1
00:00:15,356 --> 00:00:23,596
Speaker 1: Pushkin. Brandon McBee was a commodities trader. His job was

2
00:00:23,636 --> 00:00:27,236
Speaker 1: buying and selling things like natural gas and soybeans, and

3
00:00:27,396 --> 00:00:30,036
Speaker 1: in twenty seventeen, he and a few of his friends,

4
00:00:30,076 --> 00:00:34,356
Speaker 1: who were also commodities traders, got interested in crypto obviously,

5
00:00:34,796 --> 00:00:37,916
Speaker 1: and as a hobby, Brand and his friends started buying

6
00:00:38,076 --> 00:00:42,356
Speaker 1: specialized computer hardware to mine crypto, specifically to mind a

7
00:00:42,396 --> 00:00:45,556
Speaker 1: token called ether. They were setting up computers in their

8
00:00:45,596 --> 00:00:49,156
Speaker 1: offices and in garages, and then after a little while,

9
00:00:49,676 --> 00:00:52,796
Speaker 1: other traders they knew wanted in on the action, and

10
00:00:52,876 --> 00:00:56,156
Speaker 1: so Brand and his friends started buying more hardware and

11
00:00:56,276 --> 00:00:57,916
Speaker 1: charging the other traders to use it.

12
00:00:58,916 --> 00:01:02,436
Speaker 2: We were basically retrofitting warehouses in New Jersey.

13
00:01:02,436 --> 00:01:04,596
Speaker 1: That sounds sketchy, it was.

14
00:01:05,036 --> 00:01:08,956
Speaker 2: It was extremely sketchy, lots of wires everywhere, and could

15
00:01:08,956 --> 00:01:11,516
Speaker 2: it be any more different than where we're at today?

16
00:01:11,916 --> 00:01:15,236
Speaker 2: But we figured out how to run tens of thousands

17
00:01:15,236 --> 00:01:19,556
Speaker 2: of GPUs in the harshest environments imaginable GPUs.

18
00:01:20,156 --> 00:01:23,116
Speaker 1: Those were the key pieces of hardware that Brandon and

19
00:01:23,156 --> 00:01:27,276
Speaker 1: his friends were using. GPU stands for graphics processing unit,

20
00:01:27,636 --> 00:01:31,476
Speaker 1: and GPUs were originally developed to power things like video games,

21
00:01:31,956 --> 00:01:34,476
Speaker 1: but they also turn out to be really good for

22
00:01:34,516 --> 00:01:39,596
Speaker 1: some other things like mining cryptocurrency, running complex biological simulations,

23
00:01:39,916 --> 00:01:44,556
Speaker 1: and also GPUs are really good for AI. And as

24
00:01:44,596 --> 00:01:47,436
Speaker 1: Brandon and his buddies kept buying more and more GPUs,

25
00:01:47,556 --> 00:01:50,636
Speaker 1: they thought to themselves, maybe we shouldn't just rent these

26
00:01:50,636 --> 00:01:52,916
Speaker 1: out to people who want to mind crypto. Maybe we

27
00:01:52,916 --> 00:01:54,676
Speaker 1: should see if people want to use them for other

28
00:01:54,716 --> 00:01:58,076
Speaker 1: things like training and running AI. That turned out to

29
00:01:58,116 --> 00:02:02,876
Speaker 1: be a very good call. Today, companies building and running

30
00:02:02,956 --> 00:02:06,516
Speaker 1: AI models are clamoring to get access to GPUs, and

31
00:02:06,676 --> 00:02:09,876
Speaker 1: Brannon and his friend's hobby has grown into a multi

32
00:02:09,916 --> 00:02:13,836
Speaker 1: billion dollar company called core Weave that competes against companies

33
00:02:13,876 --> 00:02:22,396
Speaker 1: like Google, Microsoft, and Amazon. I'm Jacob Goldstein. This is

34
00:02:22,436 --> 00:02:24,516
Speaker 1: What's Your Problem, the show where I talk to people

35
00:02:24,516 --> 00:02:27,876
Speaker 1: who are trying to make technological progress. My guest today

36
00:02:28,036 --> 00:02:31,876
Speaker 1: is Brandon b He's the chief strategy officer at Coreweek

37
00:02:32,356 --> 00:02:34,396
Speaker 1: and I wanted to talk to Brandon because you know,

38
00:02:34,476 --> 00:02:37,076
Speaker 1: obviously we talk about AI all the time, and usually

39
00:02:37,116 --> 00:02:40,276
Speaker 1: we talk about it as this very abstract thing, but

40
00:02:40,476 --> 00:02:44,516
Speaker 1: in fact you need physical stuff for AI to exist.

41
00:02:44,716 --> 00:02:48,316
Speaker 1: You need in particular GPUs, And so if you want

42
00:02:48,316 --> 00:02:50,836
Speaker 1: to understand what's going on with AI. You need to

43
00:02:50,916 --> 00:02:54,436
Speaker 1: understand what's going on with GPUs, and Brannon is at

44
00:02:54,476 --> 00:02:55,156
Speaker 1: the center of that.

45
00:02:57,796 --> 00:02:59,836
Speaker 2: We grew up building computers, right like I grew up

46
00:02:59,836 --> 00:03:00,996
Speaker 2: building computers with my brother.

47
00:03:01,476 --> 00:03:03,316
Speaker 1: Like were you a gamer? Is that one? Yeah?

48
00:03:03,316 --> 00:03:03,916
Speaker 2: Absolutely?

49
00:03:03,996 --> 00:03:07,396
Speaker 1: Yes, yeah, like in then you were very early to GPUs.

