1
00:00:02,750 --> 00:00:04,430
Speaker 1: Bloomberg Audio Studios.

2
00:00:04,890 --> 00:00:05,569
Speaker 2: Podcasts.

3
00:00:05,890 --> 00:00:06,430
Speaker 3: Radio.

4
00:00:06,850 --> 00:00:07,270
Speaker 4: News.

5
00:00:18,060 --> 00:00:21,619
Speaker 1: Hello and welcome to another episode of the Odd Lodge Podcast.

6
00:00:21,720 --> 00:00:22,700
Speaker 1: I'm Joe Weisenthal.

7
00:00:22,940 --> 00:00:23,940
Speaker 4: And I'm Tracy Allaway.

8
00:00:24,200 --> 00:00:27,270
Speaker 1: Tracy, we're here in Chicago, second year in a row

9
00:00:27,290 --> 00:00:29,510
Speaker 1: that we've done a live show here in a Great

10
00:00:29,570 --> 00:00:30,180
Speaker 1: city for us.

11
00:00:30,250 --> 00:00:30,610
Speaker 3: That's right.

12
00:00:30,750 --> 00:00:32,550
Speaker 4: We're going to be talking to someone who we spoke

13
00:00:32,590 --> 00:00:35,379
Speaker 4: with last year, but a lot has changed in 12 months.

14
00:00:36,800 --> 00:00:39,150
Speaker 4: This is someone who is trying to build a market

15
00:00:39,310 --> 00:00:40,690
Speaker 4: for compute, right?

16
00:00:40,850 --> 00:00:41,030
Speaker 3: Right.

17
00:00:41,050 --> 00:00:42,810
Speaker 1: So we're going to be speaking with Don Wilson. He

18
00:00:42,830 --> 00:00:46,070
Speaker 1: is the founder and CEO of DRW. He is really

19
00:00:46,229 --> 00:00:49,850
Speaker 1: at the forefront of this effort to build futures markets

20
00:00:49,890 --> 00:00:52,810
Speaker 1: for things like GPUs and so forth. So check out

21
00:00:52,830 --> 00:00:56,030
Speaker 1: the conversation. We talked to you, Don, basically exactly a

22
00:00:56,050 --> 00:00:57,930
Speaker 1: year ago, last time we were here. We talked about

23
00:00:57,950 --> 00:01:01,770
Speaker 1: things like GPU futures. But in like AI time, it

24
00:01:01,800 --> 00:01:03,800
Speaker 1: might as well have been 10 years ago because so

25
00:01:03,840 --> 00:01:07,369
Speaker 1: much has changed. We want to talk GPU market structure,

26
00:01:07,430 --> 00:01:08,030
Speaker 1: all that stuff.

27
00:01:08,390 --> 00:01:09,570
Speaker 2: The last couple of days.

28
00:01:09,390 --> 00:01:12,419
Speaker 1: Have actually been kind of interesting in the AI trade world.

29
00:01:12,940 --> 00:01:17,060
Speaker 1: Everyone's watching CoreWeave. Everyone's watching CoreWeave bonds. There was an

30
00:01:17,200 --> 00:01:19,819
Speaker 1: anthropic investor meeting before the IPO.

31
00:01:19,840 --> 00:01:21,179
Speaker 3: A little jitters.

32
00:01:21,280 --> 00:01:25,770
Speaker 1: Maybe their revenue only grew up, whatever, 50% versus 51%.

33
00:01:25,770 --> 00:01:29,130
Speaker 1: I don't know the exact numbers. Give us your read

34
00:01:29,270 --> 00:01:33,060
Speaker 1: right now. Is there a little slowdown happening in AI?

35
00:01:33,530 --> 00:01:35,699
Speaker 1: Are we shifting into a lower gear? Is there anything

36
00:01:35,740 --> 00:01:37,720
Speaker 1: in the trajectory that's changed?

37
00:01:38,120 --> 00:01:43,000
Speaker 2: Any new thing that's growing this fast is not going

38
00:01:43,020 --> 00:01:49,110
Speaker 2: to grow at a constant rate. The first derivative isn't

39
00:01:49,120 --> 00:01:51,230
Speaker 2: going to be constant. The second derivative isn't going to

40
00:01:51,250 --> 00:01:54,310
Speaker 2: be constant. So to me, this is just kind of

41
00:01:54,710 --> 00:01:59,340
Speaker 2: expected volatility. I don't know, maybe people went on vacation

42
00:01:59,380 --> 00:02:02,420
Speaker 2: in August so they didn't consume as many anthropic tokens.

43
00:02:03,980 --> 00:02:06,270
Speaker 1: But you're not reading anything more into it than that.

44
00:02:06,390 --> 00:02:07,490
Speaker 3: I'm not. OK.

45
00:02:08,070 --> 00:02:11,889
Speaker 4: Wait, so with something like CoreWeave, a big part of

46
00:02:11,930 --> 00:02:14,990
Speaker 4: what you're doing with Compute Exchange and silicon data is

47
00:02:15,030 --> 00:02:17,400
Speaker 4: this idea that if you create a market for compute,

48
00:02:17,900 --> 00:02:21,560
Speaker 4: then cloud companies can hedge their future revenue streams and

49
00:02:21,580 --> 00:02:25,139
Speaker 4: the cost of capital can come down. In a week

50
00:02:25,160 --> 00:02:27,380
Speaker 4: like this, when you see the yield on a CoreWeave

51
00:02:27,440 --> 00:02:31,630
Speaker 4: bond at 15%, what does that tell you? Is the

52
00:02:31,650 --> 00:02:34,710
Speaker 4: suggestion that, well, if compute futures were up and running,

53
00:02:34,730 --> 00:02:37,370
Speaker 4: all of CoreWeave's problems would be solved?

