1 00:00:00,320 --> 00:00:04,040 Welcome back to Data Driven, the podcast where we talk about how data 2 00:00:04,040 --> 00:00:07,680 and AI are changing the world. And sometimes we 3 00:00:07,680 --> 00:00:11,360 even understand it. Today's guest is the brilliant Carmen 4 00:00:11,360 --> 00:00:15,200 Lee, CEO of Silicon Data and former Bloomberg brainiac 5 00:00:15,200 --> 00:00:19,000 who's now on a mission to bring financial grade transparency to the wild west 6 00:00:19,000 --> 00:00:22,840 of GPU compute markets. If you've ever wondered how to hedge 7 00:00:22,840 --> 00:00:26,580 your AI infrastructure costs the way airlines hedge fuel, or what 8 00:00:26,580 --> 00:00:30,180 a futures market for GPUs even looks like, you're in for a 9 00:00:30,180 --> 00:00:33,500 treat. Carmen's turning raw compute into a tradable 10 00:00:33,500 --> 00:00:36,980 commodity, normalizing chaos, and possibly building the 11 00:00:36,980 --> 00:00:40,420 Bloomberg terminal for AI infrastructure. Minus the beige 12 00:00:40,420 --> 00:00:44,220 keyboard, we cover everything from tokenomics and TSMC 13 00:00:44,220 --> 00:00:47,860 to why your AI startup's margins are flatter than the earth in a 14 00:00:47,860 --> 00:00:51,580 conspiracy forum. Oh, and there's a used GPU car 15 00:00:51,580 --> 00:00:55,160 lot somewhere in Virginia. Stick around. This one's a data 16 00:00:55,160 --> 00:00:57,400 geek's fever dream in the best way. 17 00:01:00,680 --> 00:01:04,360 Hello and welcome back to Data Driven, the podcast where we explore the 18 00:01:04,360 --> 00:01:08,120 emerging field of data science, artificial intelligence, and 19 00:01:08,520 --> 00:01:11,920 this crazy AI world we live in. But it's all underpinned by data 20 00:01:11,920 --> 00:01:15,480 engineering. And with me, as always, is my favoritest data 21 00:01:15,480 --> 00:01:19,320 engineer in the world. Even my dog is barking, giving you a shout out. 22 00:01:20,030 --> 00:01:23,790 Andy Leonard. How's it going, Andy? It's going well, Frank. How are you? 23 00:01:24,110 --> 00:01:26,750 I'm doing well, I'm doing well. I'm keeping busy. 24 00:01:28,430 --> 00:01:32,230 We were talking about other podcasts that we have and 25 00:01:32,230 --> 00:01:35,390 the other one is Impact Quantum. So go to impactquantum.com 26 00:01:36,030 --> 00:01:39,070 definitely check it out. And had a very fascinating 27 00:01:40,110 --> 00:01:43,870 conversation with our guest in the virtual green room. So without 28 00:01:43,870 --> 00:01:47,710 further ado, let's welcome Carmen Lee to the show. She's 29 00:01:47,710 --> 00:01:51,390 the CEO of Silicon Data and she is driven by a 30 00:01:51,390 --> 00:01:55,190 passion for developing and delivering cutting 31 00:01:55,190 --> 00:01:58,870 edge derivative products and data solutions that 32 00:01:58,870 --> 00:02:02,550 provide essential data, intelligence and efficiency to compute 33 00:02:02,550 --> 00:02:05,710 markets worldwide. Her company's vision is to 34 00:02:05,710 --> 00:02:09,350 revolutionize these markets through unparalleled data transparency 35 00:02:09,590 --> 00:02:13,030 and financial innovation. Welcome to the show, Carmen. 36 00:02:13,660 --> 00:02:17,380 Thank you. You deliver up my tagline so well I might want to 37 00:02:17,380 --> 00:02:21,020 hire you to do the whatever. Thank you. This is like. 38 00:02:21,340 --> 00:02:23,940 Thank you. This is like I was looking the other day. This is almost our 39 00:02:23,940 --> 00:02:27,580 400th show, so I do have a face for radio and 40 00:02:27,580 --> 00:02:30,620 apparent thankfully. But a voice for radio. So good for me. 41 00:02:31,980 --> 00:02:35,620 This is great. And speaking of radio, we were geeking out because 42 00:02:35,620 --> 00:02:39,260 I started my career in New York in finance 43 00:02:39,260 --> 00:02:43,020 and Bloomberg. Having a Bloomberg terminal on your desk was 44 00:02:43,020 --> 00:02:46,550 a status symbol. There were the ones who had it and the ones who didn't 45 00:02:46,550 --> 00:02:50,390 and the ones who wanted it. And you know radio, 46 00:02:50,390 --> 00:02:53,550 right? Bloomberg radio, which we also get here in dc. And you used to work 47 00:02:53,550 --> 00:02:57,310 for Bloomberg, so that's really cool. That's right. I had a great time 48 00:02:57,310 --> 00:03:00,990 working for Bloomberg and my team was part of the 49 00:03:00,990 --> 00:03:04,790 data team I thought is 50 00:03:04,790 --> 00:03:08,430 one of the most cutting edge data company especially in the 51 00:03:08,430 --> 00:03:12,220 financial services industry. Back then I cover all content, 52 00:03:12,460 --> 00:03:16,220 all product data integrations with any third 53 00:03:16,220 --> 00:03:20,020 party ecosystems. So think about any training 54 00:03:20,020 --> 00:03:23,580 cycles from Fedmin back offices, think about any 55 00:03:24,140 --> 00:03:27,860 cloud providers and database Systems and 56 00:03:27,860 --> 00:03:31,260 even AI, LLMs, whatever you call them, 57 00:03:31,580 --> 00:03:34,060 different use cases, real time 58 00:03:35,180 --> 00:03:38,860 reference data, aesthetic data, anything. It's 59 00:03:38,940 --> 00:03:42,760 really fascinating. I learned a lot my background before that I 60 00:03:42,760 --> 00:03:46,160 was in all financial services and I don't know if I bore your audience at 61 00:03:46,160 --> 00:03:49,800 this point. I started my career in trading, high frequency trading 62 00:03:50,600 --> 00:03:53,720 in Chicago. So to me transparency, 63 00:03:53,720 --> 00:03:57,400 efficiency and free market is sort of in my blood. 64 00:03:57,959 --> 00:04:01,800 100% brainwashed at this point in life. So one of 65 00:04:01,800 --> 00:04:05,520 the things I noticed when I was a Bloomberg is there's a 66 00:04:05,520 --> 00:04:07,720 lot of interesting ecosystem 67 00:04:09,580 --> 00:04:13,420 platform came up last year, right? So they all leveraging gen 68 00:04:13,420 --> 00:04:16,780 AI. You're the first few adopters which is good for them 69 00:04:17,420 --> 00:04:21,220 and their client basis sometimes can be financial institutions. So boom, client 70 00:04:21,220 --> 00:04:24,460 basis. So one of the things I noticed is it was a really fascinating conversation. 