50
00:03:07,596 --> 00:03:10,156
Speaker 2: Yes, yeah, in the nineties, like well as your Christmas

51
00:03:10,196 --> 00:03:12,836
Speaker 2: present it was to go pick your new hard drive,

52
00:03:13,236 --> 00:03:14,996
Speaker 2: or to get like a new sticker ram and bring

53
00:03:15,036 --> 00:03:16,676
Speaker 2: it into your computer, or to you know, raid your

54
00:03:16,716 --> 00:03:20,916
Speaker 2: uncle's office for their old compute equipment and take parts

55
00:03:20,916 --> 00:03:21,316
Speaker 2: out of it.

56
00:03:21,516 --> 00:03:21,996
Speaker 1: Uh huh.

57
00:03:22,516 --> 00:03:24,156
Speaker 2: It's just a really fun space to be.

58
00:03:24,196 --> 00:03:28,116
Speaker 1: Then let's start macro and go micro. Okay, so like macro, Like,

59
00:03:28,516 --> 00:03:32,236
Speaker 1: let me think of planet Earth, Like where on planet

60
00:03:32,356 --> 00:03:34,636
Speaker 1: Earth are the data centers that you are in?

61
00:03:35,836 --> 00:03:38,196
Speaker 2: So we operate in North America currently. I think you'll

62
00:03:38,196 --> 00:03:41,716
Speaker 2: see us expand globally within the next twelve eighteen months

63
00:03:41,996 --> 00:03:42,236
Speaker 2: or so.

64
00:03:42,276 --> 00:03:44,796
Speaker 1: Where in North America? Like how many data centers in?

65
00:03:45,556 --> 00:03:48,756
Speaker 2: Yeah? So in North America will be in about fourteen

66
00:03:48,836 --> 00:03:52,676
Speaker 2: data centers by Q one. Okay, So that's co located,

67
00:03:52,876 --> 00:03:56,596
Speaker 2: which means someone else built and operates the actual shell

68
00:03:56,716 --> 00:03:58,876
Speaker 2: and all the power and cooling and all these like

69
00:03:58,956 --> 00:04:01,916
Speaker 2: really engineering tins of components within that data center. Then

70
00:04:01,916 --> 00:04:06,036
Speaker 2: we come in and build out our infrastructure and the

71
00:04:06,076 --> 00:04:09,596
Speaker 2: infrastructure and the data center. It's one of the coolest

72
00:04:09,596 --> 00:04:11,836
Speaker 2: things you can ever walk into, right, It's like walking

73
00:04:11,836 --> 00:04:12,636
Speaker 2: into a spaceship.

74
00:04:12,876 --> 00:04:15,236
Speaker 1: I love it. So so tell me what it's like

75
00:04:15,276 --> 00:04:16,716
Speaker 1: when you go to one of these data senters. What

76
00:04:16,756 --> 00:04:18,636
Speaker 1: does it look like? Just like a giant warehouse out

77
00:04:18,676 --> 00:04:20,916
Speaker 1: in the middle of nowhere, somewhere where power is cheap.

78
00:04:22,276 --> 00:04:24,476
Speaker 2: Uh, you know, power's in there, but it's more about

79
00:04:24,676 --> 00:04:27,956
Speaker 2: redundancy of power. It's more about connectivity to like the

80
00:04:28,236 --> 00:04:31,516
Speaker 2: broader Internet backbone across North America. And it's about security

81
00:04:32,196 --> 00:04:33,876
Speaker 2: as well. So when you pull up to one of

82
00:04:33,876 --> 00:04:36,356
Speaker 2: these sites, you know it's it's twenty foot walls, it's

83
00:04:36,356 --> 00:04:40,236
Speaker 2: security guards, it's you know, multiple badges worth of security

84
00:04:40,316 --> 00:04:43,116
Speaker 2: to get back to where you need to. It's no cameras,

85
00:04:43,316 --> 00:04:47,916
Speaker 2: it's completely clean rooms like, no food, no drinks, Like

86
00:04:48,036 --> 00:04:50,716
Speaker 2: you can't have anything being pulled into the service, right

87
00:04:50,756 --> 00:04:54,076
Speaker 2: because there's so much airflow moving through you know, the

88
00:04:54,236 --> 00:04:54,876
Speaker 2: entire room.

89
00:04:55,316 --> 00:04:58,436
Speaker 1: Okay, so you badge, you badge the security guy, you park,

90
00:04:58,476 --> 00:05:01,196
Speaker 1: you get in, walk into the room. What's it look like?

91
00:05:01,356 --> 00:05:02,636
Speaker 1: Is it hot? Is allowed?

92
00:05:03,796 --> 00:05:07,636
Speaker 2: It is? It feels like it's screaming at you. It

93
00:05:07,756 --> 00:05:10,956
Speaker 2: is extremely loud. You can barely have a conversation. And

94
00:05:10,996 --> 00:05:15,316
Speaker 2: that's a reflection of the amount of power intensity ultimately, right,

95
00:05:15,356 --> 00:05:19,596
Speaker 2: because that's just fans pushing air through the servers and

96
00:05:20,676 --> 00:05:24,116
Speaker 2: very aggreat to cool them, to cool them, yes, because

97
00:05:24,116 --> 00:05:29,196
Speaker 2: they're consuming power on an extremely dense basis. Right, So

98
00:05:29,236 --> 00:05:31,196
Speaker 2: the more power you consume in a smaller area, the

99
00:05:31,236 --> 00:05:33,636
Speaker 2: more you know, socialated cooling you have to there. So

100
00:05:33,676 --> 00:05:36,996
Speaker 2: it's loud, and i'd say one of the more interesting

101
00:05:37,036 --> 00:05:42,356
Speaker 2: things that you're in there, it's the miles of fiber

102
00:05:42,396 --> 00:05:46,396
Speaker 2: optic that's required to transport information and connect everything. So

103
00:05:47,116 --> 00:05:51,876
Speaker 2: a typical let's say sixteen thousand GPU cluster that we build,

104
00:05:51,996 --> 00:05:56,596
Speaker 2: it has forty eight thousand discrete connections amongst it that

105
00:05:56,636 --> 00:05:58,996
Speaker 2: all have to be done perfectly. Then it has about

106
00:05:59,236 --> 00:06:04,676
Speaker 2: five hundred miles of fiber optic cabling run between all

107
00:06:04,676 --> 00:06:07,556
Speaker 2: those connections. So this is one sort of cluster that

108
00:06:07,636 --> 00:06:11,716
Speaker 2: you have built. Yeah, and it's like one part of

109
00:06:11,756 --> 00:06:13,916
Speaker 2: a bigger data center. Yeah, that's right.