54
00:02:37,510 --> 00:02:41,190
Speaker 2: I do think that it's highlighting the massive need for

55
00:02:41,250 --> 00:02:47,690
Speaker 2: risk management products. Think about if you're a neocloud and

56
00:02:47,710 --> 00:02:54,130
Speaker 2: you sign contracts with the most creditworthy people companies in

57
00:02:54,169 --> 00:02:57,470
Speaker 2: the world, then you're fine, right? But there are a

58
00:02:57,490 --> 00:03:02,060
Speaker 2: lot of other consumers of GPUs that are not the

59
00:03:02,100 --> 00:03:04,720
Speaker 2: most credit-worthy companies in the world. And if you're a neocloud,

60
00:03:04,740 --> 00:03:06,180
Speaker 2: you want to be able to sell to them. And

61
00:03:06,280 --> 00:03:10,100
Speaker 2: so I think what the market's saying is, if you're

62
00:03:10,560 --> 00:03:15,870
Speaker 2: selling on-demand compute or you're selling to less credit-worthy entities,

63
00:03:16,430 --> 00:03:20,150
Speaker 2: then your cost of capital is just going to be higher. Now,

64
00:03:20,550 --> 00:03:25,989
Speaker 2: imagine once GPU futures exist, you can, as a neocloud,

65
00:03:26,410 --> 00:03:29,290
Speaker 2: buy the GPUs, deploy them in your data center, sell

66
00:03:29,310 --> 00:03:34,269
Speaker 2: a strip of GPU futures, and then just sell on-demand

67
00:03:34,310 --> 00:03:38,710
Speaker 2: compute if you want. Or sell six-month contracts to less

68
00:03:38,750 --> 00:03:42,290
Speaker 2: creditworthy companies that can post a big percentage of the

69
00:03:42,770 --> 00:03:44,090
Speaker 2: contract value up front.

70
00:03:44,610 --> 00:03:47,010
Speaker 3: And the market is going to.

71
00:03:49,130 --> 00:03:52,380
Speaker 2: Reward companies that do that that basically just have on

72
00:03:52,420 --> 00:03:57,700
Speaker 2: a giant basis trade heavily, their cost of capital is

73
00:03:57,720 --> 00:03:58,540
Speaker 2: going to be much lower.

74
00:03:59,640 --> 00:04:04,300
Speaker 1: Terry mentioned, okay, the GPU futures don't exist yet. And

75
00:04:04,340 --> 00:04:07,580
Speaker 1: there's some message from the White House, like national security,

76
00:04:07,620 --> 00:04:09,700
Speaker 1: et cetera. And he's like, I don't get it. Do

77
00:04:09,740 --> 00:04:13,140
Speaker 1: you get it? Can you tell us something about why

78
00:04:13,240 --> 00:04:16,990
Speaker 1: these products don't exist yet? Maybe what the regulatory hesitancy

79
00:04:17,120 --> 00:04:18,380
Speaker 1: is about the listing?

80
00:04:19,150 --> 00:04:21,290
Speaker 4: Yeah, and just remind us what the state is of

81
00:04:21,390 --> 00:04:22,710
Speaker 4: regulatory approval at the moment.

82
00:04:22,830 --> 00:04:26,360
Speaker 2: So right now, there's a comment period that's open. The

83
00:04:26,760 --> 00:04:33,640
Speaker 2: CFTC extended it. If they don't do anything, then the

84
00:04:34,779 --> 00:04:38,669
Speaker 2: futures can get listed November 9th. Now, conveniently, November 9th

85
00:04:38,750 --> 00:04:39,890
Speaker 2: is after November 3rd.

86
00:04:39,910 --> 00:04:40,800
Speaker 3: And so maybe....

87
00:04:45,470 --> 00:04:47,830
Speaker 2: Some people in Washington will be more comfortable with the

88
00:04:47,890 --> 00:04:52,450
Speaker 2: Compute Futures launching. I don't know. I can't explain what

89
00:04:52,490 --> 00:04:55,000
Speaker 2: the concern is. I've heard all sorts of different theories.

90
00:04:57,500 --> 00:05:01,339
Speaker 2: One theory is that some of the indices are not

91
00:05:01,400 --> 00:05:02,180
Speaker 2: very robust.

92
00:05:02,580 --> 00:05:06,320
Speaker 1: Yeah, okay, so I'm actually very curious about this, the

93
00:05:06,720 --> 00:05:11,590
Speaker 1: robustness of the indices, because a lot of GPUs are

94
00:05:11,650 --> 00:05:13,429
Speaker 1: sold via bilateral offtake.

95
00:05:13,830 --> 00:05:14,289
Speaker 4: Agreements.

96
00:05:14,350 --> 00:05:14,570
Speaker 3: Right.

97
00:05:14,610 --> 00:05:17,630
Speaker 1: So it's like Anthropic has some deal with SpaceX.

98
00:05:17,710 --> 00:05:17,950
Speaker 3: Right.

99
00:05:17,990 --> 00:05:20,670
Speaker 1: To use their compute. We don't really know the price

100
00:05:20,750 --> 00:05:24,890
Speaker 1: of that. Why should anyone feel confident then when they

101
00:05:24,920 --> 00:05:29,520
Speaker 1: look at an index of GPU prices that like it

102
00:05:29,600 --> 00:05:32,820
Speaker 1: actually represents something resembling the save the market here?

103
00:05:32,920 --> 00:05:36,450
Speaker 2: Listen, I think it's buyer beware. learn about the index

104
00:05:36,490 --> 00:05:41,529
Speaker 2: that you're trading futures on. Silicon Data, that team has

105
00:05:41,550 --> 00:05:45,599
Speaker 2: done an amazing job of collecting massive amounts of data,

106
00:05:46,200 --> 00:05:51,960
Speaker 2: and it's focused on on-demand compute from neoclouds. There are

107
00:05:52,000 --> 00:05:54,960
Speaker 2: a lot of neoclouds. There are very few hyperscalers. So

108
00:05:55,240 --> 00:05:58,080
Speaker 2: their index is focused on neoclouds, of which there are many.

109
00:05:59,980 --> 00:06:06,880
Speaker 2: And there is a lot of transparency about on-demand compute. And,

110
00:06:07,460 --> 00:06:11,339
Speaker 2: you know, they've done a very good job to create

111
00:06:11,380 --> 00:06:17,210
Speaker 2: an index that, you know, will be IOSCO compliant.

112
00:06:17,650 --> 00:06:20,950
Speaker 3: That's the benchmark. Some of the other.

113
00:06:20,810 --> 00:06:25,250
Speaker 2: Providers I don't think have ever heard the phrase IOSCO compliant.

114
00:06:25,270 --> 00:06:25,930
Speaker 1: So IOSCO is the.

115
00:06:30,680 --> 00:06:34,290
Speaker 2: I don't know, it's the kind of group of international

116
00:06:34,330 --> 00:06:37,190
Speaker 2: regulators that I think in 2013 came up.

117
00:06:37,210 --> 00:06:41,150
Speaker 3: With a best practices approach for indices.

