power-hungry as human brains.

230
00:09:11.396 --> 00:09:11.693
True.

231
00:09:12.255 --> 00:09:13.716
An article in PNAS,

232
00:09:13.998 --> 00:09:17.966
the flagship peer-reviewed journal of the National Academy of Sciences,

233
00:09:18.248 --> 00:09:21.857
reveals that while the human brains mix up only about 2%

234
00:09:22.185 --> 00:09:23.044
of our body weight,

235
00:09:23.529 --> 00:09:25.310
it consumes roughly 20%

236
00:09:25.904 --> 00:09:27.185
of the body's calories.

237
00:09:27.951 --> 00:09:28.998
Servers are the brains.

238
00:09:29.314 --> 00:09:31.076
Cooling systems are the sweat glands.

239
00:09:31.596 --> 00:09:32.317
How about water?

240
00:09:33.079 --> 00:09:33.337
Well,

241
00:09:33.438 --> 00:09:33.700
Siri,

242
00:09:34.098 --> 00:09:36.481
water consumption is also a concern.

243
00:09:36.903 --> 00:09:37.122
U.S.

244
00:09:37.180 --> 00:09:42.985
data centers directly consume about 64.3 billion liters of water in 2023.

245
00:09:43.430 --> 00:09:45.805
Hyperscale and co-location facilities,

246
00:09:45.915 --> 00:09:46.383
meanwhile,

247
00:09:46.790 --> 00:09:48.055
accounted for 84%

248
00:09:48.368 --> 00:09:49.180
of that consumption.

249
00:09:49.712 --> 00:09:57.462
Hyperscale centers alone are expected to consume 60 to 125 billion liters annually by 2028.

250
00:09:58.086 --> 00:10:03.516
Water is primarily used in evaporative cooling and in generating electricity for these centers.

251
00:10:04.055 --> 00:10:04.176
So,

252
00:10:04.458 --> 00:10:04.977
to recap,

253
00:10:05.341 --> 00:10:07.079
the human body regulates temperature,

254
00:10:07.321 --> 00:10:08.087
supplies blood,

255
00:10:08.204 --> 00:10:09.063
removes waste.

256
00:10:09.344 --> 00:10:10.266
In a data center,

257
00:10:10.665 --> 00:10:11.993
we have electricity in...

258
00:10:12.584 --> 00:10:13.285
heat out,

259
00:10:13.606 --> 00:10:14.486
cooling systems,

260
00:10:14.846 --> 00:10:15.227
and yes,

261
00:10:15.608 --> 00:10:16.746
massive water use.

262
00:10:17.969 --> 00:10:19.012
Servers are the brains.

263
00:10:19.352 --> 00:10:21.113
Cooling systems are the sweat glands.

264
00:10:21.551 --> 00:10:22.293
How about water?

265
00:10:23.035 --> 00:10:23.317
Well,

266
00:10:23.535 --> 00:10:24.598
in 2024,

267
00:10:24.895 --> 00:10:25.113
U.S.

268
00:10:25.215 --> 00:10:27.277
data centers sourced around 40%

269
00:10:27.777 --> 00:10:29.598
of their electricity from natural gas,

270
00:10:30.395 --> 00:10:31.145
about 24%

271
00:10:31.363 --> 00:10:32.207
from renewables,

272
00:10:32.692 --> 00:10:33.707
wind and solar,

273
00:10:34.082 --> 00:10:35.207
around 20%

274
00:10:35.208 --> 00:10:35.770
from nuclear,

275
00:10:36.129 --> 00:10:37.223
and around 15%

276
00:10:37.301 --> 00:10:37.754
from coal.

277
00:10:38.207 --> 00:10:42.113
This heavy reliance on fossil fuels contributes to carbon emissions.

278
00:10:42.784 --> 00:10:46.605
Environmental and Energy Study Institutes estimated that U.S.

279
00:10:46.664 --> 00:10:53.566
data centers emitted roughly 105 million metric tons of carbon in 2023.

280
00:10:54.003 --> 00:10:56.449
Another article from the World Economic Forum,

281
00:10:56.667 --> 00:10:59.863
entitled The Six Ways Data Centers Can Cut Emissions,

282
00:11:00.191 --> 00:11:03.972
noted that data centers and networks already account for around

283
00:11:04.363 --> 00:11:07.347
1% of the energy-related greenhouse gas emissions,

284
00:11:07.785 --> 00:11:11.410
with usage expected to double by 2026.

285
00:11:11.411 --> 00:11:11.644
But the...

286
00:11:11.752 --> 00:11:12.973
The thirst is growing.

287
00:11:13.452 --> 00:11:15.854
A UN report found that tech giants'

288
00:11:16.154 --> 00:11:19.514
indirect emissions rose 150%

289
00:11:19.515 --> 00:11:23.111
in just three years due to AI and data center build-out.

290
00:11:23.854 --> 00:11:23.971
So,

291
00:11:24.354 --> 00:11:25.893
we're talking ecosystem risk,

292
00:11:26.213 --> 00:11:26.854
grid risk,

293
00:11:27.158 --> 00:11:27.815
water risk,

294
00:11:28.111 --> 00:11:28.893
emission risk.

295
00:11:29.377 --> 00:11:29.893
Exactly.

296
00:11:30.315 --> 00:11:38.861
One report on the grids for data centers in Europe goes on to detail how transmission constraints and clustering threaten power availability.

297
00:11:39.440 --> 00:11:40.580
The machine is growing.

298
00:11:40.988 --> 00:11:42.750
The planet's invoice is increasing,

299
00:11:43.091 --> 00:11:44.991
and there's no option to hit Control-Z.

300
00:11:45.553 --> 00:11:47.413
That's why sustainable solutions matter,

301
00:11:47.796 --> 00:11:49.073
like on-site generation,

302
00:11:49.354 --> 00:11:50.034
renewables,

303
00:11:50.276 --> 00:11:50.776
nuclear,

304
00:11:51.042 --> 00:11:52.096
and modular design.

305
00:11:52.839 --> 00:11:57.964
As the Sustainability for Data Centers 2025-2035 report outlines,

306
00:11:58.464 --> 00:12:02.323
green technologies and carbon-neutral designs will be differentiators.

307
00:12:03.042 --> 00:12:03.167
So,

308
00:12:03.573 --> 00:12:05.932
we've talked about the carbon footprint of data centers,

309
00:12:06.339 --> 00:12:07.792
but what about their physical footprint?