110
00:06:14,076 --> 00:06:17,796
Speaker 1: Okay, So let's talk. Let's talk about the cluster. You

111
00:06:17,836 --> 00:06:20,996
Speaker 1: said it has forty eight thousand connections, meaning like does

112
00:06:21,036 --> 00:06:23,076
Speaker 1: a human being when you're building it, actually have to

113
00:06:23,156 --> 00:06:27,036
Speaker 1: go and like plug forty eight thousand things into forty

114
00:06:27,036 --> 00:06:30,156
Speaker 1: eight thousand things one at a time, no or yes.

115
00:06:29,956 --> 00:06:33,196
Speaker 2: Yes, yeah, And it all has to be done correctly.

116
00:06:33,276 --> 00:06:36,476
Speaker 2: And then all the components have to provision correctly. When

117
00:06:36,476 --> 00:06:38,076
Speaker 2: I say provisioning, that means they have to get their

118
00:06:38,116 --> 00:06:41,516
Speaker 2: you know, operational instructions, like you turn something on, it

119
00:06:41,556 --> 00:06:44,276
Speaker 2: doesn't know who it is, right, like what networks of

120
00:06:44,316 --> 00:06:46,596
Speaker 2: what it's supposed to do. So you have to develop

121
00:06:46,596 --> 00:06:49,196
Speaker 2: a bunch of bunch of software that we've done that

122
00:06:49,636 --> 00:06:53,876
Speaker 2: informs the machine. Hey, hello world, here's what we'd like

123
00:06:53,916 --> 00:06:56,116
Speaker 2: for you to do. So that that's called provisioning.

124
00:06:57,516 --> 00:07:01,396
Speaker 1: Okay, so forty eight thousand connections and how many GPUs

125
00:07:01,396 --> 00:07:02,436
Speaker 1: did you say are in this cluster?

126
00:07:03,036 --> 00:07:05,796
Speaker 2: So that that's for a sixteen thousand GPU clubs.

127
00:07:06,156 --> 00:07:08,716
Speaker 1: And so so we've started at the macro, We started

128
00:07:08,716 --> 00:07:11,996
Speaker 1: a planning that we're getting, then we got the data center,

129
00:07:12,076 --> 00:07:14,476
Speaker 1: and then we got your cluster, and then now let's

130
00:07:14,476 --> 00:07:18,596
Speaker 1: get to that GPU because like that is a really interesting.

131
00:07:20,116 --> 00:07:22,196
Speaker 1: It's the center of this whole thing, right in this

132
00:07:22,276 --> 00:07:24,916
Speaker 1: cluster you're talking about, Is it, in fact the.

133
00:07:25,196 --> 00:07:27,836
Speaker 2: Nvidia H one hundred, Yes, oh go, that's right.

134
00:07:28,396 --> 00:07:31,156
Speaker 1: So I want to talk about the Nvidia H one

135
00:07:31,196 --> 00:07:33,556
Speaker 1: hundred for a minute, because like it's a big deal

136
00:07:33,596 --> 00:07:37,476
Speaker 1: in the world right now, right like everybody wants them.

137
00:07:37,716 --> 00:07:40,996
Speaker 1: Apparently you can't get them, or you can't get as

138
00:07:41,036 --> 00:07:44,956
Speaker 1: many as you want. You can search for Nvidia H

139
00:07:44,996 --> 00:07:47,996
Speaker 1: one hundred on Amazon and it comes up, and the

140
00:07:47,996 --> 00:07:51,796
Speaker 1: first one that came up for me costs thirty seven thousand,

141
00:07:51,996 --> 00:07:56,436
Speaker 1: seven hundred and ninety dollars, doesn't say weather shipping is included.

142
00:07:57,396 --> 00:08:00,316
Speaker 1: And that's just one of them. That's just one of them.

143
00:08:00,916 --> 00:08:03,436
Speaker 1: And you're talking about how many in this cluster.

144
00:08:04,396 --> 00:08:07,636
Speaker 2: So that would be sixteen thousand and a sixteen thousand

145
00:08:07,636 --> 00:08:10,316
Speaker 2: GPU cluster, and we're you know, we're building multiple of those.

146
00:08:10,956 --> 00:08:14,516
Speaker 1: And so presumably you're not paying retail one hopes you're

147
00:08:14,556 --> 00:08:18,876
Speaker 1: not paying thirty seven thousand dollars per But but what

148
00:08:18,996 --> 00:08:21,996
Speaker 1: is the order of magnitude of what the what the

149
00:08:22,036 --> 00:08:24,796
Speaker 1: GPUs cost you in one of these clusters.

150
00:08:24,836 --> 00:08:26,036
Speaker 2: We're not going to talk about that.

151
00:08:26,076 --> 00:08:30,036
Speaker 1: Okay, but Let's be clear, I'm just the same as

152
00:08:30,076 --> 00:08:31,636
Speaker 1: everyone else's thirty seven thousand.

153
00:08:32,276 --> 00:08:34,756
Speaker 2: Yeah, it's it's less. It's it's going to be comparable

154
00:08:34,836 --> 00:08:37,716
Speaker 2: with the what the hyperscalers payers, or so we think

155
00:08:37,916 --> 00:08:39,036
Speaker 2: right end of the day, we don't.