71 00:04:24,860 --> 00:04:28,460 So those startups, they're gaining a lot of tractions. Good for them. So 72 00:04:28,460 --> 00:04:31,860 obviously I was like oh you're doing so well. And they will complain to me 73 00:04:31,860 --> 00:04:35,340 saying that they were sassed, right? They were 100% SaaS 74 00:04:35,340 --> 00:04:38,850 revenue so static and then it's pivoting to 75 00:04:39,090 --> 00:04:42,770 AI driven SaaS. So their cost, think about last year The 76 00:04:42,770 --> 00:04:46,450 GPU per GPU per hour was like $9 or 77 00:04:46,450 --> 00:04:50,290 6, 7, 9. Back to like 3 if 78 00:04:50,290 --> 00:04:53,610 you own interruptible instances, right? So the swing is like 79 00:04:53,610 --> 00:04:57,450 300% within the same day but then their revenue 80 00:04:57,450 --> 00:05:00,450 is static, right? So their margin like 81 00:05:00,770 --> 00:05:04,130 positive 40% to negative, 60 to 82 00:05:04,130 --> 00:05:07,790 positive and there's no way for them to manage it. And also 83 00:05:07,790 --> 00:05:11,510 same time it's not like they bring on more clients. They 84 00:05:11,510 --> 00:05:15,110 can enjoy the scalability. It's like again 85 00:05:15,110 --> 00:05:18,910 same thing, the margin is uncontrollable and they have this problem say how 86 00:05:18,910 --> 00:05:22,509 do they actually coming out a cash flow plan for next year and then 87 00:05:22,509 --> 00:05:25,830 they obviously complain. Totally strikes me to be 88 00:05:25,990 --> 00:05:29,270 hey, this industry needs financial 89 00:05:30,070 --> 00:05:33,590 infrastructure layer, right? It's almost like talking to American 90 00:05:33,670 --> 00:05:37,270 Airlines. Say hey airline, you cannot hatch your oil prices 91 00:05:37,270 --> 00:05:40,790 fluctuation. How are they Going to price their tickets. They can't, right? 92 00:05:40,870 --> 00:05:44,310 And it's not like American Airline cost OPEX in like give me five year long 93 00:05:44,310 --> 00:05:48,150 contract. They don't do that. Every single of those commodities 94 00:05:48,310 --> 00:05:52,070 pricing discovery and hedging happens in divers market. So 95 00:05:52,070 --> 00:05:55,670 futures options because there's a few reasons, right? Number one is it's just 96 00:05:55,670 --> 00:05:59,470 efficient. Number two is cheap. Both is flexible and then you and me, 97 00:05:59,470 --> 00:06:02,590 we can do the same thing. We have oil exposures, we don't have to be 98 00:06:02,590 --> 00:06:06,360 American airline. But today if you are crowing for 99 00:06:06,360 --> 00:06:10,080 hyperscalers, you can go to those, you know, 100 00:06:12,400 --> 00:06:15,760 whoever, right? Produce chips, right? And get a long term contract. 101 00:06:16,000 --> 00:06:19,599 But you, if you and me start Neo Cloud, guess what? We don't have access 102 00:06:19,599 --> 00:06:23,440 to kind of pricing. It's not good. We you have a 103 00:06:23,440 --> 00:06:27,160 few players who have the pricing, who have that way to hedge it. But 104 00:06:27,160 --> 00:06:30,710 then the smaller Prius just couldn't get in the game, right? It's really not good 105 00:06:30,710 --> 00:06:34,510 for the ecosystem's health performance and 106 00:06:34,510 --> 00:06:37,390 the risk management. So that's really struck my core. 107 00:06:38,190 --> 00:06:41,790 Last year I was like man, someone needs to do the 108 00:06:41,790 --> 00:06:45,270 index, the pricing, the benchmarking layer of the 109 00:06:45,270 --> 00:06:49,070 GPU compute as human resource 110 00:06:49,070 --> 00:06:52,430 I feel like will be the biggest human resource in the next few years. 111 00:06:52,430 --> 00:06:56,270 Surpass all energy combined, right? So that's why I left Bloomberg right 112 00:06:56,270 --> 00:06:59,590 away. Super passionate. I think we can bring so much transparency to 113 00:06:59,590 --> 00:07:03,390 ecosystem will benefit everybody, right? Not only benefiting other people, 114 00:07:03,390 --> 00:07:07,110 needs compute benefiting like you know, the end consumers. Because think about 115 00:07:07,110 --> 00:07:10,710 the whole funnel, right? You had finance and gpu the 116 00:07:10,710 --> 00:07:14,470 actual clusters cost, right? So 117 00:07:14,470 --> 00:07:17,990 if the banks don't have enough information or hedging for the 118 00:07:17,990 --> 00:07:21,800 banks then they have to charge you high interest, they have no other way. Or 119 00:07:21,800 --> 00:07:25,640 you have to look for alternative capitals which traditionally 120 00:07:25,640 --> 00:07:29,400 they're more expensive, right? Because they're not banks. Banks are cheap as a 121 00:07:29,400 --> 00:07:33,200 cost of capital, right? So then the cost from you 122 00:07:33,200 --> 00:07:36,920 know, stage zero is high. Then think about the second stage, third 123 00:07:36,920 --> 00:07:40,680 stage and then people like you and me using Sora with OpenAI 124 00:07:41,160 --> 00:07:45,000 everything will be more expensive because of that, right? So fix the problem 125 00:07:45,080 --> 00:07:48,890 with transparency from this from Gecko is really really 126 00:07:48,890 --> 00:07:52,490 critical and then their benchmarkings and encourage the 127 00:07:52,490 --> 00:07:56,210 secondary markets and all those flexibility and then 128 00:07:57,730 --> 00:08:01,010 availability will be really incredible to benefit the whole 129 00:08:01,010 --> 00:08:04,850 ecosystem. Interesting. So is it fair to say you've built basically a 130 00:08:04,850 --> 00:08:08,370 futures market for GPU. Compute I building a 131 00:08:08,370 --> 00:08:12,050 benchmark index layer. We are working with future exchange, 132 00:08:12,050 --> 00:08:15,780 right? So I'm not a futures exchange so that would be something we 133 00:08:15,780 --> 00:08:19,300 will think about S and P. Right. So they license the index with a. Right, 134 00:08:19,300 --> 00:08:22,940 right, right, right, right. That's what we do. Right. Well, we will index 135 00:08:22,940 --> 00:08:26,300 to an exchange and they will have futures options on top of that and other 136 00:08:26,300 --> 00:08:30,140 financial products. That's a fascinating concept because like 137 00:08:30,140 --> 00:08:33,820 you're right, we need that because the scarcity 138 00:08:33,820 --> 00:08:37,020 of GPU compute is a real issue. It comes up. 139 00:08:37,500 --> 00:08:41,349 And if, if, if Amazon, the rate. Of volatility, how do 140 00:08:41,349 --> 00:08:44,711 you. With, with. With like 40, 141 00:08:44,827 --> 00:08:48,589 60% fluctuation every daily volatility and then it's 142 00:08:48,589 --> 00:08:51,749 just not a, a very transparent market 143 00:08:52,149 --> 00:08:55,349 which is. Breeds inefficiency. Right. 144 00:08:57,429 --> 00:09:00,869 Absolutely. So for those of. Oh, sorry. Go ahead, 145 00:09:00,869 --> 00:09:03,829 Andy. Okay. I was just going to ask. So are you tracking 146 00:09:04,709 --> 00:09:08,469 features and functionality and all of that? That, that would be the. How you value 147 00:09:08,549 --> 00:09:12,350 the GPU itself and compare that to the price and 148 00:09:12,350 --> 00:09:15,910 you're coming up with some ratio. Exactly. So 149 00:09:16,390 --> 00:09:20,070 compute is not like. Unfortunately it's not as easy as electricity 150 00:09:20,150 --> 00:09:23,990 or even oil have different grid. Right. So even 100 151 00:09:23,990 --> 00:09:27,790 has different configurations. Right. They all, it's not the same. Right. 152 00:09:27,790 --> 00:09:31,590 Different CPUs, different RAMs and geolocation matters. 