118
00:06:41,610 --> 00:06:43,650
Speaker 1: Oh, we should do an IOSCO episode one day.

119
00:06:45,589 --> 00:06:47,560
Speaker 4: Now that you've been working on the futures contract for

120
00:06:47,600 --> 00:06:50,159
Speaker 4: a while, can you maybe tell us what you've learned about,

121
00:06:50,180 --> 00:06:54,340
Speaker 4: I guess, the pricing trajectory for GPUs over time? Because

122
00:06:54,360 --> 00:06:56,260
Speaker 4: this is one of the concerns, right? You can say

123
00:06:56,279 --> 00:07:00,270
Speaker 4: you're bullish on the compute markets, But it doesn't necessarily

124
00:07:00,310 --> 00:07:03,580
Speaker 4: mean that you're bullish on the value of a single GPU.

125
00:07:03,620 --> 00:07:07,339
Speaker 4: It can depreciate over time. Maybe it gets more efficient

126
00:07:07,380 --> 00:07:08,800
Speaker 4: and people move on to new things.

127
00:07:09,800 --> 00:07:15,210
Speaker 2: So that is the trillion dollar question. I'm glad I

128
00:07:15,290 --> 00:07:19,170
Speaker 2: asked it. And that is why these futures need to exist.

129
00:07:19,250 --> 00:07:25,880
Speaker 2: Because nobody knows what the value of a B200 will

130
00:07:25,900 --> 00:07:30,140
Speaker 2: be four years from now, five years from now. And,

131
00:07:30,940 --> 00:07:36,310
Speaker 2: you know, and so by introducing futures contracts that trade,

132
00:07:36,650 --> 00:07:40,130
Speaker 2: you know, way out, I think initially the CMEs futures

133
00:07:40,170 --> 00:07:42,650
Speaker 2: contracts are going to go out three years and, you know,

134
00:07:43,230 --> 00:07:44,990
Speaker 2: as they become more liquid, they'll.

135
00:07:44,750 --> 00:07:47,690
Speaker 3: Go out more.

136
00:07:48,870 --> 00:07:53,220
Speaker 2: But, you know, that enables people to hedge that. Back

137
00:07:53,520 --> 00:08:00,060
Speaker 2: in November of, you know, almost a year ago, the

138
00:08:00,100 --> 00:08:05,140
Speaker 2: price of H100s, the hourly on-demand price, had just been declining, declining, declining,

139
00:08:05,200 --> 00:08:08,600
Speaker 2: and everybody was saying, why do you need GPU futures?

140
00:08:08,680 --> 00:08:09,740
Speaker 3: These things just go down.

141
00:08:10,680 --> 00:08:17,660
Speaker 2: And then, you know, right around Thanksgiving, they turned around

142
00:08:18,120 --> 00:08:21,500
Speaker 2: and kind of went up and to the right. And

143
00:08:21,640 --> 00:08:26,270
Speaker 2: what happened, you know, it was Anthropix's newest model came out,

144
00:08:26,430 --> 00:08:31,330
Speaker 2: which was a massive step up in usefulness. And all

145
00:08:31,350 --> 00:08:34,770
Speaker 2: of a sudden, there was a ton of demand for inference. And,

146
00:08:35,250 --> 00:08:42,770
Speaker 2: you know, more recently, we've seen, you know, looking at B200, B300,

147
00:08:42,929 --> 00:08:49,780
Speaker 2: and H100 prices, another surge upward in price. And, by

148
00:08:49,820 --> 00:08:52,420
Speaker 2: the way, at the same time that the Silicon Data

149
00:08:52,800 --> 00:08:54,740
Speaker 2: token index price has plummeted.

150
00:08:54,800 --> 00:08:55,699
Speaker 1: Yeah, it's interesting.

151
00:08:56,320 --> 00:08:56,740
Speaker 3: And so.

152
00:08:58,040 --> 00:09:00,460
Speaker 2: Yeah, I mean, these things are volatile. They can go up,

153
00:09:00,480 --> 00:09:05,000
Speaker 2: they can go down, and a hedging instrument is badly needed.

154
00:09:21,250 --> 00:09:25,920
Speaker 4: Just as a hypothetical, let's say GPUs depreciate over time

155
00:09:26,179 --> 00:09:29,500
Speaker 4: and there's more demand but in general the value goes

156
00:09:29,540 --> 00:09:30,980
Speaker 4: down which is something we've seen in a lot of

157
00:09:31,020 --> 00:09:34,980
Speaker 4: commodity markets right in that case does having a market

158
00:09:35,020 --> 00:09:39,360
Speaker 4: where you're able to hedge your future gpu revenue streams

159
00:09:39,760 --> 00:09:43,720
Speaker 4: does that just enable people to i guess borrow more

160
00:09:43,920 --> 00:09:49,199
Speaker 4: and overbuild versus genuinely reduce risk you're just allowing people

161
00:09:49,260 --> 00:09:52,689
Speaker 4: to borrow against the same risk, ostensibly with a hedge,

162
00:09:52,690 --> 00:09:56,569
Speaker 4: but the underlying isn't actually moving in your favor.

163
00:09:56,570 --> 00:09:58,030
Speaker 3: Does that make sense?

164
00:09:58,030 --> 00:10:00,079
Speaker 2: No. I mean, here, the reason it doesn't make sense

165
00:10:00,320 --> 00:10:03,280
Speaker 2: is that for a futures contract to exist, there has

166
00:10:03,320 --> 00:10:05,980
Speaker 2: to be a willing buyer and a willing seller, right?

167
00:10:06,420 --> 00:10:09,870
Speaker 2: So your hypothesis would work if they could just go

168
00:10:09,929 --> 00:10:12,890
Speaker 2: and sell futures contracts at some price that was made

169
00:10:12,990 --> 00:10:17,420
Speaker 2: up maybe too high, and there was kind of infinite

170
00:10:17,460 --> 00:10:21,059
Speaker 2: demand for that. Okay, well, sure. But there has to

171
00:10:21,100 --> 00:10:23,120
Speaker 2: be a rational buyer on the other side. And so

172
00:10:23,179 --> 00:10:25,570
Speaker 2: I think what we'll see is that those forward contracts

173
00:10:25,610 --> 00:10:27,370
Speaker 2: are going to be, are in fact going to be

174
00:10:27,410 --> 00:10:31,450
Speaker 2: at lower prices just because there's so much hedging demand

175
00:10:31,890 --> 00:10:33,689
Speaker 2: to sell in the forward space.