156
00:08:38,916 --> 00:08:41,836
Speaker 1: Really know how much does it cost to build one

157
00:08:41,916 --> 00:08:45,356
Speaker 1: of these clusters, like to an order of magnitude? Sure,

158
00:08:45,396 --> 00:08:47,916
Speaker 1: one hundred million dollars, is that the right order of magnitude.

159
00:08:48,996 --> 00:08:51,076
Speaker 2: It's in the hundreds of millions of dollars, that's.

160
00:08:50,916 --> 00:08:52,756
Speaker 1: Correct to build one of these clusters.

161
00:08:52,836 --> 00:08:55,156
Speaker 2: It's a built one sixteen thousand GPU cluster. But to

162
00:08:55,236 --> 00:08:57,316
Speaker 2: be clear, like we build clusters of different sizes, we

163
00:08:57,396 --> 00:09:00,716
Speaker 2: build multiples of these clusters. It's it's all kind of

164
00:09:00,756 --> 00:09:04,236
Speaker 2: based on what that end client is looking for. And

165
00:09:04,596 --> 00:09:07,036
Speaker 2: you know we do on demand compute. May all kind

166
00:09:07,076 --> 00:09:10,156
Speaker 2: of qualified as our our capital expins at the company

167
00:09:10,236 --> 00:09:12,436
Speaker 2: levels measured in the billions of dollars a year.

168
00:09:12,476 --> 00:09:15,236
Speaker 1: Sure, it's a very capital intensive business because you have

169
00:09:15,356 --> 00:09:18,316
Speaker 1: to buy and for the fundamental reason that you have

170
00:09:18,436 --> 00:09:22,236
Speaker 1: to buy these GPUs that everybody wants that are profoundly expensive.

171
00:09:22,396 --> 00:09:22,956
Speaker 2: That's correct.

172
00:09:23,676 --> 00:09:26,076
Speaker 1: I mean an interesting twist in this part of the

173
00:09:26,116 --> 00:09:28,836
Speaker 1: story is you know, in Vidia is this really interesting

174
00:09:28,876 --> 00:09:31,116
Speaker 1: company right now. Their stock has gone crazy this year

175
00:09:31,156 --> 00:09:34,596
Speaker 1: because they make the GPUs that are in wild demand

176
00:09:34,756 --> 00:09:37,676
Speaker 1: right now, and they are also an investor in your company,

177
00:09:37,796 --> 00:09:41,956
Speaker 1: right which is very interesting. Like how does that piece

178
00:09:42,036 --> 00:09:44,396
Speaker 1: of it, that in Vidia investment in core weave, Like,

179
00:09:44,396 --> 00:09:47,276
Speaker 1: how does that work in terms of the broader dynamics.

180
00:09:48,116 --> 00:09:51,476
Speaker 2: Yeah, I think you see strategic investors on cap tables

181
00:09:51,756 --> 00:09:56,676
Speaker 2: pretty often. We have a differentiated product, We have a

182
00:09:56,796 --> 00:10:00,436
Speaker 2: fantastic way to get their infrastructure into the market, but

183
00:10:01,276 --> 00:10:04,316
Speaker 2: we don't have any preferential treatment. We don't have any prioritization.

184
00:10:04,876 --> 00:10:07,476
Speaker 1: We're just not easier for you to get the GPUs.

185
00:10:07,756 --> 00:10:09,436
Speaker 2: No, it is not easier for us to get the

186
00:10:09,516 --> 00:10:10,676
Speaker 2: GPUs whatsoever.

187
00:10:12,036 --> 00:10:15,996
Speaker 1: So these GPUs, these Nvidia GPUs, why is this the

188
00:10:16,076 --> 00:10:19,276
Speaker 1: one that everybody wants? What's so special about the H

189
00:10:19,356 --> 00:10:19,836
Speaker 1: one hundred?

190
00:10:20,356 --> 00:10:22,356
Speaker 2: So Jensen and the Vidia team did this.

191
00:10:22,796 --> 00:10:25,196
Speaker 1: Jensen is the founder c Yeah.

192
00:10:25,396 --> 00:10:30,676
Speaker 2: Yes, did this amazing job of enabling the artificial intelligence

193
00:10:30,676 --> 00:10:33,756
Speaker 2: space from an early early onset, like back in the

194
00:10:33,796 --> 00:10:37,916
Speaker 2: early twenty tens, by releasing the drivers, the software that

195
00:10:38,036 --> 00:10:41,836
Speaker 2: actually control the GPUs into the market and making it

196
00:10:42,076 --> 00:10:45,836
Speaker 2: accessible to those developers. Right, so when you go into

197
00:10:46,036 --> 00:10:49,556
Speaker 2: like GitHub, for example, you search AI project, they're all

198
00:10:49,556 --> 00:10:52,836
Speaker 2: going to come back referencing Kuda drivers, which is the

199
00:10:53,116 --> 00:10:55,076
Speaker 2: software in Cressent right there, using x eighty six.

200
00:10:55,356 --> 00:10:59,436
Speaker 1: So Nvidia very cleverly sort of created the language that

201
00:10:59,556 --> 00:11:01,796
Speaker 1: most of the people who write software, who write AI

202
00:11:01,916 --> 00:11:05,476
Speaker 1: software speak essentially you, that's totally right. And so if

203
00:11:05,676 --> 00:11:08,036
Speaker 1: there might be another chip where the sort of specs

204
00:11:08,196 --> 00:11:12,476
Speaker 1: are comparable, but the language is different, and everybody speaks

205
00:11:12,556 --> 00:11:15,236
Speaker 1: this one language and there's sort of a network effect

206
00:11:15,236 --> 00:11:17,596
Speaker 1: at play where it's like everybody develops on this chip

207
00:11:17,636 --> 00:11:19,916
Speaker 1: because everybody develops on this chip, and so that is

208
00:11:20,076 --> 00:11:23,036
Speaker 1: like a classic killer network effect.

209
00:11:24,156 --> 00:11:26,156
Speaker 2: Yep, they just did an amazing job.