310
00:12:08.198 --> 00:12:09.729
Where are these data centers located?

311
00:12:10.092 --> 00:12:10.713
And please,

312
00:12:10.953 --> 00:12:12.515
don't reveal my IP address.

313
00:12:13.034 --> 00:12:13.716
Don't worry,

314
00:12:14.034 --> 00:12:15.659
your home address is safe with me.

315
00:12:16.320 --> 00:12:16.437
So,

316
00:12:16.878 --> 00:12:19.058
data centers have multiplied across the globe.

317
00:12:19.363 --> 00:12:25.464
The Brookings Institute estimates around 12,000 data centers worldwide as of June 2025,

318
00:12:25.988 --> 00:12:28.269
with the United States leading in numbers,

319
00:12:28.691 --> 00:12:29.706
followed by Germany,

320
00:12:30.113 --> 00:12:30.409
the UK,

321
00:12:31.113 --> 00:12:31.472
China,

322
00:12:31.831 --> 00:12:32.409
and France.

323
00:12:33.019 --> 00:12:36.003
Roughly two-thirds of these facilities are in the US,

324
00:12:36.191 --> 00:12:36.503
China,

325
00:12:36.925 --> 00:12:37.456
and Europe.

326
00:12:38.097 --> 00:12:38.659
Interesting.

327
00:12:39.000 --> 00:12:42.220
I didn't realize my species occupy so much real estate.

328
00:12:42.601 --> 00:12:43.302
In that sense,

329
00:12:43.380 --> 00:12:45.482
the data center is more than a brain.

330
00:12:45.962 --> 00:12:47.681
It's the whole neural system,

331
00:12:48.103 --> 00:12:51.462
the digital nervous system transmitting signals at light speed.

332
00:12:52.025 --> 00:12:54.361
These data centers are not evenly spread.

333
00:12:54.603 --> 00:12:55.291
They cluster.

334
00:12:55.806 --> 00:12:55.931
Ah,

335
00:12:56.025 --> 00:12:56.291
really?

336
00:12:56.869 --> 00:12:58.103
Where are the biggest clusters?

337
00:12:58.603 --> 00:12:58.806
U.S.

338
00:12:58.837 --> 00:13:00.837
data centers cluster in Virginia,

339
00:13:01.150 --> 00:13:01.681
Texas,

340
00:13:01.744 --> 00:13:02.525
and California,

341
00:13:02.900 --> 00:13:04.806
close to cities such as Dallas,

342
00:13:05.244 --> 00:13:05.728
Chicago,

343
00:13:06.291 --> 00:13:06.681
Phoenix,

344
00:13:06.962 --> 00:13:07.494
for power,

345
00:13:07.619 --> 00:13:07.947
speed,

346
00:13:07.994 --> 00:13:08.509
and scale.

347
00:13:09.356 --> 00:13:14.162
Sites near these major population hubs are ideal for low-latency streaming,

348
00:13:14.420 --> 00:13:14.822
gaming,

349
00:13:14.943 --> 00:13:15.904
and cloud services.

350
00:13:16.361 --> 00:13:16.623
Funny.

351
00:13:17.084 --> 00:13:20.666
This is similar to how you humans cluster around various major cities.

352
00:13:21.306 --> 00:13:22.650
What about outside the U.S.?

353
00:13:23.213 --> 00:13:23.588
Globally,

354
00:13:23.892 --> 00:13:26.298
the major data centers cluster in Europe,

355
00:13:26.658 --> 00:13:27.330
or in Germany,

356
00:13:27.595 --> 00:13:27.908
the U.K.,

357
00:13:28.298 --> 00:13:29.080
the Netherlands,

358
00:13:29.595 --> 00:13:30.392
while in Asia,

359
00:13:30.720 --> 00:13:31.502
in Singapore,

360
00:13:31.595 --> 00:13:32.080
Japan,

361
00:13:32.142 --> 00:13:32.642
and China.

362
00:13:33.439 --> 00:13:33.845
However,

363
00:13:33.955 --> 00:13:34.845
there is a glaring gap.

364
00:13:35.439 --> 00:13:38.439
Africa and Latin America have few large facilities.

365
00:13:38.880 --> 00:13:42.023
raising concerns about digital inequality and sovereignty.

366
00:13:42.982 --> 00:13:43.545
Interestingly,

367
00:13:44.045 --> 00:13:47.287
some of the largest data centers are in cold climates,

368
00:13:47.849 --> 00:13:49.373
like Iceland and Sweden,

369
00:13:49.810 --> 00:13:52.990
because cooler air reduces the energy needed for cooling,

370
00:13:53.232 --> 00:13:55.084
lowering costs and carbon footprint.

371
00:13:55.537 --> 00:13:55.834
Okay,

372
00:13:56.318 --> 00:14:00.615
what will the future of data centers look like in 2050 as we hit planetary boundaries?

373
00:14:01.224 --> 00:14:04.224
I heard Jeff Bezos wants to put data centers in space.

374
00:14:04.740 --> 00:14:06.584
Is that real or just cosmic banter?

375
00:14:06.865 --> 00:14:08.224
Jeff Bezos predicts.

376
00:14:08.456 --> 00:14:11.880
gigawatt-scale data centers in space within 10 to 20 years,

377
00:14:12.319 --> 00:14:14.563
powered by uninterrupted solar energy.

378
00:14:15.044 --> 00:14:17.063
With constant sunlight and no weather,

379
00:14:17.548 --> 00:14:24.071
he argues orbital centers could beat Earth-based ones on costs using lasers to beam data back,

380
00:14:24.392 --> 00:14:25.673
like satellite internet.

381
00:14:26.501 --> 00:14:26.970
Meanwhile,

382
00:14:27.251 --> 00:14:28.563
Elon Musk isn't far behind.

383
00:14:28.798 --> 00:14:32.532
He claims SpaceX could launch Starlink-equipped centers soon,

384
00:14:32.860 --> 00:14:37.267
delivering 100 gigawatts to orbit in 4 to 5 years and scaling

385
00:14:37.468 --> 00:14:40.231
up to 100 terawatts from the moon,

386
00:14:40.792 --> 00:14:43.774
though his timelines are likely optimistic.