176
00:10:34,350 --> 00:10:36,550
Speaker 3: And there will need to be, you know.

177
00:10:36,750 --> 00:10:41,380
Speaker 2: Investors who want to generate a return by buying those

178
00:10:41,400 --> 00:10:42,180
Speaker 2: futures contracts.

179
00:10:43,700 --> 00:10:46,360
Speaker 1: What is the– if we cast out a few years,

180
00:10:46,780 --> 00:10:53,940
Speaker 1: what is the sort of ecosystem of clouds and inference

181
00:10:54,620 --> 00:10:56,939
Speaker 1: providers that needs to exist for this to be a

182
00:10:56,980 --> 00:11:01,929
Speaker 1: rich market? Because if the future of AI is just–

183
00:11:02,190 --> 00:11:04,510
Speaker 1: let's say the future of AI were just anthropic and

184
00:11:04,550 --> 00:11:09,860
Speaker 1: open AI. And the future of cloud were just meta, Alphabet,

185
00:11:10,370 --> 00:11:13,130
Speaker 1: and Amazon. There wouldn't need to be a futures market.

186
00:11:13,150 --> 00:11:16,630
Speaker 1: It would just be all bilateral agreements, offtake agreements, et cetera.

187
00:11:16,860 --> 00:11:17,030
Speaker 2: Yep.

188
00:11:17,300 --> 00:11:21,950
Speaker 1: So how much like. How fat of a tale of

189
00:11:22,050 --> 00:11:26,550
Speaker 1: non-open AI and anthropic inference do we need such that

190
00:11:26,610 --> 00:11:29,950
Speaker 1: this is a useful vault market that cannot just be

191
00:11:29,990 --> 00:11:31,410
Speaker 1: solved by bilateral agreements?

192
00:11:31,510 --> 00:11:33,790
Speaker 2: Well, I mean, I think we're already there. I mean,

193
00:11:33,830 --> 00:11:37,130
Speaker 2: look at the growing use of open source models. I mean,

194
00:11:37,170 --> 00:11:44,380
Speaker 2: it's exploded. And you're seeing companies with many billion dollar

195
00:11:44,580 --> 00:11:48,920
Speaker 2: valuations that have just been created to put open source

196
00:11:48,970 --> 00:11:51,170
Speaker 2: models on GPUs and sell the tokens.

197
00:11:51,190 --> 00:11:52,290
Speaker 3: Right.

198
00:11:52,850 --> 00:11:58,470
Speaker 2: You know, like fireworks and together. And so we already

199
00:11:58,530 --> 00:12:00,550
Speaker 2: have a really rich ecosystem.

200
00:12:01,030 --> 00:12:01,929
Speaker 1: Is it rich enough yet?

201
00:12:02,530 --> 00:12:03,930
Speaker 3: For sure. Okay, interesting.

202
00:12:04,470 --> 00:12:04,630
Speaker 2: Yeah.

203
00:12:05,080 --> 00:12:06,240
Speaker 4: Should we talk about perpetuals?

204
00:12:07,420 --> 00:12:08,079
Speaker 3: Yeah.

205
00:12:08,559 --> 00:12:09,540
Speaker 2: Can I ask another?

206
00:12:09,860 --> 00:12:10,660
Speaker 3: Yeah, go for it.

207
00:12:10,679 --> 00:12:14,120
Speaker 1: Can I ask, on GPU futures, So one of the

208
00:12:14,280 --> 00:12:18,150
Speaker 1: other things that people say, the skeptics, is they say

209
00:12:18,730 --> 00:12:25,229
Speaker 1: GPU or compute is not homogenous enough for a futures market.

210
00:12:25,470 --> 00:12:28,550
Speaker 1: And there's a few different things. They say there's different

211
00:12:28,610 --> 00:12:32,820
Speaker 1: software arrays, and that could significantly change the value. There's

212
00:12:32,900 --> 00:12:35,840
Speaker 1: also the fact that there's, at least in the inference

213
00:12:35,880 --> 00:12:39,300
Speaker 1: part of AI, there's a lot of other players who

214
00:12:39,340 --> 00:12:45,120
Speaker 1: want AMD and and etched and all these other tranium

215
00:12:45,540 --> 00:12:47,820
Speaker 1: and whatever the Microsoft one was and.

216
00:12:47,740 --> 00:12:48,840
Speaker 3: TPUs, et cetera.

217
00:12:49,250 --> 00:12:52,180
Speaker 1: And this is just way too diverse of a market

218
00:12:52,640 --> 00:12:57,090
Speaker 1: to really be captured in an index. How confident are

219
00:12:57,130 --> 00:12:59,630
Speaker 1: you that either the market is going to be homogenous

220
00:12:59,690 --> 00:13:03,670
Speaker 1: enough for futures or that there will be enough sophistication

221
00:13:03,730 --> 00:13:07,570
Speaker 1: about the sort of basis between the H100 index versus

222
00:13:07,590 --> 00:13:11,670
Speaker 1: something similar such that this market can serve a diverse

223
00:13:12,429 --> 00:13:13,130
Speaker 1: provider base?

224
00:13:13,370 --> 00:13:13,530
Speaker 3: Yeah.

225
00:13:13,590 --> 00:13:16,190
Speaker 2: I mean, first of all, the amount of money going into,

226
00:13:16,570 --> 00:13:20,729
Speaker 2: you know, just H-100s or just B-200s is large enough

227
00:13:21,030 --> 00:13:25,200
Speaker 2: to support a single futures market. Now, in addition to that,

228
00:13:25,429 --> 00:13:29,580
Speaker 2: these other things are very correlated. And so, you know,

229
00:13:29,620 --> 00:13:32,720
Speaker 2: people will be able to trade a basis on these things.

230
00:13:33,260 --> 00:13:39,480
Speaker 3: Versus whatever it is. And, you know, maybe... obviously, you know,

231
00:13:39,520 --> 00:13:40,080
Speaker 3: these are going.

232
00:13:39,980 --> 00:13:46,339
Speaker 2: To be off of kind of the, the baseline standard configuration, whatever. But,

233
00:13:46,640 --> 00:13:50,510
Speaker 2: and so maybe, uh, somebody's planning on doing a training

234
00:13:50,550 --> 00:13:53,110
Speaker 2: run and they need a giant cluster. Well, that's going

235
00:13:53,130 --> 00:13:56,270
Speaker 2: to bear a premium, of course. But, but, you know,

236
00:13:56,929 --> 00:14:00,550
Speaker 2: if the demand for inference goes up massively, then guess what?