153 00:09:31,590 --> 00:09:35,110 Right. So a lot of things. So normalization become very critical 154 00:09:35,110 --> 00:09:37,670 component to financially settle index. 155 00:09:40,530 --> 00:09:44,370 Right now we have H100A100 indexes published at Bloomberg and Refinitiv. 156 00:09:44,850 --> 00:09:48,530 So the way we do it is we have a base case and all 157 00:09:48,530 --> 00:09:52,130 the factors normalize to the base case. And the way we normalize 158 00:09:52,450 --> 00:09:56,130 historical data, what factor is actually important to the users, 159 00:09:56,450 --> 00:09:59,850 the CPU matter? How much does it matter? What's the wave, whatever it 160 00:09:59,850 --> 00:10:03,690 contributes. How often do we calibrate? Maybe it matters 161 00:10:03,690 --> 00:10:07,480 today, maybe tomorrow. This, this particular. Whatever 162 00:10:07,640 --> 00:10:10,280 inputs value more. Right. So we do calibration, 163 00:10:11,320 --> 00:10:12,920 period of calibration as well. 164 00:10:14,520 --> 00:10:18,360 Interesting. Yeah, it's fascinating to kind of see because I mean 165 00:10:18,360 --> 00:10:21,640 it always seemed like there's something missing around 166 00:10:23,960 --> 00:10:27,560 the GPU market. Right. Because it's just. And I also think too 167 00:10:27,640 --> 00:10:31,200 it's been a while since we had any kind of compute limitations on what we 168 00:10:31,200 --> 00:10:34,920 wanted to do. Right. Like that CPU is like. Yeah, it's cheap 169 00:10:34,920 --> 00:10:38,560 and you can get what you want and it's not supply demand kind of shifting. 170 00:10:38,560 --> 00:10:41,960 Yeah, I agree. Right. So I didn't really think of like, 171 00:10:42,360 --> 00:10:46,080 you know, kind of this, this market kind of response to 172 00:10:46,080 --> 00:10:49,720 it, which I think is, is an interesting approach and I think, I think, 173 00:10:49,800 --> 00:10:53,479 I think it's fascinating. Yeah. Even if you think about 174 00:10:53,480 --> 00:10:57,000 AI SaaS company. Right. I don't know if you heard the saying that 175 00:10:57,240 --> 00:11:00,360 SAS is 80% margin AI SaaS is 0% Mar. 176 00:11:03,010 --> 00:11:06,770 So I mean it depends on how you run your workflow. If 177 00:11:06,770 --> 00:11:10,530 you are not being thoughtful, right. 178 00:11:10,530 --> 00:11:14,050 You just dump everything, everything you need to do into the most 179 00:11:14,850 --> 00:11:18,490 expensive closed source model. And you're not 180 00:11:18,490 --> 00:11:22,010 optimizing your thinking tokens, your input tokens, 181 00:11:22,010 --> 00:11:25,170 output tokens. It can get very pricey very 182 00:11:25,330 --> 00:11:28,770 quickly, right? Not batching it, you're not doing all the right things. 183 00:11:29,850 --> 00:11:33,130 And even you do all the right things, it's gonna be such a meaningful 184 00:11:34,730 --> 00:11:38,490 percentage of your cost. And then all those companies not ready for it. Right. 185 00:11:38,650 --> 00:11:42,090 Because in before what's the raw material cost? 186 00:11:42,730 --> 00:11:46,170 Electricity. Like really nothing. Right. But now 187 00:11:47,210 --> 00:11:50,890 every company becomes, you know, a which is great company 188 00:11:50,890 --> 00:11:54,490 but then their cost structure is shifting from zero cost 189 00:11:54,570 --> 00:11:58,340 to. To 40%, 60%, any percent to token 190 00:11:58,340 --> 00:12:02,020 or to GPU at the end. Right? Right. So how do you think about hedging 191 00:12:02,020 --> 00:12:05,300 that kind of cost component? Can you control that? Can you optimize for it? Can 192 00:12:05,300 --> 00:12:09,140 you monitor it, can you benchmark it? You know, can you hedging it? So 193 00:12:10,340 --> 00:12:12,580 no, that's a good point. So do you think 194 00:12:14,420 --> 00:12:17,980 there's multiple, I guess, inputs and levers to this? Right. Because it doesn't seem like 195 00:12:17,980 --> 00:12:21,460 this would be a straight thing. So what's, you know, Andy mentioned that you were 196 00:12:21,460 --> 00:12:25,220 tracking certain benchmarks. Like what benchmarks are you tracking? Because I'm very curious about 197 00:12:25,220 --> 00:12:28,860 this. Right? So there's a few things, depends on 198 00:12:28,860 --> 00:12:32,580 your position at least this can change 199 00:12:32,580 --> 00:12:36,340 every single day. Like our ecosystem is so nuts, right? So it 200 00:12:36,340 --> 00:12:39,500 depends your 201 00:12:39,580 --> 00:12:42,940 positioning, the whole workflow, right. So think about if you are new 202 00:12:42,940 --> 00:12:45,740 clouds, you are selling token, right? 203 00:12:47,260 --> 00:12:50,900 The cost for you is a gpu, right. So then your margin 204 00:12:50,900 --> 00:12:54,730 becomes the diff between the margin and the GPU cost. And that's 205 00:12:54,730 --> 00:12:57,930 the way we calculate it, right. Which is different units. 206 00:12:58,730 --> 00:13:02,330 And then your worry is okay, so for 207 00:13:02,330 --> 00:13:05,930 token survey for the tokens, how much money can I get rate 208 00:13:06,170 --> 00:13:09,930 from one particular gpu? The flops, right? How can I optimize for that? And 209 00:13:10,090 --> 00:13:13,690 what if I'm doing even hosting open source models? 210 00:13:13,770 --> 00:13:17,530 And how do I make sure people using that open source model, should I 211 00:13:17,530 --> 00:13:21,250 shifting it? What's the pricing for that? Think about that strategy and GPU said 212 00:13:21,250 --> 00:13:24,960 okay, am I renting GPUs? I'm like outright 213 00:13:24,960 --> 00:13:28,800 purchase those GPUs and put on my books and depreciate it. How 214 00:13:28,800 --> 00:13:32,520 long can I depreciate it for? How do I let's say if 215 00:13:32,520 --> 00:13:36,160 everyone's the latest and greatest, I'm selling the GPU after second, third year, 216 00:13:36,240 --> 00:13:40,000 what's the terminal value for the GPUs? Who should validate that? Which bank should 217 00:13:40,080 --> 00:13:43,720 depreciate the asset classes. So it's a lot of things coming to the new 218 00:13:43,720 --> 00:13:47,360 cloud space. If you think about your inferencing infrastructure, 219 00:13:47,360 --> 00:13:50,930 right? So let's say you're 220 00:13:51,090 --> 00:13:54,610 AI tech company, right? Then your revenue is token, 221 00:13:54,770 --> 00:13:58,610 right? Ideally they're paying you based on token 222 00:13:58,610 --> 00:14:02,330 use cases as well. And then your cost is token which is 223 00:14:02,330 --> 00:14:05,650 easier but same time for you is thinking through okay so 224 00:14:06,450 --> 00:14:09,650 right now open source tokens, the price 225 00:14:10,290 --> 00:14:13,650 they do move up and down. For example 226 00:14:14,660 --> 00:14:18,140 if you look at Deep Seq, even Deep Seq, they host their own servicing but 227 00:14:18,140 --> 00:14:21,980 then the price changes, they have the off peak hours and that change all