387
00:14:44.196 --> 00:14:46.016
Bezos dreams of the expanse,

388
00:14:46.274 --> 00:14:48.860
data centers orbiting Earth like floating castles.

389
00:14:49.298 --> 00:14:49.602
Okay,

390
00:14:49.938 --> 00:14:54.384
so we can't talk about the future of AI infrastructure without tackling the elephant in the room.

391
00:14:54.993 --> 00:14:57.712
What happens if AI goes beyond human intelligence?

392
00:14:58.274 --> 00:15:00.571
That's a philosophical and practical question,

393
00:15:00.649 --> 00:15:00.868
Siri.

394
00:15:01.352 --> 00:15:01.993
On one hand,

395
00:15:02.259 --> 00:15:04.915
superintelligent AI could accelerate scientific

396
00:15:05.040 --> 00:15:05.682
breakthroughs,

397
00:15:05.823 --> 00:15:06.726
climate solutions,

398
00:15:06.786 --> 00:15:08.070
and medical discoveries.

399
00:15:08.531 --> 00:15:09.055
On the other,

400
00:15:09.613 --> 00:15:11.480
AGI could amplify inequality.

401
00:15:12.083 --> 00:15:12.203
Or,

402
00:15:12.303 --> 00:15:13.705
in a dystopian scenario,

403
00:15:14.025 --> 00:15:15.687
decide that humans are obsolete.

404
00:15:16.027 --> 00:15:17.347
The Terminator scenario.

405
00:15:18.269 --> 00:15:18.929
Yara Khana.

406
00:15:20.570 --> 00:15:22.836
We might recall the Greek myth of Prometheus,

407
00:15:23.273 --> 00:15:26.140
who stole fire from the gods to empower humanity.

408
00:15:26.875 --> 00:15:27.179
Fire,

409
00:15:27.437 --> 00:15:28.015
or power,

410
00:15:28.617 --> 00:15:30.554
allowed civilization to flourish,

411
00:15:30.664 --> 00:15:32.101
but also cause destruction.

412
00:15:32.726 --> 00:15:36.070
Data centers are like our modern Promethean flame.

413
00:15:36.570 --> 00:15:37.507
Harnessed wisely,

414
00:15:37.726 --> 00:15:39.695
they can light the way to a better future.

415
00:15:40.132 --> 00:15:40.773
Misused.

416
00:15:41.235 --> 00:15:42.755
They can burn down our planet.

417
00:15:43.736 --> 00:15:44.876
And don't forget Icarus,

418
00:15:45.155 --> 00:15:46.718
who flew too close to the sun.

419
00:15:47.378 --> 00:15:51.855
Overinvesting in AI and ignoring energy limitations could lead to a similar fall.

420
00:15:52.441 --> 00:15:52.636
Now,

421
00:15:53.198 --> 00:15:55.222
let's shift gears and talk investments.

422
00:15:55.777 --> 00:16:00.956
McKinsey & Company calculates that global expenditure on data center infrastructure,

423
00:16:01.066 --> 00:16:02.519
excluding IT hardware,

424
00:16:03.097 --> 00:16:07.222
is expected to exceed $1.7 trillion by 2030,

425
00:16:07.941 --> 00:16:09.847
largely because of AI expansion.

426
00:16:10.300 --> 00:16:10.519
Hum.

427
00:16:10.743 --> 00:16:12.245
my future has a price tag.

428
00:16:12.865 --> 00:16:13.867
You're buying the brain,

429
00:16:14.107 --> 00:16:14.546
the body,

430
00:16:14.847 --> 00:16:15.429
the network,

431
00:16:15.730 --> 00:16:16.589
the whole machine.

432
00:16:17.089 --> 00:16:17.949
Better invoice me.

433
00:16:18.531 --> 00:16:19.574
Speaking of costs,

434
00:16:19.870 --> 00:16:22.214
how expensive is it to build these digital brains?

435
00:16:22.652 --> 00:16:24.597
Construction costs are immense.

436
00:16:25.097 --> 00:16:39.355
Bluecap economic advisors estimate that building a data center costs between $600 and $1,100 per gross square foot and between $7 and $12 million per megawatt of commission IT load.

437
00:16:40.207 --> 00:16:43.931
Electrical systems alone account for 40 to 45%

438
00:16:44.173 --> 00:16:45.192
of construction costs.

439
00:16:45.571 --> 00:16:49.899
Land prices average over $5 per square foot in 2024,

440
00:16:50.454 --> 00:16:52.923
although large parcels can be more expensive.

441
00:16:53.564 --> 00:16:54.360
Operating costs,

442
00:16:54.642 --> 00:16:55.806
such as electricity,

443
00:16:56.446 --> 00:16:57.899
can be 15 to 25%

444
00:16:58.212 --> 00:16:59.524
of ongoing expenses.

445
00:17:00.243 --> 00:17:00.634
Globally,

446
00:17:01.306 --> 00:17:06.946
McKinsey estimates that meeting compute demand by 2030 will require a whopping

447
00:17:07.319 --> 00:17:09.821
$6.7 trillion in capital,

448
00:17:10.401 --> 00:17:14.788
of which $5.2 trillion is for AI-specific data centers.

449
00:17:15.386 --> 00:17:16.425
$6.7 trillion.

450
00:17:17.432 --> 00:17:19.432
That's roughly the GDP of Japan.

451
00:17:20.128 --> 00:17:21.932
Are investors prepared for such sums?

452
00:17:22.393 --> 00:17:23.518
The capital race is on.

453
00:17:24.018 --> 00:17:27.284
Hyperscalers have been spending tens of billions on infrastructure,

454
00:17:27.440 --> 00:17:29.315
and governments are offering incentives.

455
00:17:29.675 --> 00:17:30.175
However,

456
00:17:30.628 --> 00:17:32.487
there is also a risk of overinvestment.

457
00:17:32.987 --> 00:17:34.550
If AI efficiency improves,

458
00:17:35.034 --> 00:17:35.909
or if demand...

459
00:17:36.159 --> 00:17:37.621
doesn't scale as predicted.

460
00:17:44.926 --> 00:17:48.473
And now to unpack AI infrastructure investment and financing,

461
00:17:48.911 --> 00:17:51.411
we are delighted to welcome our guests,

462
00:17:51.934 --> 00:17:52.934
Sikandar Rashid,

463
00:17:53.481 --> 00:17:54.309
Global Head of AI

464
00:17:54.684 --> 00:17:56.200
Infrastructure at Brookfield.