210
00:11:27,876 --> 00:11:33,996
Speaker 1: Interesting, you're competing against Amazon, Microsoft, Google, and Oracle, so

211
00:11:34,436 --> 00:11:36,636
Speaker 1: more or less the biggest companies in the world. Yes,

212
00:11:36,756 --> 00:11:41,716
Speaker 1: you're like four guys who were just in a warehouse

213
00:11:41,756 --> 00:11:45,996
Speaker 1: in New Jersey, Like what what happened there?

214
00:11:46,156 --> 00:11:46,196
Speaker 2: Like?

215
00:11:46,276 --> 00:11:47,116
Speaker 1: How is that happening?

216
00:11:48,716 --> 00:11:52,356
Speaker 2: It's ultimately it was through that specialization dot that really

217
00:11:52,436 --> 00:11:55,516
Speaker 2: goddess thing. Like you have an at scale existing market

218
00:11:55,756 --> 00:11:58,116
Speaker 2: and you figure out how to do something better. You've

219
00:11:58,276 --> 00:12:03,196
Speaker 2: decommoditized a market through performance. So the analogy I like

220
00:12:03,356 --> 00:12:05,396
Speaker 2: is like, we figured out how to refine a barrel

221
00:12:05,436 --> 00:12:06,036
Speaker 2: of oil better.

222
00:12:06,156 --> 00:12:09,756
Speaker 1: Yeah, so everyone has to come use us because you

223
00:12:09,876 --> 00:12:11,956
Speaker 1: can do it more efficiently, more cheaply.

224
00:12:12,276 --> 00:12:14,876
Speaker 2: Yeah, which means it's more efficient for our clients. At

225
00:12:14,876 --> 00:12:15,436
Speaker 2: the end of the day.

226
00:12:15,636 --> 00:12:20,276
Speaker 1: Presumably these companies are spending billions, tens of billions, maybe

227
00:12:20,436 --> 00:12:22,476
Speaker 1: hundreds of billions of dollars to catch up with you.

228
00:12:23,476 --> 00:12:28,156
Speaker 1: I mean, I wouldn't be inclined to bet against them, respectfully.

229
00:12:28,316 --> 00:12:31,076
Speaker 2: I couldn't agree with you more. They absolutely Isn't that

230
00:12:31,236 --> 00:12:31,676
Speaker 2: bad for you?

231
00:12:31,836 --> 00:12:32,756
Speaker 1: Isn't that scary for you?

232
00:12:33,036 --> 00:12:33,076
Speaker 2: Like?

233
00:12:33,156 --> 00:12:35,316
Speaker 1: Why are you talking about to be like sprinting to

234
00:12:35,396 --> 00:12:37,996
Speaker 1: try and not get caught by Google and Amazon?

235
00:12:38,996 --> 00:12:42,716
Speaker 2: Look, these are amazing companies with amazing depth of engineering

236
00:12:42,836 --> 00:12:45,716
Speaker 2: talent and a massive amount of resources. To your point,

237
00:12:45,716 --> 00:12:48,276
Speaker 2: they're the largest companies on the Planet's super smart. It's

238
00:12:48,516 --> 00:12:49,156
Speaker 2: very good.

239
00:12:48,996 --> 00:12:51,116
Speaker 1: At building big, efficient things.

240
00:12:52,276 --> 00:12:56,116
Speaker 2: Yes, but they are generalized. These are huge teams, and

241
00:12:56,236 --> 00:12:58,436
Speaker 2: we both know how long it takes to move a

242
00:12:58,556 --> 00:13:02,436
Speaker 2: big ship in a new direction. And I mean, to

243
00:13:02,516 --> 00:13:04,156
Speaker 2: put a sense of scale to it, They're working with

244
00:13:04,236 --> 00:13:07,836
Speaker 2: tens of thousands of engineers. We're in the hundreds of engineers,

245
00:13:07,876 --> 00:13:10,596
Speaker 2: and it just allows for us to focus on a

246
00:13:10,836 --> 00:13:13,876
Speaker 2: task much more discreetly. This is all we do. We

247
00:13:14,076 --> 00:13:19,276
Speaker 2: build supercomputers. I'm not trying to serve every single corner

248
00:13:19,396 --> 00:13:22,156
Speaker 2: of the cloud. I'm trying to serve the part of

249
00:13:22,236 --> 00:13:24,796
Speaker 2: the cloud that needs access to supercomputers, and we've just

250
00:13:24,876 --> 00:13:26,356
Speaker 2: become really good at doing it.

251
00:13:30,996 --> 00:13:32,956
Speaker 1: We'll be back in a minute after a break for

252
00:13:33,076 --> 00:13:54,836
Speaker 1: some ants. I know some of your clients I've read about, right,

253
00:13:54,996 --> 00:13:57,596
Speaker 1: I imagine some might be private, but some are public, right, Like, so,

254
00:13:57,956 --> 00:13:59,676
Speaker 1: who are some of the people who are using the

255
00:13:59,756 --> 00:14:00,676
Speaker 1: things that you have built?

256
00:14:01,196 --> 00:14:04,556
Speaker 2: Some of the public ones. Inflection AI is one that

257
00:14:05,156 --> 00:14:08,516
Speaker 2: we really enjoy working with that team, especially on Jeanie most.

258
00:14:08,396 --> 00:14:09,956
Speaker 1: Of us solely my right, who was one of the

259
00:14:09,996 --> 00:14:11,956
Speaker 1: founders of DeepMind and who just wrote a book and

260
00:14:11,956 --> 00:14:14,396
Speaker 1: who are trying to book for the show. That's his company, right,

261
00:14:14,436 --> 00:14:16,476
Speaker 1: that's the company that he founded. That's correct Mind, And

262
00:14:16,556 --> 00:14:19,316
Speaker 1: so they are training like a big sort of a

263
00:14:19,716 --> 00:14:22,956
Speaker 1: GBT size model, right, and they're using your computers to

264
00:14:22,996 --> 00:14:23,196
Speaker 1: do it.