237
00:14:00,570 --> 00:14:02,699
Speaker 2: That training run is also going to be more expensive.

238
00:14:02,840 --> 00:14:05,060
Speaker 2: So it's just a basis between the things.

239
00:14:05,080 --> 00:14:05,640
Speaker 3: That's one thing.

240
00:14:05,679 --> 00:14:10,640
Speaker 2: The other thing is, um, that eventually, and, you know,

241
00:14:10,679 --> 00:14:13,420
Speaker 2: we may be getting close, you'll be able to trade

242
00:14:13,740 --> 00:14:18,360
Speaker 2: futures contracts on the output, on the tokens themselves.

243
00:14:18,679 --> 00:14:20,570
Speaker 3: Oh, token futures. Token futures.

244
00:14:21,110 --> 00:14:24,710
Speaker 2: And so, you know, again, there's, of course, a diverse

245
00:14:25,130 --> 00:14:28,030
Speaker 2: number of tokens and people will put up the same objections.

246
00:14:28,530 --> 00:14:34,410
Speaker 2: But I think that over time, that's going to be possible.

247
00:14:34,470 --> 00:14:38,190
Speaker 2: And By the way, that will introduce a really interesting possibility,

248
00:14:38,690 --> 00:14:42,050
Speaker 2: which is, now think about if you go.

249
00:14:42,150 --> 00:14:45,620
Speaker 3: And rent GPUs, you pay the on-demand price.

250
00:14:46,100 --> 00:14:48,500
Speaker 2: And then you load up an open source model, and

251
00:14:48,520 --> 00:14:53,360
Speaker 2: you start selling tokens through OpenRouter or something. Well, can

252
00:14:53,400 --> 00:14:59,280
Speaker 2: you do that profitably or not? Well, that profit margin

253
00:14:59,320 --> 00:15:00,740
Speaker 2: that you can generate.

254
00:15:00,260 --> 00:15:01,410
Speaker 3: Is akin to.

255
00:15:04,330 --> 00:15:09,250
Speaker 2: What an oil refinery does or what a soybean crush

256
00:15:09,310 --> 00:15:09,970
Speaker 2: plant does.

257
00:15:10,730 --> 00:15:12,750
Speaker 3: So we're going to have a crack spread.

258
00:15:13,010 --> 00:15:15,800
Speaker 2: So in oil, we call it a crack spread. In soybeans,

259
00:15:15,840 --> 00:15:19,140
Speaker 2: we call it a crush spread. So you heard it

260
00:15:19,180 --> 00:15:24,340
Speaker 2: here first. At DRW, we call this the crunch spread.

261
00:15:24,660 --> 00:15:25,280
Speaker 1: That's great.

262
00:15:25,780 --> 00:15:28,670
Speaker 4: Wait, is this token futures or something you're actively working on,

263
00:15:28,730 --> 00:15:29,290
Speaker 4: it sounds like?

264
00:15:29,470 --> 00:15:30,250
Speaker 2: Yes. Okay.

265
00:15:30,270 --> 00:15:30,390
Speaker 3: Okay.

266
00:15:31,090 --> 00:15:33,730
Speaker 4: What sort of stage would you say you're at?

267
00:15:33,750 --> 00:15:37,670
Speaker 2: I mean, it's the thought stage, but I mean, here,

268
00:15:38,070 --> 00:15:45,290
Speaker 2: to be fair, the silicon data token index is providing

269
00:15:45,430 --> 00:15:49,720
Speaker 2: some pretty interesting data and is widely discussed.

270
00:15:49,840 --> 00:15:53,160
Speaker 3: So that's a start.

271
00:15:53,880 --> 00:15:56,840
Speaker 4: And so just going back to GPU futures for a second,

272
00:15:56,860 --> 00:16:00,130
Speaker 4: so every time a new chip comes out, that needs

273
00:16:00,150 --> 00:16:03,340
Speaker 4: to be a separate contract and a new pool of liquidity.

274
00:16:03,360 --> 00:16:05,010
Speaker 4: Is that right? Is that how you're thinking of it?

275
00:16:05,080 --> 00:16:05,400
Speaker 2: Correct.

276
00:16:05,680 --> 00:16:05,900
Speaker 4: Okay.

277
00:16:05,980 --> 00:16:08,460
Speaker 2: And how do you, I mean, it seems kind of challenging.

278
00:16:08,500 --> 00:16:09,820
Speaker 3: You have to like keep.

279
00:16:09,700 --> 00:16:13,950
Speaker 4: Up with the technological cycle, which is very rapid.

280
00:16:14,310 --> 00:16:17,330
Speaker 3: It's rapid, but not that rapid. Right.

281
00:16:17,650 --> 00:16:20,490
Speaker 2: And, and you're probably not going to launch a futures

282
00:16:20,530 --> 00:16:23,630
Speaker 2: contract for every single one. You're going to launch futures

283
00:16:23,670 --> 00:16:26,980
Speaker 2: contracts for the ones that are in the major ones. And, uh,

284
00:16:27,570 --> 00:16:29,610
Speaker 2: You know, but that's okay. I mean, there are different,

285
00:16:29,850 --> 00:16:33,030
Speaker 2: you know, think about how many electricity contracts.

286
00:16:32,620 --> 00:16:36,740
Speaker 3: There are for every little node. And I mean, you know,

287
00:16:37,080 --> 00:16:37,660
Speaker 3: thousands of them.

288
00:16:37,800 --> 00:16:40,040
Speaker 1: There's a lot of oil contracts too. We just usually

289
00:16:40,060 --> 00:16:43,640
Speaker 1: talk about two of them typically. Just out of curiosity,

290
00:16:43,680 --> 00:16:46,340
Speaker 1: and this is actually sort of separate from the futures conversation,

291
00:16:46,360 --> 00:16:49,790
Speaker 1: but I guess like as a technologist with multiple businesses,

292
00:16:50,010 --> 00:16:56,570
Speaker 1: et cetera, are you, how assailable is NVIDIA's dominance today? Like,

293
00:16:56,590 --> 00:16:59,470
Speaker 1: what's your read on that? Because this is a constant question,

294
00:16:59,830 --> 00:17:02,650
Speaker 1: whether any of these other chip makers can actually eat

295
00:17:02,690 --> 00:17:05,490
Speaker 1: into NVIDIA's position. What's your take there?