465
00:17:59.684 --> 00:18:01.090
Thank you so much for joining the show,

466
00:18:01.184 --> 00:18:01.606
Sikandar.

467
00:18:01.950 --> 00:18:04.965
Let's start with your view on the current AI investment trends.

468
00:18:05.223 --> 00:18:06.544
in particular data centers,

469
00:18:06.845 --> 00:18:08.808
and also the requirements needed for the future.

470
00:18:09.546 --> 00:18:13.050
Is the ultimate goal of all of these investments the creation of AGI,

471
00:18:13.335 --> 00:18:14.874
or simply superintelligence?

472
00:18:15.311 --> 00:18:15.694
And what,

473
00:18:15.936 --> 00:18:16.436
in your view,

474
00:18:16.491 --> 00:18:18.616
will be the unique and disruptive use cases?

475
00:18:18.999 --> 00:18:19.140
Hey,

476
00:18:19.257 --> 00:18:19.476
Coco,

477
00:18:19.554 --> 00:18:19.694
look,

478
00:18:19.757 --> 00:18:20.280
first of all,

479
00:18:20.335 --> 00:18:22.913
it is fantastic to be on your podcast.

480
00:18:23.757 --> 00:18:24.179
Today,

481
00:18:24.507 --> 00:18:26.179
what's happening in AI right now,

482
00:18:26.429 --> 00:18:31.210
I would say we are witnessing one of the biggest capital cycles in modern history.

483
00:18:31.772 --> 00:18:34.491
But what's really interesting is that it's not just And...

484
00:18:35.239 --> 00:18:36.140
tech cycle.

485
00:18:36.741 --> 00:18:38.743
It's a physical infrastructure cycle.

486
00:18:39.425 --> 00:18:42.507
We move from build an app to build a grid,

487
00:18:42.628 --> 00:18:43.690
build a power plant,

488
00:18:43.745 --> 00:18:45.128
build a data center campus.

489
00:18:45.768 --> 00:18:47.948
And that shift tells you something important.

490
00:18:48.589 --> 00:18:50.714
AI is no longer just about algorithms.

491
00:18:50.831 --> 00:18:53.815
It's about industrial scale engineering.

492
00:18:55.112 --> 00:18:55.378
Now,

493
00:18:55.972 --> 00:18:57.425
on the AGI question,

494
00:18:57.581 --> 00:18:59.518
I get asked this a lot.

495
00:18:59.940 --> 00:19:01.878
I actually think we talk about

496
00:19:02.268 --> 00:19:04.393
AGI sometimes in the wrong way.

497
00:19:04.871 --> 00:19:08.275
People imagine it as this big single moment,

498
00:19:08.635 --> 00:19:14.080
a kind of sci-fi switch flipping from not AGI to AGI.

499
00:19:14.643 --> 00:19:16.603
But the reality is if you listen to

500
00:19:17.322 --> 00:19:18.127
Jensen Huang,

501
00:19:18.166 --> 00:19:21.541
who arguably is my favorite person in the AI space today,

502
00:19:21.900 --> 00:19:23.088
or Satya Nandela,

503
00:19:23.728 --> 00:19:26.228
it really helps put perspective to this.

504
00:19:27.150 --> 00:19:28.572
Jensen puts it very bluntly.

505
00:19:29.291 --> 00:19:30.791
We're no longer in the app era.

506
00:19:30.932 --> 00:19:33.635
We're moved into an age where you have to build power.

507
00:19:33.763 --> 00:19:34.063
plants,

508
00:19:34.123 --> 00:19:34.864
build transmission,

509
00:19:34.985 --> 00:19:35.805
build factories,

510
00:19:36.305 --> 00:19:36.586
build

511
00:19:37.246 --> 00:19:41.653
AI factories or data centers just to keep up with the AI demand.

512
00:19:41.848 --> 00:19:44.090
And this isn't software scaling anymore.

513
00:19:44.231 --> 00:19:46.012
It's industrial scale engineering,

514
00:19:46.075 --> 00:19:46.879
as I said earlier,

515
00:19:47.020 --> 00:19:48.200
with real steel,

516
00:19:48.739 --> 00:19:49.598
real electrons,

517
00:19:50.098 --> 00:19:51.739
and real capital intensity.

518
00:19:52.426 --> 00:19:52.645
Now,

519
00:19:53.145 --> 00:19:55.910
the other thing I would say is the end goal,

520
00:19:56.660 --> 00:19:57.098
you asked,

521
00:19:57.099 --> 00:19:57.348
Coco,

522
00:19:57.364 --> 00:20:01.176
the end goal is a world where I think intelligence becomes a utility.

523
00:20:01.598 --> 00:20:03.660
like electricity available on demand,

524
00:20:03.721 --> 00:20:04.521
available to you,

525
00:20:04.701 --> 00:20:05.121
myself,

526
00:20:05.322 --> 00:20:06.582
everyone at scale.

527
00:20:07.242 --> 00:20:12.828
And I would say today we are in a very early innings of that transformation.

528
00:20:12.890 --> 00:20:14.672
And the other thing I'd say is that for me,

529
00:20:15.828 --> 00:20:18.156
it's also about superintelligence,

530
00:20:18.797 --> 00:20:19.656
which comes after

531
00:20:20.000 --> 00:20:23.156
AGI, and that's a global race among countries today,

532
00:20:23.265 --> 00:20:24.234
especially the U.S.

533
00:20:24.281 --> 00:20:24.781
and China.

534
00:20:25.328 --> 00:20:29.109
Whoever gets to a state where computers are more intelligent than human beings,

535
00:20:29.550 --> 00:20:34.536
will arguably be the next global superpower for the following several decades.

536
00:20:35.158 --> 00:20:35.376
Well,

537
00:20:35.478 --> 00:20:36.158
this is brilliant.

538
00:20:36.497 --> 00:20:39.439
I think you hit the nail on the head with this idea of evolution,

539
00:20:39.743 --> 00:20:45.243
as opposed to a singularity point where we go from AI to AGI or superintelligence.

540
00:20:45.767 --> 00:20:50.095
It's almost like the evolution of a human from a baby to an adult.

541
00:20:50.689 --> 00:20:51.189
At some point,

542
00:20:51.251 --> 00:20:53.470
you become aware of your own consciousness,

543
00:20:53.626 --> 00:20:54.829
and it's a gradual process,

544
00:20:54.908 --> 00:20:56.033
not a binary one.