265
00:14:24,316 --> 00:14:25,396
Speaker 2: That's right, yeap?

266
00:14:25,516 --> 00:14:26,796
Speaker 1: Who else you know?

267
00:14:26,916 --> 00:14:29,196
Speaker 2: A smaller team that we just love. The engineers at

268
00:14:29,276 --> 00:14:34,236
Speaker 2: is a company called novel Ai. Novel Ai has kind

269
00:14:34,236 --> 00:14:36,916
Speaker 2: of a storytelling based inference product and market. They do

270
00:14:36,996 --> 00:14:40,516
Speaker 2: text to image, text to story. But there's we have

271
00:14:40,716 --> 00:14:43,196
Speaker 2: over six hundred AI clients on our platform today.

272
00:14:43,796 --> 00:14:47,556
Speaker 1: And are you still doing you know, drug discovery stuff.

273
00:14:47,556 --> 00:14:49,796
Speaker 1: Are you still doing entertainment stuff? Or has AI? Is

274
00:14:49,836 --> 00:14:52,436
Speaker 1: the demand for AI and willingness to pay just so

275
00:14:52,636 --> 00:14:54,276
Speaker 1: high that it's mostly.

276
00:14:54,036 --> 00:14:56,756
Speaker 2: That now, that's just where all the attention is is

277
00:14:56,836 --> 00:14:59,556
Speaker 2: on AI. But we're absolutely cruiting clients on the media

278
00:14:59,596 --> 00:15:02,756
Speaker 2: and entertainment side and the drug discovery side.

279
00:15:02,756 --> 00:15:05,316
Speaker 1: And is it getting way more expensive for those clients

280
00:15:05,476 --> 00:15:07,676
Speaker 1: because of the AI demand? Are they in the sort

281
00:15:07,716 --> 00:15:11,076
Speaker 1: of unfortunate kind of backwash of like this thing they

282
00:15:11,156 --> 00:15:13,716
Speaker 1: need to use just got wildly more expensive because there's

283
00:15:13,756 --> 00:15:16,076
Speaker 1: so much demand and so much money flowing in for

284
00:15:16,196 --> 00:15:17,156
Speaker 1: these other use cases.

285
00:15:17,756 --> 00:15:21,356
Speaker 2: We have not increased pricing due to that. I think

286
00:15:21,756 --> 00:15:24,316
Speaker 2: what you've seen instead is just it's just more difficult

287
00:15:24,316 --> 00:15:24,956
Speaker 2: for people to get.

288
00:15:24,836 --> 00:15:27,356
Speaker 1: To Supply is constrained. Price hasn't gone up, but supply

289
00:15:27,476 --> 00:15:29,876
Speaker 1: is constrained. Well, let's let's talk about that, because supply

290
00:15:29,996 --> 00:15:32,796
Speaker 1: constraints are like an interesting part of the story right there.

291
00:15:32,876 --> 00:15:35,836
Speaker 1: Was a quote from a guy at OpenAI who's like,

292
00:15:35,956 --> 00:15:38,116
Speaker 1: what people, what my friends talk about when we get

293
00:15:38,156 --> 00:15:40,716
Speaker 1: together is like, how are we going to get more GPUs?

294
00:15:40,836 --> 00:15:42,956
Speaker 1: Like have you heard where are their GPUs? It's like

295
00:15:43,396 --> 00:15:45,876
Speaker 1: it's like mad Max or something, but instead of fuel,

296
00:15:46,276 --> 00:15:50,876
Speaker 1: what everybody needs is GPUs. Like that is really interesting, right,

297
00:15:50,956 --> 00:15:55,396
Speaker 1: Like one doesn't think of technology these you know, tech

298
00:15:55,476 --> 00:15:59,476
Speaker 1: firms with all the money they could want constrained because

299
00:15:59,556 --> 00:16:04,196
Speaker 1: they just can't get these chips, these GPUs at any price.

300
00:16:04,356 --> 00:16:06,676
Speaker 1: Like that's a really interesting state of affairs.

301
00:16:07,796 --> 00:16:10,996
Speaker 2: Yes, and I think the driver ultimately look like AI

302
00:16:11,156 --> 00:16:14,556
Speaker 2: software adoption is the steepest adoption curve we've ever seen,

303
00:16:15,396 --> 00:16:19,836
Speaker 2: and we're trying to build supercomputers at that pace, and

304
00:16:19,836 --> 00:16:23,276
Speaker 2: it's a massive logistical and engineering challenge build to keep

305
00:16:23,396 --> 00:16:23,916
Speaker 2: pace with that.

306
00:16:24,356 --> 00:16:27,516
Speaker 1: Duct right, stuff doesn't scale like tof.

307
00:16:27,636 --> 00:16:30,956
Speaker 2: Is a massive engineering challenge as well, Like it.

308
00:16:31,036 --> 00:16:33,316
Speaker 1: Are the GPUs the rate limiting step? I mean, even

309
00:16:33,316 --> 00:16:34,796
Speaker 1: if you had them, it would be hard, But is

310
00:16:34,876 --> 00:16:39,156
Speaker 1: the fundamental it's a good question bottleneck the GPUs.

311
00:16:40,756 --> 00:16:44,236
Speaker 2: GPUs themselves are an extremely high demand, but increasingly it's

312
00:16:44,356 --> 00:16:48,076
Speaker 2: data center space, which which we've found has been extremely

313
00:16:48,236 --> 00:16:50,956
Speaker 2: interesting because data center space is an area that's been

314
00:16:51,076 --> 00:16:54,596
Speaker 2: undercapitalized or underinvested in for probably the last decade or so,

315
00:16:55,276 --> 00:16:59,036
Speaker 2: and there's only a certain number of data centers that

316
00:16:59,116 --> 00:17:03,716
Speaker 2: can actually run this infrastructure due to that increase in

317
00:17:03,956 --> 00:17:09,436
Speaker 2: power density relative to kind of legacy infratis sructual.