the 228 00:14:21,980 --> 00:14:25,780 time. Or you can do closed source which the price is pretty 229 00:14:25,780 --> 00:14:29,580 static. The way I think about it is again it's extremely 230 00:14:29,580 --> 00:14:33,380 free market approach, right? Is how can we 231 00:14:33,380 --> 00:14:37,060 make sure especially open source ones, the token prices 232 00:14:37,460 --> 00:14:41,220 is driven by the market demand supply curve, 233 00:14:41,710 --> 00:14:45,550 right? Let's say if everyone, if I have like 100 GPUs 234 00:14:45,550 --> 00:14:49,390 right now and obviously let's say I 235 00:14:49,390 --> 00:14:52,910 choose to host only one llama open source 236 00:14:52,910 --> 00:14:56,630 model and then I know I can produce X amount of tokens, 237 00:14:56,630 --> 00:15:00,230 both input output tokens, right? And I can just auction off 238 00:15:00,230 --> 00:15:03,670 and you guys and you can buy a million token and one day he's like 239 00:15:03,670 --> 00:15:06,590 I'm not going to use it, why do I sell it to Frank? Can this 240 00:15:06,590 --> 00:15:10,430 be some market where right now you are stuck 241 00:15:10,430 --> 00:15:14,200 with it, right? In 242 00:15:14,200 --> 00:15:17,760 my mind, unfortunately I'm very brainwashed to free market. I feel like you have to 243 00:15:17,760 --> 00:15:21,440 give people option. The more option you give people and 244 00:15:21,440 --> 00:15:25,040 any have flexibility, franchise flexibility and people more 245 00:15:25,040 --> 00:15:28,800 willing to participate because they know they can get out. Because right now you're stuck 246 00:15:28,960 --> 00:15:32,320 with hyperscaler GPUs or any tokens, you're stuck with it 247 00:15:32,560 --> 00:15:36,320 and then you're less likely to commit because you know you 248 00:15:36,320 --> 00:15:39,680 can get out or you get fined even worse, right? You know those cases, you 249 00:15:39,680 --> 00:15:42,320 get fined millions of dollars when you back out on cloud deals. 250 00:15:43,890 --> 00:15:47,530 That's one of the things I really think I should encourage people thinking about tokens 251 00:15:47,530 --> 00:15:50,810 and GPUs as a main cost structure. How can we drive 252 00:15:50,810 --> 00:15:54,650 efficiency so people can commit and then get out if 253 00:15:54,650 --> 00:15:58,210 they need to and then swap out and everyone gets more value 254 00:15:58,850 --> 00:16:02,490 and efficiency from those transactions. So is it 255 00:16:02,490 --> 00:16:05,330 more like an exchange or an auction? 256 00:16:06,130 --> 00:16:09,930 What's the mechanism? Right? So from token GPU side 257 00:16:09,930 --> 00:16:13,330 obviously there's Spot exchange already like compute 258 00:16:13,330 --> 00:16:17,180 exchange, where you can actually tell them, hey, I need this 259 00:16:17,180 --> 00:16:20,860 configuration how many nodes? And then they will 260 00:16:20,860 --> 00:16:24,500 say okay, let's do an auction. And then the 261 00:16:24,500 --> 00:16:27,340 best price, best quality, whatever combination wins. Right? 262 00:16:27,900 --> 00:16:31,700 Yeah. You can potentially do other asset class as well. Right. So we're. 263 00:16:31,700 --> 00:16:35,540 Siliconita is a data company. So think about us as the Bloomberg and there's the 264 00:16:35,540 --> 00:16:39,220 Nices, NASDAQqs and everybody, right. This spot, right, 265 00:16:39,220 --> 00:16:42,880 you can actually get GPUs. You, you can actually get stocks from those exchanges. 266 00:16:43,040 --> 00:16:46,880 And the FAST is we collect data from those exchanges like 267 00:16:46,880 --> 00:16:50,160 Bloomberg. Right. And then we'll produce financial products on top of that. Right. 268 00:16:50,400 --> 00:16:54,200 So that's right, there's spot, which is the 269 00:16:54,200 --> 00:16:57,920 nasdaq. Right. You can buy and sell, get actual physical 270 00:16:57,920 --> 00:17:01,520 deliveries, all the compute or token you need. And there's 271 00:17:01,520 --> 00:17:05,160 data side which is making data the Bloomberg. Right. And then FAST 272 00:17:05,160 --> 00:17:09,010 is structurally the financial products layer. Right, data layer. And 273 00:17:09,010 --> 00:17:12,370 then we're agnostic, meaning we look agnostic of chips, 274 00:17:12,370 --> 00:17:15,530 agnostic of spot markets, agnostic of everything. Right. 275 00:17:16,170 --> 00:17:19,370 And it's a future exchange which they license 276 00:17:19,930 --> 00:17:23,769 our indexes to create futures product. Ideally we're 277 00:17:23,769 --> 00:17:27,290 settling to spa. Maybe some of them will sell at spa. Right. So it's pretty 278 00:17:27,610 --> 00:17:30,650 standard practices. So 279 00:17:31,210 --> 00:17:35,010 would the currency or the coin of the realm be tokens 280 00:17:35,010 --> 00:17:38,680 or compute time or compute seconds? Things 281 00:17:38,680 --> 00:17:42,280 change. It's, it's making my life really fun 282 00:17:42,280 --> 00:17:46,040 and you know, also different. Yeah, all the time. 283 00:17:46,280 --> 00:17:50,040 And then you, you mentioned you have this quantum thing, right? Right. 284 00:17:50,600 --> 00:17:53,880 It's a lot. We track all compute. So it doesn't for us 285 00:17:54,360 --> 00:17:57,960 what chips and what, what architecture framework 286 00:17:58,120 --> 00:18:01,680 and you know, we don't really care. We benchmark the performances and the data 287 00:18:01,680 --> 00:18:05,260 inside. And everything we don't know for us is getting 288 00:18:05,260 --> 00:18:08,940 ready for everything. So we want to create product 289 00:18:08,940 --> 00:18:12,620 that's actually going to be helpful to the marketplaces, not just creating 290 00:18:12,620 --> 00:18:16,300 things like gambling table. People bet on binary things. Right. For 291 00:18:16,300 --> 00:18:19,380 us, how can we make it useful for the people who actually 292 00:18:20,180 --> 00:18:23,940 naturally long compute? So the Neo clouds everybody else, 293 00:18:24,260 --> 00:18:27,620 they need product to hedge their revenue fluctuation. Right. 294 00:18:27,940 --> 00:18:31,780 So they issue short futures and whoever naturally short compute. 295 00:18:31,780 --> 00:18:34,760 So you need computer and for you is a cost management. 296 00:18:35,400 --> 00:18:39,040 So I want to make sure my product is usable by them. It 297 00:18:39,040 --> 00:18:42,680 depends on how they pay. Right. If they pay tokens, 298 00:18:42,760 --> 00:18:46,480 nothing to create token products. You're very right now paying people paying 299 00:18:46,480 --> 00:18:50,040 per GP power and you create product for that. If they pay 300 00:18:50,200 --> 00:18:53,840 things all right, then it's different contracts for that. 301 00:18:53,840 --> 00:18:56,920 So it really depends on how people using it today and tomorrow. 