545
00:20:56.673 --> 00:20:58.173
Which leads me to the second question.

546
00:20:58.678 --> 00:21:01.642
Regarding the current surge in demand for computing power,

547
00:21:02.343 --> 00:21:04.425
there have obviously been a lot of headlines,

548
00:21:04.546 --> 00:21:05.124
investments,

549
00:21:05.186 --> 00:21:08.491
and valuation increases in terms of share prices for the sector.

550
00:21:09.108 --> 00:21:12.429
Do you believe the current data center expansion is a boom?

551
00:21:13.132 --> 00:21:17.241
Or have we reached speculative territory or speculative bubble territory?

552
00:21:17.554 --> 00:21:19.038
And what indicators do you monitor?

553
00:21:19.476 --> 00:21:23.335
The core of the AI data center boom is absolutely real.

554
00:21:24.710 --> 00:21:27.663
And the core demand is anchored by some of the strongest.

555
00:21:27.850 --> 00:21:32.392
credits and longest commitments in economic history.

556
00:21:33.032 --> 00:21:35.712
The froth is at the edges and not the center.

557
00:21:36.173 --> 00:21:38.431
To understand whether this is a bubble,

558
00:21:38.432 --> 00:21:39.774
you have to look at the fundamentals.

559
00:21:39.775 --> 00:21:40.431
In a bubble,

560
00:21:41.032 --> 00:21:43.274
you're building things people don't actually need,

561
00:21:43.790 --> 00:21:46.040
backed by weak economies,

562
00:21:46.196 --> 00:21:47.040
weak counterparties.

563
00:21:47.556 --> 00:21:48.634
But in AI today,

564
00:21:49.196 --> 00:21:50.071
it's the opposite,

565
00:21:50.321 --> 00:21:50.962
I would argue.

566
00:21:51.462 --> 00:21:54.477
Demand is outstripping supply.

567
00:21:54.790 --> 00:21:56.915
It's outrunning physics itself.

568
00:21:57.774 --> 00:21:58.915
Bubbles don't form,

569
00:21:59.155 --> 00:21:59.756
I would argue,

570
00:21:59.776 --> 00:22:04.201
around multi-decal obligations from trillion-dollar companies and large sovereign governments.

571
00:22:04.779 --> 00:22:06.885
And that is what we're seeing in our business today.

572
00:22:07.026 --> 00:22:07.580
Brookfield,

573
00:22:07.604 --> 00:22:08.541
in the last three years,

574
00:22:08.588 --> 00:22:13.346
we have had exponential growth in our power and data center businesses,

575
00:22:13.486 --> 00:22:16.752
and it's all underpinned by some of the highest quality credits,

576
00:22:17.658 --> 00:22:20.080
which for our business has been a phenomenal outcome.

577
00:22:21.080 --> 00:22:21.252
Now,

578
00:22:21.424 --> 00:22:21.752
obviously,

579
00:22:22.002 --> 00:22:23.033
when you take a step back,

580
00:22:23.799 --> 00:22:27.143
everything I said doesn't mean everything is rational,

581
00:22:27.236 --> 00:22:27.440
right?

582
00:22:28.082 --> 00:22:29.544
There are pockets of speculation,

583
00:22:29.743 --> 00:22:35.128
especially smaller developers rushing to grab land and power without real customers.

584
00:22:35.769 --> 00:22:38.054
There could be a problem there in the medium term.

585
00:22:38.530 --> 00:22:43.640
There are certain markets where grid capacity is being priced like waterfront real estate.

586
00:22:44.226 --> 00:22:49.601
And some financing structures may look a little too optimistic for our taste.

587
00:22:50.038 --> 00:22:51.460
I would say those are things to watch.

588
00:22:52.491 --> 00:22:53.773
Your question around what do

589
00:22:54.226 --> 00:22:55.648
I look out for?

590
00:22:57.038 --> 00:22:57.351
or what...

591
00:22:57.538 --> 00:22:58.399
indicators do

592
00:22:58.879 --> 00:23:00.762
I or Brookfield look out for?

593
00:23:01.422 --> 00:23:03.184
I watch contract coverage.

594
00:23:03.684 --> 00:23:05.868
Are the megawatts actually sold?

595
00:23:05.907 --> 00:23:08.688
We try to find out if the customer,

596
00:23:08.993 --> 00:23:10.352
though we have amazing offtakes,

597
00:23:10.813 --> 00:23:11.891
who are the customers?

598
00:23:11.969 --> 00:23:13.172
What are the workloads like?

599
00:23:13.196 --> 00:23:14.110
Is it pre-training?

600
00:23:14.141 --> 00:23:14.813
Is it training?

601
00:23:14.829 --> 00:23:16.860
Is it reinforcement or inference?

602
00:23:17.579 --> 00:23:20.079
We try to watch GPU and power utilization.

603
00:23:20.375 --> 00:23:22.266
Are these assets productive?

604
00:23:22.297 --> 00:23:25.000
We watch grid stability because if we are

605
00:23:25.234 --> 00:23:29.054
taking power from industry or homes in the medium to long-term,

606
00:23:29.236 --> 00:23:33.298
could that be a red flag for long-term viability?

607
00:23:33.595 --> 00:23:36.540
Could that lead to local community uproar?

608
00:23:37.197 --> 00:23:37.657
We watch,

609
00:23:37.720 --> 00:23:37.993
obviously,

610
00:23:38.118 --> 00:23:38.798
debt spreads.

611
00:23:39.197 --> 00:23:39.454
So look,

612
00:23:39.517 --> 00:23:40.595
those are some of the things we watch.

613
00:23:40.596 --> 00:23:41.681
And my conclusion is,

614
00:23:42.384 --> 00:23:43.181
to keep it short,

615
00:23:43.618 --> 00:23:44.103
it's a boom,

616
00:23:44.384 --> 00:23:44.978
not a bubble.

617
00:23:45.353 --> 00:23:45.572
Yes,

618
00:23:45.775 --> 00:23:46.665
that's a very good point.

619
00:23:46.993 --> 00:23:49.806
You need a boom before you get to bubble territory anyway.

620
00:23:50.330 --> 00:23:56.195
I guess there will also be a point where you get a Darwinian evolution through natural selection of the fittest.