296
00:17:06,210 --> 00:17:08,920
Speaker 2: Sure. I think that the space is going to continue

297
00:17:08,940 --> 00:17:14,060
Speaker 2: to grow. And I can't, it's unlikely that NVIDIA will

298
00:17:14,100 --> 00:17:16,760
Speaker 2: be able to maintain the market share that they have maintained.

299
00:17:17,000 --> 00:17:25,070
Speaker 2: But I'm pretty confident that NVIDIA will continue.

300
00:17:24,690 --> 00:17:25,410
Speaker 3: To grow as a company.

301
00:17:41,609 --> 00:17:43,790
Speaker 4: You know, last year when we spoke, we were talking

302
00:17:43,810 --> 00:17:47,100
Speaker 4: about AI use in general at DRW. And I think

303
00:17:47,140 --> 00:17:51,060
Speaker 4: you said people were using agents to sort of debate

304
00:17:51,260 --> 00:17:54,980
Speaker 4: investment theories or risk allocation and things like that. But

305
00:17:55,030 --> 00:17:58,570
Speaker 4: you weren't using AI to actually deploy risk at that time.

306
00:17:59,430 --> 00:18:01,840
Speaker 4: Joe mentioned that things change very, very fast in the

307
00:18:01,859 --> 00:18:04,300
Speaker 4: world of AI. 12 months later, has that changed? What's

308
00:18:04,320 --> 00:18:05,120
Speaker 4: AI doing for you now?

309
00:18:05,340 --> 00:18:09,860
Speaker 2: Yeah, no, I mean, there's definitely more adoption. There's more frontline.

310
00:18:09,920 --> 00:18:10,820
Speaker 3: Adoption, for sure.

311
00:18:11,680 --> 00:18:13,040
Speaker 4: Has your token budget gone up?

312
00:18:13,960 --> 00:18:17,320
Speaker 3: The token budget goes up. Yes, it goes up.

313
00:18:17,380 --> 00:18:22,340
Speaker 2: The token use is like this exponential chart. It's pretty interesting.

314
00:18:22,640 --> 00:18:25,850
Speaker 1: Do you have, there's another question, like across your businesses,

315
00:18:26,310 --> 00:18:30,510
Speaker 1: have you figured out optimal internal token allocation. We've asked

316
00:18:30,530 --> 00:18:33,590
Speaker 1: this to a lot of executives and at least up

317
00:18:33,630 --> 00:18:35,490
Speaker 1: until recent, they're like, we're just still trying to figure

318
00:18:35,530 --> 00:18:38,050
Speaker 1: this out, but you must have everyone, of course, everyone

319
00:18:38,090 --> 00:18:41,680
Speaker 1: wants more. I assume everyone always wants the frontier token cost.

320
00:18:42,040 --> 00:18:46,140
Speaker 1: Have you solved the question of like internal token allocation yet?

321
00:18:46,160 --> 00:18:48,340
Speaker 2: Well, here, I mean, I think the most important thing

322
00:18:48,560 --> 00:18:52,620
Speaker 2: is to be thoughtful about how you route the, the

323
00:18:53,000 --> 00:18:57,010
Speaker 2: inference requests and, and, And not every inference request needs

324
00:18:57,050 --> 00:19:02,959
Speaker 2: to go to a frontier model. And so that's the

325
00:19:02,980 --> 00:19:05,419
Speaker 2: thing that we're trying to optimize for, is minimize token

326
00:19:05,440 --> 00:19:09,720
Speaker 2: spend by optimally routing inference requests.

327
00:19:10,760 --> 00:19:14,790
Speaker 4: Now shall we ask about perpetuals? OK, perps, go.

328
00:19:15,609 --> 00:19:17,949
Speaker 2: Why is Terry wrong about perps? I mean, listen, here.

329
00:19:24,940 --> 00:19:27,800
Speaker 3: In my view, are... Terry's in the audience, by the way.

330
00:19:27,960 --> 00:19:28,280
Speaker 2: I know.

331
00:19:28,320 --> 00:19:30,159
Speaker 4: But you're also his client, so.

332
00:19:30,380 --> 00:19:30,680
Speaker 3: I know.

333
00:19:30,859 --> 00:19:37,810
Speaker 2: And Terry and I enjoy talking about this. My view

334
00:19:37,990 --> 00:19:41,910
Speaker 2: is that perps serve a useful economic purpose.

335
00:19:42,970 --> 00:19:48,370
Speaker 3: They are futures contracts. And I mean, here.

336
00:19:48,910 --> 00:19:52,840
Speaker 2: I'll give you a couple examples. The, you know, a

337
00:19:52,900 --> 00:19:57,210
Speaker 2: lot of people like to buy the crude oil ETF. Yeah,

338
00:19:57,230 --> 00:20:00,190
Speaker 2: the crude oil ETF generally, you know, at one point,

339
00:20:00,230 --> 00:20:01,889
Speaker 2: it was always in the front month future, and it

340
00:20:01,930 --> 00:20:04,390
Speaker 2: broadcast to the world exactly what day it would roll

341
00:20:04,430 --> 00:20:07,090
Speaker 2: from that futures to the next futures contract. And I

342
00:20:07,109 --> 00:20:08,730
Speaker 2: don't know, there was probably a little bit of slippage

343
00:20:08,780 --> 00:20:11,580
Speaker 2: in that, a little bit of inefficiency. Now they've kind

344
00:20:11,619 --> 00:20:14,040
Speaker 2: of like, now they do it over a window and

345
00:20:14,140 --> 00:20:17,080
Speaker 2: they hold the first three contracts and whatever.

346
00:20:18,060 --> 00:20:19,260
Speaker 3: But there's still a lot of slippage.

347
00:20:19,320 --> 00:20:23,130
Speaker 2: Now imagine a PERP that is tied to, let's say,

348
00:20:23,150 --> 00:20:25,090
Speaker 2: the Brent Index.

349
00:20:25,910 --> 00:20:30,369
Speaker 3: And an ETF that just buys that.

350
00:20:31,690 --> 00:20:36,390
Speaker 2: PERP. I mean, I think that that would be much

351
00:20:36,609 --> 00:20:42,210
Speaker 2: more efficient experience for the ETF. And of course, that

352
00:20:42,250 --> 00:20:45,070
Speaker 2: doesn't mean that that's a useful hedging instrument for an

353
00:20:45,210 --> 00:20:45,930
Speaker 2: oil producer.