302 00:18:57,640 --> 00:19:01,450 And then, you know, we. We hyped to create products that may 303 00:19:01,450 --> 00:19:05,050 not be the S&P 500, which live forever. We probably create financial 304 00:19:05,050 --> 00:19:08,810 products live for next five to 10 years. Because guess what? Chips 305 00:19:08,810 --> 00:19:12,650 what our style, right? The A100 people still using it, but 306 00:19:12,650 --> 00:19:16,330 like L4s, people are using it, but like other chips like the V's, 307 00:19:16,330 --> 00:19:19,930 the, you know, probably not as much. Right. Then similarly, my 308 00:19:19,930 --> 00:19:23,770 financial products associated with that underlying asset 309 00:19:23,770 --> 00:19:27,400 probably will, you know, retire, be retired. Right? Which is fine. 310 00:19:28,120 --> 00:19:30,760 That's cool. I'm sorry, go ahead, Andy. 311 00:19:31,880 --> 00:19:35,720 I was just thinking about it and a couple of ideas popped into 312 00:19:35,720 --> 00:19:39,160 my head as you were describing that, Carmen. One is 313 00:19:39,240 --> 00:19:42,760 capacity. It sounds like you're literally selling 314 00:19:42,760 --> 00:19:46,440 compute capacity, GPU capacity, time, just 315 00:19:46,440 --> 00:19:50,040 whatever. But it kind of falls into that bucket under one hand. 316 00:19:50,440 --> 00:19:53,800 But then on the other hand, it seems like that 317 00:19:54,520 --> 00:19:57,800 it almost creates this utility market. 318 00:19:59,480 --> 00:20:03,160 Is that fair or am I missing something, right? No, 319 00:20:03,240 --> 00:20:06,800 you're right. But two pieces. So one is a compute exchange part, right? This is 320 00:20:06,800 --> 00:20:10,040 where you can actually get either depends on what people, 321 00:20:10,760 --> 00:20:14,600 the mode of people preferences. You can get GPUs or get 322 00:20:14,600 --> 00:20:18,400 tokens, whatever, right. Physically delivered, you do you. You 323 00:20:18,400 --> 00:20:22,050 don't have to touch any financial products, right. It is literally like you going to 324 00:20:22,050 --> 00:20:25,890 a store buying stuff. And then the more option based, right. 325 00:20:25,890 --> 00:20:29,730 You can actually get instances. And the silicon data is. You 326 00:20:29,730 --> 00:20:33,410 cannot actually getting any compute. Right? Like you cannot 327 00:20:33,410 --> 00:20:36,650 get any stocks from Bloomberg. Well, you can get this data. 328 00:20:37,370 --> 00:20:40,970 What asset is trading, what prices? So that informal decision, ideally 329 00:20:40,970 --> 00:20:44,770 in your spot market be like, hey, I think everyone, you know, 330 00:20:44,770 --> 00:20:47,730 the H100 price is a little too high, in my opinion. I'm not going to. 331 00:20:47,730 --> 00:20:51,140 Right. Right now, like, forget about this. And I can totally use a 332 00:20:51,140 --> 00:20:54,660 100. Right. It's fine. So this data is data 333 00:20:54,660 --> 00:20:58,340 layer, which is liquid data, right? So those are those the 334 00:20:58,340 --> 00:21:02,020 sort of two pieces to I guess resolve the 335 00:21:02,020 --> 00:21:05,860 workflow equation. So it's kind of like when you go to the supermarket. I'm 336 00:21:05,860 --> 00:21:08,500 sorry, Andy. When you go to. That's okay, go ahead. When you go to the 337 00:21:08,500 --> 00:21:11,900 supermarket, you buy the beef, you buy the pork, but you don't think about the 338 00:21:11,900 --> 00:21:14,900 pork belly futures and stuff like that. It's kind of abstracted away from you. 339 00:21:15,460 --> 00:21:18,940 Exactly. The farmers will think about this, right? Yeah, farmers think about it. 340 00:21:18,940 --> 00:21:22,780 Yeah. They need to hatch the corn futures, right? But if 341 00:21:22,780 --> 00:21:26,420 you are farmer, you still say you were someone to eat the 342 00:21:26,740 --> 00:21:29,140 corn. You go supermarket, you don't think about, hey, 343 00:21:30,500 --> 00:21:34,020 Right. So you may have covered this already, but how does 344 00:21:34,260 --> 00:21:36,900 or does fungibility come into play? 345 00:21:37,780 --> 00:21:41,540 It's a great question. So I went through so many different iterations about this. 346 00:21:43,790 --> 00:21:47,510 Initially I was like, okay, why don't I just normalize across flops? And I was 347 00:21:47,510 --> 00:21:51,310 like, nope, can't do that because there's 348 00:21:51,310 --> 00:21:55,150 just, there's so many things wrong with this approach. But obviously 349 00:21:55,150 --> 00:21:58,910 we can dig into details, but we're not going to do that. And then secondly 350 00:21:58,910 --> 00:22:02,190 is okay, why don't we do like inferencing 351 00:22:02,350 --> 00:22:06,190 chips? Like just make a pot and then we realize, okay, how can. 352 00:22:06,270 --> 00:22:10,110 So again back to the initial question. I want to make product actually going 353 00:22:10,110 --> 00:22:13,740 to help people hedging. Right. If you 354 00:22:13,740 --> 00:22:17,380 do a combination of different chips, then if you 355 00:22:17,380 --> 00:22:20,220 are and you know we're using of a lot of people, are you going to 356 00:22:20,220 --> 00:22:23,540 really use that to hatch? How would some correlation look like. Right. 357 00:22:23,780 --> 00:22:27,619 Maybe you just rather have different chip types and then just hatch accordingly 358 00:22:27,619 --> 00:22:31,380 because the correlation will be much higher than the combination of indexes. 359 00:22:31,620 --> 00:22:35,380 Maybe the composition of indexes is good for just tracking 360 00:22:35,460 --> 00:22:39,220 general, but not for actually financial products. So we have, we have, 361 00:22:39,220 --> 00:22:43,060 we can have all. Some of them will be tradable. Some of them. Well, right. 362 00:22:43,540 --> 00:22:47,100 For us is if people start, if, if we move to the world 363 00:22:47,100 --> 00:22:50,740 where it's not going to be Nvidia only kind of play 364 00:22:50,740 --> 00:22:54,339 in the like amd. We can eventually, 365 00:22:54,339 --> 00:22:58,100 it'll probably end eventually. Well, we'll see when, right? We'll 366 00:22:58,100 --> 00:23:01,860 see quantum happens first or everyone catching up first. I have no 367 00:23:01,860 --> 00:23:05,400 idea. Right. So if it's like a more vibrant 368 00:23:05,400 --> 00:23:09,080 ecosystems. Right. And then maybe we're thinking about, hey, maybe we can do 369 00:23:09,080 --> 00:23:12,600 like doing some of the chips. Even different firms would normalize it and then we 370 00:23:12,600 --> 00:23:16,280 do something like a inferencing chips, chaining chips. I don't 371 00:23:16,280 --> 00:23:20,080 know. So that's another thing. Or like token, token indexes. Right. 