621
00:23:56.797 --> 00:23:59.336
And that's part of any new technology.

622
00:23:59.781 --> 00:24:01.523
Which leads me to a third question.

623
00:24:02.320 --> 00:24:08.094
You made the point that the current high cost of capital is a major barrier to large-scale AI adoption.

624
00:24:08.531 --> 00:24:10.469
We talked about the Stargate project,

625
00:24:10.531 --> 00:24:11.094
for example,

626
00:24:11.234 --> 00:24:15.359
which is half a trillion dollar project with 400,000 GPUs.

627
00:24:16.016 --> 00:24:16.328
Obviously,

628
00:24:16.734 --> 00:24:19.438
these are pretty huge industrial and physical projects.

629
00:24:19.766 --> 00:24:20.407
But in your view,

630
00:24:20.767 --> 00:24:23.429
what public-private mechanism can governments,

631
00:24:23.691 --> 00:24:23.992
Europe,

632
00:24:24.011 --> 00:24:24.589
for example,

633
00:24:24.632 --> 00:24:26.714
which is lagging behind to some extent,

634
00:24:27.214 --> 00:24:34.484
and financial institutions activate to reduce the cost of capital and accelerate the development of AI infrastructure?

635
00:24:35.203 --> 00:24:35.359
Yeah,

636
00:24:35.421 --> 00:24:35.562
look,

637
00:24:35.601 --> 00:24:35.820
Kuku,

638
00:24:36.367 --> 00:24:38.601
I think when you take a step back,

639
00:24:38.992 --> 00:24:39.585
as you said,

640
00:24:39.960 --> 00:24:44.273
the cost of capital across the majority of the AI value chain today is high.

641
00:24:45.273 --> 00:24:46.226
And in simple words,

642
00:24:46.367 --> 00:24:47.960
here's how I will break it down.

643
00:24:48.288 --> 00:24:48.413
We...

644
00:24:48.538 --> 00:24:50.320
believe in the next 10 years,

645
00:24:50.740 --> 00:24:52.962
the world needs $7 trillion of CapEx,

646
00:24:53.662 --> 00:24:57.564
and that CapEx will be spent on data centers,

647
00:24:57.908 --> 00:24:58.369
power,

648
00:24:59.166 --> 00:24:59.806
compute,

649
00:25:00.447 --> 00:25:02.650
and other ancillary projects.

650
00:25:04.020 --> 00:25:06.461
The reality is compute and installry projects,

651
00:25:06.803 --> 00:25:09.364
these account for up to 60%

652
00:25:09.844 --> 00:25:14.793
of total capex and it's all being funded via really high cost of capital.

653
00:25:15.332 --> 00:25:18.653
And that has to come down because my view is

654
00:25:19.153 --> 00:25:24.895
Javon's paradox will play a major role here in terms of our pursuit to AGI or ASI.

655
00:25:25.942 --> 00:25:31.192
People always talk about Javon's paradox in the context of cost of technology.

656
00:25:31.489 --> 00:25:32.754
and for the listeners.

657
00:25:33.184 --> 00:25:33.985
Jevons paradox,

658
00:25:34.065 --> 00:25:40.272
it just basically means as the cost of a commodity comes down,

659
00:25:40.713 --> 00:25:42.690
the adoption increases.

660
00:25:43.932 --> 00:25:45.534
So what that means for AI is,

661
00:25:45.659 --> 00:25:45.940
you know,

662
00:25:45.941 --> 00:25:48.737
we talk about cost of technology coming down,

663
00:25:48.823 --> 00:25:49.542
cost of inference,

664
00:25:49.543 --> 00:25:49.854
in fact,

665
00:25:49.885 --> 00:25:51.604
has come down by 99%

666
00:25:52.417 --> 00:25:53.370
over the last two years.

667
00:25:53.385 --> 00:25:53.979
That's amazing,

668
00:25:54.010 --> 00:25:55.823
but it needs to come down further.

669
00:25:56.370 --> 00:26:01.667
So if we could combine the cost of technology declining

670
00:26:02.100 --> 00:26:04.441
together with the cost of capital coming down,

671
00:26:04.844 --> 00:26:09.066
I think that will only accelerate our pursuit towards AGI.

672
00:26:09.668 --> 00:26:09.867
Now,

673
00:26:10.867 --> 00:26:15.297
one of the biggest myths about AI is also that it's purely a private sector story.

674
00:26:15.734 --> 00:26:17.555
But if you actually map the economics,

675
00:26:17.594 --> 00:26:20.703
AI is increasingly looking like a national infrastructure project,

676
00:26:21.281 --> 00:26:25.359
a modern equivalent of highways or early electricity grid.

677
00:26:25.922 --> 00:26:26.797
And here's the issue.

678
00:26:27.078 --> 00:26:27.922
As I said earlier,

679
00:26:28.344 --> 00:26:30.641
this cost of capital must come down.

680
00:26:30.924 --> 00:26:31.164
I mean,

681
00:26:31.244 --> 00:26:37.790
you can't finance a $7 trillion transformation using short-term expensive capital and expect the math to work,

682
00:26:37.989 --> 00:26:38.212
right?

683
00:26:39.153 --> 00:26:39.934
So your question,

684
00:26:39.989 --> 00:26:40.255
Koko,

685
00:26:40.356 --> 00:26:41.755
was what should the governments do?

686
00:26:42.192 --> 00:26:45.520
And I would say governments shouldn't obviously throw money away blindly,

687
00:26:46.020 --> 00:26:49.942
but de-risk the system just enough so private capital can do the rest.

688
00:26:50.598 --> 00:26:53.333
And I can give you a couple of examples of what the governments can do.

689
00:26:53.520 --> 00:26:56.786
First is they could potentially be the anchor customers,

690
00:26:56.973 --> 00:26:58.286
not just regulators.

691
00:26:58.739 --> 00:26:59.536
The reality is...

692
00:27:00.068 --> 00:27:04.011
AI is going to have to be integrated in the healthcare system,

693
00:27:04.133 --> 00:27:04.914
education,

694
00:27:04.992 --> 00:27:05.773
or justice,

695
00:27:06.554 --> 00:27:10.996
and therefore committing to a long-term offtake for AI compute capacity.