354
00:20:46,070 --> 00:20:46,730
Speaker 3: It's not.

355
00:20:47,230 --> 00:20:50,950
Speaker 2: An oil producer needs to hedge way out the curve. Great.

356
00:20:51,400 --> 00:20:55,359
Speaker 2: You can have monthly and quarterly futures and PERPs side

357
00:20:55,400 --> 00:20:55,840
Speaker 2: by side.

358
00:20:55,880 --> 00:21:00,300
Speaker 3: They actually augment.

359
00:20:58,580 --> 00:20:59,179
Speaker 2: One another and.

360
00:21:00,710 --> 00:21:03,090
Speaker 3: And that's a good thing. Here's another example.

361
00:21:03,570 --> 00:21:08,250
Speaker 2: Suppose you are a retail investor and you own a

362
00:21:08,890 --> 00:21:14,810
Speaker 2: hard-to-borrow stock. And all the hedge funds want to short.

363
00:21:14,890 --> 00:21:19,450
Speaker 2: They're paying 100% annualized to short this stock. If you

364
00:21:19,490 --> 00:21:24,129
Speaker 2: have that stock in your retail account, your broker is

365
00:21:24,190 --> 00:21:26,950
Speaker 2: probably not saying to you, hey, by the way, we're

366
00:21:26,970 --> 00:21:29,109
Speaker 2: going to give you 300 basis points a day because

367
00:21:29,130 --> 00:21:31,990
Speaker 2: you're holding this Hard to borrow stock.

368
00:21:32,570 --> 00:21:34,670
Speaker 3: Okay, that would be 1,000% annualized, but.

369
00:21:34,550 --> 00:21:41,699
Speaker 2: 30 basis points a day. But whatever. So you're not

370
00:21:41,740 --> 00:21:44,240
Speaker 2: getting the benefit of owning that hard-to-borrow stock.

371
00:21:44,280 --> 00:21:46,100
Speaker 3: Now, if instead you own the.

372
00:21:46,080 --> 00:21:51,820
Speaker 2: PER, then mechanically it's going to pass through. And so

373
00:21:52,080 --> 00:21:58,650
Speaker 2: it's a real opportunity to democratize the financial markets.

374
00:21:59,490 --> 00:22:03,240
Speaker 1: Right, so democratize. This word democratize, I've never really liked

375
00:22:03,320 --> 00:22:05,280
Speaker 1: it because it just sounds good, but I don't know

376
00:22:05,300 --> 00:22:06,980
Speaker 1: what it means. But in this case, it makes sense

377
00:22:07,080 --> 00:22:10,830
Speaker 1: because in this case, what you're saying is there is

378
00:22:10,850 --> 00:22:14,350
Speaker 1: an economic value that is not being captured currently by

379
00:22:14,430 --> 00:22:17,910
Speaker 1: retail holders of certain stocks because no one's making the offer.

380
00:22:18,210 --> 00:22:20,369
Speaker 1: But the demand would be there if they held that

381
00:22:20,430 --> 00:22:21,210
Speaker 1: same exposure in.

382
00:22:21,210 --> 00:22:22,050
Speaker 3: A different way. Correct.

383
00:22:22,130 --> 00:22:24,710
Speaker 2: And the same thing for the ETF, the oil ETF.

384
00:22:24,750 --> 00:22:27,230
Speaker 2: I mean, that's probably a retail investor that wants to

385
00:22:27,270 --> 00:22:28,390
Speaker 2: have exposure to oil.

386
00:22:28,750 --> 00:22:30,250
Speaker 3: Here's a more efficient way to do it.

387
00:22:30,650 --> 00:22:33,970
Speaker 4: I get that it might be more useful and economically

388
00:22:34,030 --> 00:22:36,750
Speaker 4: beneficial to some, but if the regulation says that a

389
00:22:36,790 --> 00:22:40,640
Speaker 4: futures contract has to have an expiration date and these don't,

390
00:22:41,100 --> 00:22:41,820
Speaker 4: why should it not?

391
00:22:41,840 --> 00:22:42,680
Speaker 3: Where does it say that?

392
00:22:43,380 --> 00:22:44,580
Speaker 2: I haven't read that. Does it not?

393
00:22:45,940 --> 00:22:46,600
Speaker 3: Not that I've seen.

394
00:22:47,220 --> 00:22:49,590
Speaker 1: I don't know. So Terry said they're swaps.

395
00:22:49,869 --> 00:22:50,850
Speaker 2: And you said they're futures.

396
00:22:50,890 --> 00:22:51,650
Speaker 1: Why are they not swaps?

397
00:22:51,670 --> 00:22:54,990
Speaker 4: I feel like this should be an easily settled debate.

398
00:22:55,090 --> 00:22:58,130
Speaker 2: OK. So what's the difference? And I asked this question

399
00:22:58,170 --> 00:23:02,710
Speaker 2: when Dodd-Frank was being drafted. And what's the difference.

400
00:23:02,410 --> 00:23:05,870
Speaker 3: Between a future and a swap? I have no idea. OK.

401
00:23:05,890 --> 00:23:09,700
Speaker 2: Well, the answer that I got from the person drafting

402
00:23:09,760 --> 00:23:13,880
Speaker 2: the legislation for Barney Frank was, everybody.

403
00:23:13,240 --> 00:23:20,409
Speaker 3: Knows that swaps are risky. What? I'm sorry.

404
00:23:20,430 --> 00:23:27,060
Speaker 2: A swap that is cleared has the same economic risk

405
00:23:27,600 --> 00:23:33,650
Speaker 2: to society, to the holder, as a futures contract does.

406
00:23:34,450 --> 00:23:36,730
Speaker 3: I mean, we've gotten like lost.

407
00:23:36,970 --> 00:23:39,180
Speaker 2: We've like lost our minds with nomenclature.

408
00:23:40,230 --> 00:23:43,970
Speaker 4: But can you address directly Terry's point about leverage in

409
00:23:44,010 --> 00:23:45,510
Speaker 4: that case? Like how do you manage the.

410
00:23:45,430 --> 00:23:46,190
Speaker 2: Leverage from these?

411
00:23:46,260 --> 00:23:47,020
Speaker 3: No, I'm sorry.