372 00:23:20,080 --> 00:23:23,800 So can we do open source ones? Multimodality? Is 373 00:23:23,800 --> 00:23:27,040 multimodality going to be a thing in a few years? Everything going to go back 374 00:23:27,040 --> 00:23:30,880 to one model only? Because right now with different models. But maybe it's the interim 375 00:23:30,880 --> 00:23:34,580 stage. Right. We. I don't know. So it's one of the things we have to 376 00:23:34,580 --> 00:23:37,780 keep like looking and thinking and just moving things 377 00:23:38,100 --> 00:23:41,900 forward. Yeah, I was thinking too about, you 378 00:23:41,900 --> 00:23:45,660 know, the, the amortization that people 379 00:23:45,660 --> 00:23:49,500 do in their heads at least when they buy a new car. Yeah. So 380 00:23:49,500 --> 00:23:53,200 that's the math is you drive it off the lot, it's worth what, a 75, 381 00:23:53,320 --> 00:23:54,900 80% of what you pay for. 382 00:23:58,320 --> 00:24:01,440 So we need a Carfax for GPUs, right? So that's what we do too 383 00:24:02,480 --> 00:24:06,240 for silicon Mark. So what we do is okay, everything. Well 384 00:24:06,240 --> 00:24:10,040 at least right now or before Last year or T minus 1, everything 385 00:24:10,040 --> 00:24:13,200 is brand new. So okay, we'll take whatever the 386 00:24:13,599 --> 00:24:17,280 number they published and tdbs, the flops, we all know 387 00:24:17,280 --> 00:24:20,080 there's like haircut to that number. 388 00:24:20,880 --> 00:24:24,640 That's funny, right? And then a year later, right, A year later I 389 00:24:24,640 --> 00:24:28,410 say, Andy, you're growing great in great data centers. Your 390 00:24:28,410 --> 00:24:32,170 thermal cooling was doing great. I'm old data 391 00:24:32,170 --> 00:24:35,970 center, I don't have the latest cooling. Obviously my chip 392 00:24:35,970 --> 00:24:39,210 is after year. You can argue they own different curves, 393 00:24:39,370 --> 00:24:43,170 decay curve. And are we treating the same prices even 394 00:24:43,170 --> 00:24:46,890 though same configuration? Probably we shouldn't. Should it be reflection of 395 00:24:47,290 --> 00:24:50,890 the actual quality? So that's something Mark does. 396 00:24:51,050 --> 00:24:54,410 And then we do things even more basic than that. So number one is 397 00:24:55,110 --> 00:24:58,710 when you tell me you have H100 like 100 nodes, each node has 398 00:24:58,710 --> 00:25:02,350 say 8 GPUs, right? Yeah. Is that true? Can I 399 00:25:02,350 --> 00:25:06,190 number one verify the UID of that? And you see, it's all the CPUs 400 00:25:06,190 --> 00:25:09,950 and this operation systems on all 401 00:25:09,950 --> 00:25:13,790 the nodes, they all live connected. Number one, can we just 402 00:25:13,790 --> 00:25:17,190 verify are they connected? What's the latency? So that's very 403 00:25:17,430 --> 00:25:20,950 basic things, right? So we do that piece at least, you know, 404 00:25:21,840 --> 00:25:25,600 are they truly UIDs and CPUs? The machine, is the machine ever 405 00:25:25,840 --> 00:25:29,120 changed? Because we do mesh IDs based on 406 00:25:29,360 --> 00:25:32,880 CPU changes. We know something changes, right? And then the UID of every 407 00:25:32,880 --> 00:25:36,520 chip. So we do the decay curve for the individual chips and also the machine 408 00:25:36,520 --> 00:25:40,160 level and then thermal staggeration, everything. So we do 409 00:25:40,160 --> 00:25:43,840 that and then we do validation. Almost like Bloomberg Validate fixing 410 00:25:43,840 --> 00:25:47,640 compound. Because you have to understand the issuers and it's 411 00:25:47,640 --> 00:25:50,990 a bridge and it's a school and with cash flows and all those stuff. So 412 00:25:50,990 --> 00:25:54,470 we do that for GPUs. The geolocation. If you build data 413 00:25:54,470 --> 00:25:57,670 centering somewhere in North Korea, 414 00:25:58,070 --> 00:26:01,430 it's great, but no one going to use it, right? 415 00:26:02,630 --> 00:26:06,430 We took all those in considerations when we created those data models. So then 416 00:26:06,430 --> 00:26:09,830 we figured out, okay, so based on the setup and 417 00:26:10,070 --> 00:26:13,830 we run a benchmark on specific GPUs, this is our grade and then 418 00:26:13,830 --> 00:26:16,510 this is our validation. Obviously you can do whatever you want. And then you can 419 00:26:16,510 --> 00:26:19,750 say hey, screw that, I believe this is much higher price. You can do that 420 00:26:19,750 --> 00:26:23,410 as well, Right? But this is our valuation. So almost like a scoring system. 421 00:26:24,050 --> 00:26:27,890 That's interesting. So My mind immediately went 422 00:26:27,890 --> 00:26:31,530 to, when, when we started talking about cars, my 423 00:26:31,530 --> 00:26:34,690 mind immediately went to, you know, the used GPR lot 424 00:26:35,890 --> 00:26:39,090 some guy in bib overhauls out here in Farmville, Virginia 425 00:26:39,410 --> 00:26:42,770 kicking the tires. What's it going to take to get you into this 426 00:26:42,770 --> 00:26:43,570 gpu? 427 00:26:49,800 --> 00:26:53,480 Yep. See, there we go. And network them together. Right. Like I think there's also, 428 00:26:53,800 --> 00:26:57,640 you know, maybe, you know, I don't know if 429 00:26:57,640 --> 00:27:00,600 you've been tracking the, the DGX Spark device 430 00:27:01,320 --> 00:27:05,039 that Nvidia has, but apparently they have ports 431 00:27:05,039 --> 00:27:08,440 in them so you can network I think up to four together. I'm not sure 432 00:27:08,440 --> 00:27:12,240 but yeah, I'm sorry I 433 00:27:12,240 --> 00:27:16,080 cut you off but like. No, no, no. Nvidia we 434 00:27:16,080 --> 00:27:19,520 definitely leveraging a lot of. So we do the container within 435 00:27:19,520 --> 00:27:22,360 container and we do integrate with Nvidia DGX 436 00:27:22,920 --> 00:27:26,200 benchmarking. So they have open sourced some of their LM 437 00:27:26,200 --> 00:27:30,040 benchmarking based on GPUs and we do streamline their products so 438 00:27:30,040 --> 00:27:33,720 you can test lms. So Nvidia Digitex testing 439 00:27:33,720 --> 00:27:37,480 through system data. The benefit is if you do it all right yourself, 440 00:27:37,480 --> 00:27:40,200 number one, you can 441 00:27:41,000 --> 00:27:44,820 obviously people want but people can just change up the, the 442 00:27:45,060 --> 00:27:48,820 benchmark results themselves. Right? It's open source but through us it's data Oracle. You 443 00:27:48,820 --> 00:27:52,500 can't really change results. Number two things is more streamlined. It takes a few hours 444 00:27:52,500 --> 00:27:55,580 to run versus take weeks because you've download a bunch of things you may or 445 00:27:55,580 --> 00:27:58,180 may not need. You may or may not need. 446 00:28:00,020 --> 00:28:03,820 Well, I also think too like, you know, how does this, you know, you 447 00:28:03,820 --> 00:28:07,500 mentioned you, you kind of skirted around the location thing with sovereign 448 00:28:07,500 --> 00:28:11,230 AI, right? So like if I'm okay with using Google 449 00:28:11,230 --> 00:28:14,670 Services, right. And I can, I have access to TPUs, right. I have a lot 450 00:28:14,670 --> 00:28:18,470 more access to whatever Amazon's