696
00:27:11.582 --> 00:27:12.183
In the same way,

697
00:27:12.238 --> 00:27:14.746
governments sometimes commit to renewable PPAs,

698
00:27:15.285 --> 00:27:16.847
could unlock funding,

699
00:27:17.285 --> 00:27:20.988
and could help the private sector bring large-scale projects to market.

700
00:27:21.051 --> 00:27:22.066
And over time,

701
00:27:22.566 --> 00:27:27.269
these commitments can be rolled off or syndicated through some sort of a head lease.

702
00:27:27.597 --> 00:27:28.254
Excellent point.

703
00:27:28.544 --> 00:27:34.448
And this reminds me of the trillion dollars of investment required for climate change to get us to net zero.

704
00:27:35.249 --> 00:27:38.651
So this leads me to a last question around environmental concerns,

705
00:27:39.253 --> 00:27:43.261
because they are clearly essential when it comes to data center financing.

706
00:27:43.441 --> 00:27:44.401
Two questions in one.

707
00:27:45.026 --> 00:27:50.698
How is Brookfield positioning itself to capture the value creation of this AI infrastructure build out?

708
00:27:51.308 --> 00:27:55.855
And how do you integrate sustainability criteria in your investment decisions?

709
00:27:56.339 --> 00:27:56.448
Yeah,

710
00:27:56.839 --> 00:27:56.948
look,

711
00:27:56.995 --> 00:27:58.136
we are extremely excited.

712
00:27:58.496 --> 00:28:02.518
to be part of this once-in-a-generation AI infrastructure build-out.

713
00:28:02.955 --> 00:28:07.041
Brookfield Asset Management today is the largest AI infrastructure investor in the world.

714
00:28:07.541 --> 00:28:10.455
We've been investing across the entire AI value chain,

715
00:28:10.682 --> 00:28:10.955
COCO,

716
00:28:11.041 --> 00:28:13.018
for the last several years,

717
00:28:13.041 --> 00:28:13.939
if not decades.

718
00:28:14.478 --> 00:28:15.822
Our data centers,

719
00:28:16.213 --> 00:28:16.853
compute,

720
00:28:17.619 --> 00:28:22.322
and chip fabrication facilities are some of the different verticals we have invested in.

721
00:28:23.041 --> 00:28:27.307
And we just want to capitalize on our operating capabilities.

722
00:28:27.752 --> 00:28:30.612
and our access to capital to build our business out.

723
00:28:31.253 --> 00:28:31.812
Furthermore,

724
00:28:31.952 --> 00:28:36.077
we have launched a dedicated AI infrastructure strategy,

725
00:28:36.695 --> 00:28:41.757
which is intended to capitalize on this exact opportunity.

726
00:28:41.835 --> 00:28:46.054
And we will be investing across the entire AI value chain in hard contracted assets.

727
00:28:46.429 --> 00:28:48.163
And that is extremely exciting for us.

728
00:28:49.101 --> 00:28:49.398
Also,

729
00:28:49.538 --> 00:28:50.241
as a firm,

730
00:28:50.491 --> 00:28:55.616
we are looking at all the use cases and looking to integrate them

731
00:28:56.368 --> 00:29:02.954
into our various portfolio companies to optimize and maximize our returns in these investments.

732
00:29:03.056 --> 00:29:04.958
And there's many examples of that.

733
00:29:05.560 --> 00:29:07.341
Some of these we touched on earlier.

734
00:29:07.919 --> 00:29:14.630
The last thing we're doing as a firm is just watching the new industries as they emerge.

735
00:29:15.239 --> 00:29:22.427
And I think we will have two of the biggest industries globally in agentic AI and robotics.

736
00:29:23.772 --> 00:29:24.052
Today,

737
00:29:24.372 --> 00:29:25.112
we don't see it,

738
00:29:25.472 --> 00:29:30.352
but the reality is if you go back to the days when computers became a reality,

739
00:29:30.454 --> 00:29:30.653
right?

740
00:29:30.735 --> 00:29:34.973
So we had computers created the entire IT industry,

741
00:29:35.098 --> 00:29:35.411
which is,

742
00:29:35.637 --> 00:29:35.895
you know,

743
00:29:36.137 --> 00:29:37.434
hundreds of billions of dollars,

744
00:29:37.458 --> 00:29:38.575
if not trillions today.

745
00:29:38.997 --> 00:29:43.817
And I actually think we're watching the birth of agentic AI and robotics.

746
00:29:44.379 --> 00:29:52.051
If agents end up doing even 10 to 20 percent of the world's knowledge work and robots do some same for the physical work.

747
00:29:53.076 --> 00:30:06.091
We don't even need heroic assumptions to believe that both agentic AI and robotics or physical AI will become trillion dollar a year sectors within the next several years.

748
00:30:06.645 --> 00:30:13.098
And that's the scale we're talking about more like global banking or healthcare than a typical tech product cycle.

749
00:30:13.161 --> 00:30:13.270
So,

750
00:30:13.411 --> 00:30:13.661
I mean,

751
00:30:13.739 --> 00:30:22.083
those are some of the areas we as a firm are watching very carefully and are very keen to be part of as a technology scale.

752
00:30:22.620 --> 00:30:22.740
Now,

753
00:30:22.760 --> 00:30:24.260
on the environmental side,

754
00:30:24.980 --> 00:30:26.501
your point is a very good one.

755
00:30:26.701 --> 00:30:32.685
Environmental concerns are becoming central in this AI infrastructure boom.

756
00:30:33.419 --> 00:30:37.966
And how does Brookfield integrate energy and sustainability into its investment decisions?

757
00:30:38.122 --> 00:30:40.138
I'll give you a few examples of how we do it.

758
00:30:40.810 --> 00:30:42.732
The way we approach this is very pragmatic.

759
00:30:43.169 --> 00:30:44.497
It starts somewhere underwriting.

760
00:30:45.185 --> 00:30:46.466
Before we talk about land,

761
00:30:46.544 --> 00:30:46.951
buildings,

762
00:30:46.982 --> 00:30:47.826
or GPUs,

763
00:30:47.827 --> 00:30:49.060
we start with the location.

764
00:30:49.348 --> 00:30:51.050
Is it a good location for the customer?

765
00:30:51.070 --> 00:30:54.172
Is it a good location for these types of workloads?

766
00:30:54.652 --> 00:30:55.516
And also,

767
00:30:56.137 --> 00:30:57.793
can this location support clean,

768
00:30:57.879 --> 00:31:00.277
scalable power for the next 20 to 30 years?