412
00:23:47,480 --> 00:23:51,640
Speaker 2: Again, like there's this confusion. Well, if it's a perp,

413
00:23:52,140 --> 00:23:55,380
Speaker 2: it must have massive leverage and auto deleveraging and all

414
00:23:55,420 --> 00:23:56,260
Speaker 2: these horrible things.

415
00:23:57,220 --> 00:23:57,600
Speaker 4: Wrong.

416
00:23:58,400 --> 00:24:01,150
Speaker 2: You can, by the way, have a quarterly futures contract

417
00:24:01,190 --> 00:24:04,330
Speaker 2: with massive leverage and auto deleveraging and all these horrible things.

418
00:24:05,090 --> 00:24:09,750
Speaker 2: You don't. you could list a perp with reasonable leverage

419
00:24:09,930 --> 00:24:13,060
Speaker 2: and no auto deleveraging, and it would be a very

420
00:24:13,160 --> 00:24:14,610
Speaker 2: useful economic instrument.

421
00:24:15,290 --> 00:24:17,550
Speaker 4: Joe, next year, we're going to cut ourselves out of

422
00:24:17,600 --> 00:24:20,320
Speaker 4: this conversation entirely and just have Terry and John to do.

423
00:24:20,200 --> 00:24:21,040
Speaker 3: This on stage.

424
00:24:21,280 --> 00:24:24,179
Speaker 1: It's crazy to think some of our most animated moments

425
00:24:24,240 --> 00:24:26,359
Speaker 1: on the podcast will now be, what is the definition

426
00:24:26,400 --> 00:24:27,000
Speaker 1: of a swap?

427
00:24:27,460 --> 00:24:28,750
Speaker 3: But this is why.

428
00:24:30,330 --> 00:24:33,950
Speaker 1: These are the conversations that we like to have. Okay,

429
00:24:34,090 --> 00:24:36,950
Speaker 1: when 24-7 trading inevitably arrives?

430
00:24:37,650 --> 00:24:38,270
Speaker 2: It's already here.

431
00:24:38,290 --> 00:24:39,109
Speaker 3: What are you talking about?

432
00:24:39,130 --> 00:24:39,590
Speaker 1: That's true.

433
00:24:39,790 --> 00:24:40,230
Speaker 2: It is here.

434
00:24:40,280 --> 00:24:43,080
Speaker 1: You're right. Does that change people's lives? Is that annoying?

435
00:24:43,160 --> 00:24:45,179
Speaker 1: Is it annoying to have to think about risk at

436
00:24:45,180 --> 00:24:46,320
Speaker 1: 1 a.m. on a Sunday?

437
00:24:47,660 --> 00:24:48,900
Speaker 3: Just from a personal life.

438
00:24:48,960 --> 00:24:51,159
Speaker 4: Won't someone think of a financial journalist?

439
00:24:51,200 --> 00:24:53,540
Speaker 2: I mean, first of all, we've been trading crypto since 2014. Yeah,

440
00:24:53,580 --> 00:24:54,149
Speaker 2: so you've been 24-7.

441
00:24:54,140 --> 00:24:56,240
Speaker 3: So I'm used to markets moving 24-7. I will tell

442
00:24:56,290 --> 00:24:58,340
Speaker 3: you that our options market makers....

443
00:25:06,510 --> 00:25:10,710
Speaker 2: Are not happy about the prospect of, you know, making

444
00:25:10,790 --> 00:25:12,630
Speaker 2: SOFR options markets over weekends.

445
00:25:12,730 --> 00:25:16,619
Speaker 3: Yeah. And you know what? I'm very sympathetic to that.

446
00:25:16,640 --> 00:25:17,619
Speaker 1: But they're happening.

447
00:25:19,000 --> 00:25:19,420
Speaker 3: You know what?

448
00:25:19,619 --> 00:25:21,940
Speaker 2: It's just going to happen. We're going to have to

449
00:25:21,980 --> 00:25:24,400
Speaker 2: figure out how to deal with it. All right.

450
00:25:24,600 --> 00:25:27,480
Speaker 1: Don Wilson, thank you so much for coming back to

451
00:25:27,530 --> 00:25:30,010
Speaker 1: Outlaws Live. We're just going to do it again next year,

452
00:25:30,030 --> 00:25:31,510
Speaker 1: the year after that, and the year after that.

453
00:25:31,550 --> 00:25:31,930
Speaker 2: That's right.

454
00:25:31,970 --> 00:25:32,490
Speaker 1: Thank you so much.

455
00:25:45,570 --> 00:25:48,550
Speaker 4: That was another episode of the Odd Thoughts Podcast recorded

456
00:25:48,750 --> 00:25:51,790
Speaker 4: live in Chicago. I'm Tracy Alloway. You can follow me

457
00:25:51,970 --> 00:25:52,889
Speaker 4: at Tracy Alloway.

458
00:25:53,030 --> 00:25:55,750
Speaker 1: And I'm Joe Weisenthal. You can follow me at The Stalwart.

459
00:25:56,020 --> 00:25:59,440
Speaker 1: Follow our producers, Carmen Rodriguez at Carmen Arman, Dashiell Bennett

460
00:25:59,540 --> 00:26:02,680
Speaker 1: at Dashbot, Cale Brooks at Cale Brooks, and Kevin Lozano

461
00:26:02,840 --> 00:26:05,859
Speaker 1: at Kevin Lloyd Lozano. For more OddLots content, go to

462
00:26:05,880 --> 00:26:09,159
Speaker 1: Bloomberg.com slash OddLots for the daily newsletter and all of

463
00:26:09,220 --> 00:26:11,619
Speaker 1: our episodes. And you can chat about all these topics

464
00:26:11,640 --> 00:26:15,620
Speaker 1: 24-7 in our Discord, Discord.gg slash OddLots.

465
00:26:15,960 --> 00:26:18,459
Speaker 4: And if you enjoyed this conversation, then please leave us

466
00:26:18,520 --> 00:26:21,590
Speaker 4: a positive review on your favorite podcast platform. If you're

467
00:26:21,609 --> 00:26:24,330
Speaker 4: watching the video, then please like or leave a comment

468
00:26:24,420 --> 00:26:25,609
Speaker 4: or better yet, subscribe.

469
00:26:26,130 --> 00:26:27,190
Speaker 1: Thanks for watching and listening.

470
00:26:39,200 --> 00:26:39,649
Speaker 3: Thank you.