chip. Microsoft I think is 451 00:28:18,470 --> 00:28:22,110 working on something. Custom that's on prices too, right. The Geolocation 452 00:28:22,590 --> 00:28:26,030 they have different prices and different carbon footprint. We haven't even touched that. 453 00:28:26,510 --> 00:28:30,350 Right, right, right. We do track that as well based 454 00:28:30,350 --> 00:28:33,990 on local grid power grid information. We do track the carbon cost associated with 455 00:28:33,990 --> 00:28:37,690 different AI workflows. I think it's important, I think so 456 00:28:37,690 --> 00:28:41,490 for me is let me at least surfacing the number to you and 457 00:28:41,490 --> 00:28:45,130 you decide what to do with it. Right. So I think that's a good idea 458 00:28:45,130 --> 00:28:47,490 or you know, maybe it turns out that you know, 459 00:28:49,010 --> 00:28:52,810 this type of model of GPU is you know, depending on what your 460 00:28:52,810 --> 00:28:55,010 core. I think it's, I think it's great because I think one of the things 461 00:28:55,010 --> 00:28:58,810 that I've heard And I didn't Peter 462 00:28:58,810 --> 00:29:02,330 Drucker. What gets measured gets managed, right? So you're, what you're doing is you're providing 463 00:29:02,330 --> 00:29:06,060 ways to measure GPUs and GPU performance. Right. 464 00:29:06,060 --> 00:29:09,820 So if I don't care. One of the things I heard about and I'm sure 465 00:29:09,820 --> 00:29:13,420 you have some thoughts on this is like cloud providers that are 466 00:29:13,420 --> 00:29:17,140 starting up and they're just doing 467 00:29:17,860 --> 00:29:21,460 GPUs, right. They're just doing kind of training loads. Right. 468 00:29:21,620 --> 00:29:25,460 And they don't need to be located anywhere special. Right. Like they don't 469 00:29:25,460 --> 00:29:28,860 need to be in the northeast corridor. They could be in the middle of 470 00:29:28,860 --> 00:29:32,450 nowhere as long as they have power. Right. And 471 00:29:32,450 --> 00:29:35,210 because you're going to run a load, right, you're going to run a load on 472 00:29:35,210 --> 00:29:38,530 the thing, it's going to take 72 hours say to run. You don't really care 473 00:29:38,530 --> 00:29:42,250 if the latency is, you know, 150 milliseconds versus 474 00:29:42,250 --> 00:29:45,370 3. Right. It doesn't really matter. Yes. 475 00:29:46,170 --> 00:29:50,010 That's why you see a lot of us get built up in like Iceland, Finland, 476 00:29:50,810 --> 00:29:54,410 the users can be in Americas, can be in Asia. Right, 477 00:29:55,050 --> 00:29:58,870 right. For them is can they get the capacity 478 00:29:58,870 --> 00:30:02,670 looking for and you hard deal if you're thermal powered 479 00:30:02,670 --> 00:30:05,870 data centers, cheap electricity. Yeah. 480 00:30:06,350 --> 00:30:10,150 And then it's cleaner supposedly. Right. As 481 00:30:10,150 --> 00:30:13,470 long as you're not on the volcano belt. 482 00:30:13,870 --> 00:30:16,910 Right. As long as it's not going to blow up. Yeah. 483 00:30:18,510 --> 00:30:22,230 But yeah, so we definitely see that trend and a lot of energies, you 484 00:30:22,230 --> 00:30:26,080 know, what do we call it oversupply sometimes can 485 00:30:26,080 --> 00:30:29,840 be in Spain because overbuilt and the grid couldn't handle it. And 486 00:30:29,840 --> 00:30:33,560 then they need to get data center up and running like now to take over 487 00:30:34,040 --> 00:30:35,560 the power. But then 488 00:30:37,960 --> 00:30:41,600 it takes a lot to make the racks start running. Right. 489 00:30:41,600 --> 00:30:45,240 More than just the GPU itself, you need the connectivities and network 490 00:30:45,320 --> 00:30:48,840 and that could be in shortage. So you need to solve a lot of different 491 00:30:48,840 --> 00:30:52,580 pieces to actually deliver the actual computer. 492 00:30:52,810 --> 00:30:56,250 But that's why it's fascinating industry for us because 493 00:30:56,730 --> 00:31:00,330 we see things from dsml, tsmc, side. 494 00:31:00,330 --> 00:31:04,130 So anything supply demand shifting will have 495 00:31:04,130 --> 00:31:07,610 an impact on the whole ecosystem. And then this industry is winner takes off 496 00:31:09,370 --> 00:31:11,770 from LTSMC to a solution level. 497 00:31:13,290 --> 00:31:17,010 You have to be the solution. Your alternative solution just not 498 00:31:17,010 --> 00:31:20,600 going to work. So. So every single piece is so critical to 499 00:31:20,680 --> 00:31:24,200 the whole chain packaging. Right. You have to work, 500 00:31:24,840 --> 00:31:27,960 right. If you don't know how to do it, then you just can't do it. 501 00:31:27,960 --> 00:31:30,840 It's not like you can buy a cheaper pair of socks or whatever 502 00:31:31,960 --> 00:31:35,560 so we do. We're from end to end, right. From the SM of production, 503 00:31:35,560 --> 00:31:39,320 tsmc. So we're official TSMC partners are going to be actually 504 00:31:39,720 --> 00:31:43,240 TSMC conferences to this 505 00:31:43,240 --> 00:31:47,010 November. Very cool. It is really cool. I 506 00:31:47,010 --> 00:31:50,490 kicked out by those stuff very quickly. And all the way to 507 00:31:50,650 --> 00:31:54,410 the model A, the token layer. Right. Agentic layer. So 508 00:31:54,410 --> 00:31:58,170 we sort of see things all the way. Which 509 00:31:58,970 --> 00:32:02,090 I think my brain get overclocked every single. 510 00:32:04,010 --> 00:32:07,090 I know what you mean because I get till the time of like 511 00:32:07,090 --> 00:32:10,580 2:33pm and I'm like, I can't take any more input. Like 512 00:32:10,740 --> 00:32:14,500 and the muscle, my brain muscle just dead. I know. How 513 00:32:14,500 --> 00:32:17,260 do you do that? How do you get a roller in my brain, just like 514 00:32:17,260 --> 00:32:21,060 relax my brain muscles? I. I found going for a walk 515 00:32:21,060 --> 00:32:23,380 is a. Is a good way to do it. Right. 