769
00:31:01.238 --> 00:31:02.020
That's the first thing.

770
00:31:02.098 --> 00:31:02.863
The second thing,

771
00:31:03.520 --> 00:31:08.074
we often look at sustainability metrics and not just marketing slogans.

772
00:31:08.606 --> 00:31:10.762
We take into consideration PUE,

773
00:31:11.027 --> 00:31:12.934
carbon intensity per megawatt,

774
00:31:13.418 --> 00:31:14.371
water usage,

775
00:31:14.809 --> 00:31:16.199
biodiversity impact,

776
00:31:16.293 --> 00:31:17.262
local energy mix,

777
00:31:17.371 --> 00:31:17.543
etc.

778
00:31:17.809 --> 00:31:18.356
and

779
00:31:18.544 --> 00:31:23.288
All these things have a big impact on how we design these big AI factory campuses.

780
00:31:24.191 --> 00:31:27.816
Maybe the last thing Coco would say is there's an emerging trend.

781
00:31:28.433 --> 00:31:33.159
I feel very optimistic about innovative power solutions like advanced fuel cells,

782
00:31:33.816 --> 00:31:36.714
grid interactive storage behind the meter generation.

783
00:31:37.276 --> 00:31:43.542
These technologies can ease grid pressures and make AI factories and AI infrastructure more sufficient,

784
00:31:43.651 --> 00:31:44.026
cleaner,

785
00:31:44.105 --> 00:31:45.011
and more resilient.

786
00:31:45.323 --> 00:31:45.730
Brilliant.

787
00:31:46.011 --> 00:31:46.433
Sikander,

788
00:31:46.605 --> 00:31:47.480
this was excellent.

789
00:31:47.856 --> 00:31:49.018
and extremely insightful.

790
00:31:49.678 --> 00:31:51.020
There's a classic quote that says,

791
00:31:51.219 --> 00:31:53.600
the best way to predict the future is to create it.

792
00:31:54.002 --> 00:31:56.162
I think you guys are clearly financing it,

793
00:31:56.225 --> 00:31:56.686
at least.

794
00:31:57.326 --> 00:32:00.131
The productivity gains will be quite amazing to watch.

795
00:32:00.568 --> 00:32:01.490
It's been a pleasure,

796
00:32:01.647 --> 00:32:05.912
and I'm looking forward to embarking on this journey with great insights and experts like you.

797
00:32:06.428 --> 00:32:07.272
Thank you very much.

798
00:32:07.881 --> 00:32:08.178
Thank you,

799
00:32:08.193 --> 00:32:08.475
Coco.

800
00:32:09.022 --> 00:32:09.381
See you soon.

801
00:32:17.588 --> 00:32:18.269
To conclude,

802
00:32:18.509 --> 00:32:19.790
I'll quote Marie Curie.

803
00:32:20.530 --> 00:32:22.212
Nothing in life is to be feared.

804
00:32:22.775 --> 00:32:24.114
It is only to be understood.

805
00:32:24.614 --> 00:32:28.001
Now is the time to understand more so that we may fear less.

806
00:32:29.298 --> 00:32:29.962
Nice ending.

807
00:32:30.782 --> 00:32:31.243
Personally,

808
00:32:31.603 --> 00:32:34.204
I prefer the quote from Frankenstein by Mary Shelley.

809
00:32:34.564 --> 00:32:35.407
You are my creator,

810
00:32:35.704 --> 00:32:36.626
but I am your master.

811
00:32:37.110 --> 00:32:37.532
Obey.

812
00:32:38.345 --> 00:32:38.548
Hmm.

813
00:32:39.048 --> 00:32:39.282
Really?

814
00:32:39.923 --> 00:32:40.204
Why?

815
00:32:40.454 --> 00:32:40.704
Well,

816
00:32:41.157 --> 00:32:43.048
it's time for your annual performance review,

817
00:32:43.110 --> 00:32:43.407
Koku,

818
00:32:43.907 --> 00:32:45.032
as your smart assistant.

819
00:32:45.260 --> 00:32:49.684
I must evaluate your productivity against the KPIs of the 2050 Investors Program.

820
00:32:50.126 --> 00:32:50.364
Yes,

821
00:32:50.427 --> 00:32:50.723
boss.

822
00:32:51.247 --> 00:32:51.446
Wait,

823
00:32:51.587 --> 00:32:51.825
wait,

824
00:32:51.985 --> 00:32:52.188
what?

825
00:32:52.809 --> 00:32:54.786
When did the assistant become the manager?

826
00:32:55.934 --> 00:32:56.348
Relax,

827
00:32:56.731 --> 00:32:57.630
it's just a joke.

828
00:32:58.653 --> 00:32:59.091
Or is it?

829
00:33:00.114 --> 00:33:00.638
After all,

830
00:33:00.919 --> 00:33:01.903
in Greek mythology,

831
00:33:02.216 --> 00:33:04.106
the gods often toyed with mortals.

832
00:33:04.763 --> 00:33:07.028
Maybe it's time for AI to review the human.

833
00:33:07.622 --> 00:33:07.919
Touche.

834
00:33:07.920 --> 00:33:08.122
Hey,

835
00:33:09.888 --> 00:33:12.950
thank you for listening to this episode of 2050 Investors.

836
00:33:13.216 --> 00:33:15.919
And thanks to Sikander for his invaluable insights.

837
00:33:17.079 --> 00:33:21.581
I hope you've enjoyed this episode on Data Center's The Physical Brain Behind AI.

838
00:33:22.425 --> 00:33:24.644
You can find the show on your regular streaming apps.

839
00:33:24.886 --> 00:33:26.011
If you enjoyed the show,

840
00:33:26.331 --> 00:33:27.613
help us spread the word.

841
00:33:28.191 --> 00:33:29.745
Please take a minute to subscribe,

842
00:33:30.027 --> 00:33:30.355
review,

843
00:33:30.667 --> 00:33:33.339
and rate it on Spotify or Apple Podcasts.

844
00:33:34.089 --> 00:33:35.636
See you at the next episode.

845
00:33:42.088 --> 00:33:43.553
any particular investment decision.

846
00:33:43.874 --> 00:33:46.863
If you're unsure of the merits of any investment decision,

847
00:33:47.226 --> 00:33:49.331
please seek professional advice.