318
00:17:09,476 --> 00:17:12,316
Speaker 1: Right, So, like in this given amount of space, the

319
00:17:12,676 --> 00:17:16,116
Speaker 1: kinds of clusters you're building, kinds of machines you're building,

320
00:17:16,116 --> 00:17:18,116
Speaker 1: are going to use way more power. And that's a

321
00:17:18,156 --> 00:17:20,316
Speaker 1: problem because these in a given amount of these BEATA

322
00:17:20,356 --> 00:17:23,956
Speaker 1: centers aren't set up to provide that much power per

323
00:17:24,116 --> 00:17:24,876
Speaker 1: unit space.

324
00:17:24,716 --> 00:17:28,156
Speaker 2: And cooling most more importantly as well, like you have

325
00:17:28,316 --> 00:17:30,076
Speaker 2: to be able to cool that area, and.

326
00:17:30,116 --> 00:17:32,476
Speaker 1: They generate more heat because they consume more.

327
00:17:32,396 --> 00:17:36,036
Speaker 2: Power exactly right. So there's only a certain number of

328
00:17:36,076 --> 00:17:38,236
Speaker 2: data centers in the US I'm really speaking to the

329
00:17:38,356 --> 00:17:41,996
Speaker 2: US here that can run that. And now that's the bottleneck,

330
00:17:42,036 --> 00:17:43,436
Speaker 2: and that's going to be the bottleneck for the next

331
00:17:43,476 --> 00:17:47,196
Speaker 2: two years arguably, because like all that space has now

332
00:17:47,236 --> 00:17:47,756
Speaker 2: been leased.

333
00:17:48,156 --> 00:17:49,716
Speaker 1: What are you trying to figure out that you have

334
00:17:49,836 --> 00:17:50,716
Speaker 1: not yet figured out?

335
00:17:51,996 --> 00:17:56,636
Speaker 2: It's that scaling, it's that onslaught of demand. I think

336
00:17:56,716 --> 00:18:00,396
Speaker 2: we're we have it figured out, We know the engineering,

337
00:18:00,636 --> 00:18:03,276
Speaker 2: we have the right product, we have the capital, the

338
00:18:03,396 --> 00:18:05,396
Speaker 2: funding to go do it. We have the right people

339
00:18:05,476 --> 00:18:08,236
Speaker 2: on the ground, we have the right management team. Now

340
00:18:08,316 --> 00:18:12,076
Speaker 2: it's just growth like we have the next twelve to

341
00:18:12,116 --> 00:18:15,596
Speaker 2: eighteen months in front of us outlined, and it's just

342
00:18:15,916 --> 00:18:19,556
Speaker 2: go build, like, be focused on building. It's arguably some

343
00:18:19,636 --> 00:18:22,916
Speaker 2: of the largest technology infrastructure projects and the world are

344
00:18:22,956 --> 00:18:24,116
Speaker 2: happening at our company.

345
00:18:24,436 --> 00:18:26,116
Speaker 1: When you think about like the next year, you have

346
00:18:26,236 --> 00:18:30,596
Speaker 1: to spend billions of dollars, billions of dollars, it's like

347
00:18:30,756 --> 00:18:34,316
Speaker 1: three guys who are playing around in the garage four

348
00:18:34,436 --> 00:18:39,876
Speaker 1: years ago spending billions of dollars, like you worried you

349
00:18:39,996 --> 00:18:41,276
Speaker 1: might do it wrong.

350
00:18:42,676 --> 00:18:46,956
Speaker 2: If I worried, so, I think it's an excellent question.

351
00:18:47,516 --> 00:18:52,756
Speaker 2: But we've also had some of the most well informed,

352
00:18:53,276 --> 00:18:57,516
Speaker 2: intelligent participants in the artificial intelligence space look at our

353
00:18:57,596 --> 00:19:00,596
Speaker 2: engineering teams, look at our product and say that's who

354
00:19:00,596 --> 00:19:03,756
Speaker 2: I'm going with. We're doing it every day, We're building

355
00:19:03,796 --> 00:19:06,076
Speaker 2: it every day, and now it's just keep doing it,

356
00:19:06,196 --> 00:19:09,596
Speaker 2: remain focused, keep building. Don't get this dude.

357
00:19:11,156 --> 00:19:14,156
Speaker 1: Now you're just grinding me like you seem super successful,

358
00:19:14,196 --> 00:19:15,556
Speaker 1: but you're telling me is now you just have to

359
00:19:16,036 --> 00:19:18,276
Speaker 1: grind it out for a year and like every day.

360
00:19:18,356 --> 00:19:26,276
Speaker 2: Grind it out and keep growing. Yes, we'll be back in.

361
00:19:26,316 --> 00:19:41,196
Speaker 1: A minute with the lightning Okay, now let's do the

362
00:19:41,276 --> 00:19:45,236
Speaker 1: lightning round. How many screens do you have? How many

363
00:19:45,276 --> 00:19:47,956
Speaker 1: monitors do you have on your computer? Six? It's like

364
00:19:48,116 --> 00:19:50,516
Speaker 1: former commodities trader once, commodities trader always.

365
00:19:50,836 --> 00:19:55,396
Speaker 2: I can't help it. I literally can't help it.

366
00:19:55,476 --> 00:19:57,756
Speaker 1: It's so like one screen to you, you feel like

367
00:19:57,796 --> 00:20:00,236
Speaker 1: you're looking at an Apple Watch or something. One monitor.

368
00:20:01,716 --> 00:20:03,316
Speaker 1: Do you use all six monitors?

369
00:20:04,356 --> 00:20:07,436
Speaker 2: I do? I use all six. Now, whether it's all

370
00:20:07,476 --> 00:20:09,156
Speaker 2: relevant at the same time, I don't know, but but

371
00:20:09,876 --> 00:20:10,556
Speaker 2: I do use.

372
00:20:10,396 --> 00:20:12,556
Speaker 1: Also looking at six different things.

373
00:20:13,396 --> 00:20:18,196
Speaker 2: No, it's like dedicated chat Windows. There's dedicated you know, browser, Windows,

374
00:20:18,276 --> 00:20:22,196
Speaker 2: information Windows. I can get stats on all of our infrastructure.

375
00:20:23,956 --> 00:20:26,116
Speaker 2: I just can't help it. I can't get that commodity

376
00:20:26,156 --> 00:20:26,756
Speaker 2: trader out of me.

377
00:20:27,636 --> 00:20:28,476
Speaker 1: Will you go to eight?

378
00:20:29,316 --> 00:20:33,556
Speaker 2: I've been at eight. I stepped down to six. Six.

379
00:20:34,716 --> 00:20:39,916
Speaker 1: It hurt me a bit it As a former commodities trader,

380
00:20:40,116 --> 00:20:44,076
Speaker 1: what do you understand about the world that most people don't?

381
00:20:46,036 --> 00:20:52,196
Speaker 2: Risk management say more so acknowledging that there's always going

382
00:20:52,276 --> 00:20:56,396
Speaker 2: to be trade offs and what you do, but reducing

383
00:20:56,596 --> 00:21:00,076
Speaker 2: the potential downside of that trade off as much as

384
00:21:00,156 --> 00:21:06,036
Speaker 2: possible through through hedging, through negotiations, through contractual agreements, and

385
00:21:06,436 --> 00:21:07,956
Speaker 2: being able to filter out the noise.

386
00:21:09,996 --> 00:21:17,596
Speaker 1: Overrated or underrated? Bitcoin underrated, Ethereum overrated. I would not

387
00:21:17,916 --> 00:21:23,236
Speaker 1: have guessed Bitcoin underrated, Ethereum overrated, given in particular your

388
00:21:23,276 --> 00:21:26,996
Speaker 1: experience that you started your company mining ethereum. Why why

389
00:21:27,156 --> 00:21:29,276
Speaker 1: is eum overrated and why is Bitcoin underrated?

390
00:21:29,996 --> 00:21:33,196
Speaker 2: I think ethereum had so many promises.

391
00:21:32,756 --> 00:21:34,436
Speaker 1: Of what it would do global computer.

392
00:21:35,516 --> 00:21:38,356
Speaker 2: I just what are they delivered on. I just haven't

393
00:21:38,396 --> 00:21:43,476
Speaker 2: seen like any real use cases. They've never got enterprise adoption, nothing,

394
00:21:43,516 --> 00:21:45,956
Speaker 2: whereas Bitcoin is getting government It was.

395
00:21:45,956 --> 00:21:50,236
Speaker 1: Supposed to be new global money. It was. So why

396
00:21:50,276 --> 00:21:51,316
Speaker 1: do you think it's underrated?

397
00:21:53,556 --> 00:21:56,596
Speaker 2: Because I think it's still moving that direction. It has

398
00:21:56,676 --> 00:22:02,316
Speaker 2: the potential for it. It remains really interesting. It's slow moving,

399
00:22:03,516 --> 00:22:06,396
Speaker 2: but it's it's making progress, whereas I feel like Ethereum

400
00:22:06,436 --> 00:22:10,636
Speaker 2: has kind of been more sideways, if anything, interesting to down.

401
00:22:10,516 --> 00:22:13,116
Speaker 1: Even what's one time when you gave up on something?

402
00:22:14,476 --> 00:22:16,676
Speaker 2: Oh my gosh, I mean in the trading world, every

403
00:22:16,716 --> 00:22:19,596
Speaker 2: single day, like.

404
00:22:19,636 --> 00:22:22,036
Speaker 1: You abandon a thesis, basically abandon a trade.

405
00:22:22,156 --> 00:22:25,396
Speaker 2: Is that? Yeah, I mean look like you're you're wrong

406
00:22:25,756 --> 00:22:27,476
Speaker 2: every single day. And that was like one of the

407
00:22:27,556 --> 00:22:30,676
Speaker 2: really interesting kind of outcomes of spending a decade in

408
00:22:30,756 --> 00:22:33,956
Speaker 2: that career is is learning how to be wrong. You're

409
00:22:33,996 --> 00:22:36,276
Speaker 2: wrong all the time, but it's it's like, how do

410
00:22:36,396 --> 00:22:39,476
Speaker 2: you admit you're wrong, accept it, learn from it, and

411
00:22:39,596 --> 00:22:41,836
Speaker 2: move on to the next. And I thought that the

412
00:22:41,876 --> 00:22:45,596
Speaker 2: commodity trading world was just such an fascinating platform for that.

413
00:22:46,156 --> 00:22:48,116
Speaker 2: Just get it reinforced every single day. It's not like

414
00:22:48,156 --> 00:22:53,196
Speaker 2: a quarterly performance, It's an intra day, twenty times a day.

415
00:22:53,396 --> 00:22:55,636
Speaker 1: Sing it on eight screens in front of your face

416
00:22:55,716 --> 00:22:56,396
Speaker 1: in real time.

417
00:22:56,556 --> 00:23:00,316
Speaker 2: I said, red numbers telling you how badly you're doing.

418
00:23:00,436 --> 00:23:04,716
Speaker 2: Like it's immediately quantified. So I've been wrong a lot.

419
00:23:05,236 --> 00:23:07,036
Speaker 1: Great, I appreciate your time.

420
00:23:07,716 --> 00:23:09,996
Speaker 2: Yeah, that was fantastic. Really enjoyed the conversation.

421
00:23:15,876 --> 00:23:19,276
Speaker 1: Brandon McNee is the chief strategy Officer at Corweath

422
00:23:21,276 --> 00:23:21,316
Speaker 2: M