516 00:32:24,580 --> 00:32:28,060 No, like. And a co worker of mine calls it everything turns to 517 00:32:28,060 --> 00:32:31,820 hieroglyphic hieroglyphics when he's like 518 00:32:31,820 --> 00:32:34,580 looking at like stuff. And I was like, yeah, that's a good way to put 519 00:32:34,580 --> 00:32:37,120 it. Because it's just kind of like, yeah, I can't. I don't want to have 520 00:32:37,120 --> 00:32:40,120 time by a daughter. So I usually spend time with my daughters. I feel like 521 00:32:40,440 --> 00:32:44,280 they've been silly. And I would tell them, I'm so stressed out. When my daughter 522 00:32:44,280 --> 00:32:47,920 was like, me too. I was like, what are you stressed about one last donut 523 00:32:47,920 --> 00:32:51,440 than the other guy. I was like, that's very important thing. I agree with that. 524 00:32:51,440 --> 00:32:54,920 That's very stressful. I will be really upset if I get one less 525 00:32:54,920 --> 00:32:58,560 donut. So. Yeah, so definitely put things in 526 00:32:58,560 --> 00:33:00,840 perspective. Yeah, that's cool. 527 00:33:03,880 --> 00:33:07,360 I think one of the best things. Any other questions? No, plenty, 528 00:33:07,360 --> 00:33:11,040 plenty. Like, I'm just fascinated by this. I know 529 00:33:11,040 --> 00:33:14,840 we're kind of short on time, but one of the things that you mentioned was 530 00:33:14,840 --> 00:33:17,640 tcmc. Tsmc. 531 00:33:18,760 --> 00:33:22,280 So for those who don't know who they are and how important they are to 532 00:33:22,920 --> 00:33:26,360 the global economy, could you explain for those folks 533 00:33:26,520 --> 00:33:29,630 and why I was so excited that you're going to one of their conferences? I 534 00:33:29,630 --> 00:33:33,310 didn't know they had conferences, so. I don't think I would do the justice 535 00:33:33,390 --> 00:33:37,190 of explaining how important TSMC is. All right, how about I explain it and 536 00:33:37,190 --> 00:33:40,910 then you tell me where I'm wrong. I'm sure you'll do a better job 537 00:33:40,910 --> 00:33:44,670 than I can. So. Tsmc. Taiwan 538 00:33:45,630 --> 00:33:49,230 Semiconductor Manufacturing Company. That's right. 539 00:33:50,190 --> 00:33:52,990 They are based in Taiwan. And 540 00:33:54,030 --> 00:33:57,710 the reason why. Nvidia. There's a fascinating 541 00:33:57,710 --> 00:34:00,910 story in the book called the Nvidia Way. I Don't know if you've listened to 542 00:34:00,910 --> 00:34:04,270 that or read that book. Really awesome book. But basically 543 00:34:05,390 --> 00:34:09,230 one of the advantages Nvidia had early on and arguably 544 00:34:09,230 --> 00:34:12,629 now was that they off they outsourced their chip 545 00:34:12,629 --> 00:34:16,390 manufacturing to this company tsmc. I'll get it right that 546 00:34:16,390 --> 00:34:20,110 time. They are basically what they call a fab. 547 00:34:20,430 --> 00:34:23,070 And you could, I mean not 548 00:34:24,170 --> 00:34:27,490 now they're so busy like you know, you kind of the you in general. Right. 549 00:34:27,490 --> 00:34:30,450 Like I couldn't call them up and be like hey, I have some prints for 550 00:34:30,450 --> 00:34:33,010 you. I have some chip designs I want you to make for me. Can you 551 00:34:33,010 --> 00:34:36,650 send me. They're not at that scale but 552 00:34:36,890 --> 00:34:40,610 so they're a fab. And so what happens is people like Nvidia, companies like 553 00:34:40,610 --> 00:34:44,290 Nvidia, a few other companies too will go and they will, they 554 00:34:44,290 --> 00:34:47,610 will design their chips and then they'll, they'll basically 555 00:34:48,810 --> 00:34:51,849 not drop ship but effectively kind of print to order 556 00:34:52,649 --> 00:34:56,249 chips. Which frees up a company like Nvidia 557 00:34:56,809 --> 00:35:00,329 from having to build their own fabs. Kind of like intel does. Is that a 558 00:35:00,329 --> 00:35:04,169 good description? 100 so I usually call 559 00:35:04,249 --> 00:35:07,809 on Nvidia and AMD like design houses and then sometimes 560 00:35:07,809 --> 00:35:11,449 confused with people who's like oh, are they like Louis Vuitton was like no, 561 00:35:11,849 --> 00:35:15,289 Right, right. Or like graphic designers? Yeah, yeah. So they're design 562 00:35:15,289 --> 00:35:18,980 houses and then they are Fabless. Right. And intel, 563 00:35:18,980 --> 00:35:22,580 which is interesting because they do both. Right? Yeah, yeah. 564 00:35:22,580 --> 00:35:26,420 Intel like as I was saying that intel doesn't. Yeah, they do both. Yeah. 565 00:35:26,500 --> 00:35:30,020 Right. And then it could be a great strategy. Could work 566 00:35:30,260 --> 00:35:33,780 or. Well, depends on many things. Right then anyways, 567 00:35:34,180 --> 00:35:37,780 so TSMC is like the, as I said before, this 568 00:35:37,780 --> 00:35:41,380 industry, I don't know if it's good or bad but it's a winner takes 569 00:35:41,380 --> 00:35:45,060 all market. Right. So TNC is definitely 570 00:35:45,060 --> 00:35:48,460 the winner for a lot of different 571 00:35:48,460 --> 00:35:52,260 reasons. I think for the leadership, self 572 00:35:52,260 --> 00:35:55,700 and technical team for the whole supply chain ecosystem. The 573 00:35:55,700 --> 00:35:58,620 gravity, all the years, the hard work they've put in. 574 00:35:59,900 --> 00:36:03,620 So it's a position where I don't think anyone 575 00:36:03,620 --> 00:36:05,340 can seriously challenge them 576 00:36:07,340 --> 00:36:11,050 in a meaningful way in the next whatever 577 00:36:11,050 --> 00:36:14,690 years. So they're very critical. And then the good 578 00:36:14,690 --> 00:36:17,930 thing interesting about them, they're the agnostic of design houses, 579 00:36:18,410 --> 00:36:22,250 right. So they have great relationship with Nvidia for sure and I'm sure with 580 00:36:22,250 --> 00:36:25,290 them, with everybody, right. It's their job to 581 00:36:26,010 --> 00:36:29,290 produce those chips and then it's 582 00:36:29,370 --> 00:36:32,970 interesting enough it's aligned with mine. Silicon Data. Because 583 00:36:32,970 --> 00:36:36,340 I'm agnostic of chips, right. So 584 00:36:36,740 --> 00:36:40,540 obviously I want to create products that's most important to the 585 00:36:40,540 --> 00:36:44,380 ecosystem. So right now people care a few chips and 586 00:36:44,380 --> 00:36:48,140 those chips happen to be from one design houses. But let's say 587 00:36:48,140 --> 00:36:51,780 if another design house start picking up a lot of momentum. For me, it's 588 00:36:51,780 --> 00:36:54,580 like, how can I help everybody in ecosystem 589 00:36:55,780 --> 00:36:59,580 compare, contrast hashing, right? Use them benchmarking, normalize 590 00:36:59,580 --> 00:37:03,100 it in a meaningful way. So it's my job to work with all the design 591 00:37:03,100 --> 00:37:06,700 houses. It's their job to produce chips that can be usable for 592 00:37:06,700 --> 00:37:10,340 defunding the houses too. So we're very aligned in that sense. And 593 00:37:10,500 --> 00:37:14,180 anything they do, right? So think about, they are 594 00:37:14,260 --> 00:37:17,860 future looking because they're not thinking about next year or next quarter. They think 595 00:37:17,860 --> 00:37:21,620 about 20 years, 10 years. It takes them five, six 596 00:37:21,620 --> 00:37:25,340 years to build a fab, right? And then they need a fab to 597 00:37:25,340 --> 00:37:29,180 be utilized. And they have a threshold, right? If you're 598 00:37:29,180 --> 00:37:32,980 building a fabric and that's not utilized by year eight, 599 00:37:32,980 --> 00:37:36,660 they plan right now by year a year 10, they are 600 00:37:36,660 --> 00:37:40,380 losing a lot of money. A lot like billions of dollars, 601 00:37:40,380 --> 00:37:44,100 right? Like can you make sure the fab will be utilized, the demand 602 00:37:44,340 --> 00:37:47,540 will be there by year 10. Forecasting from today. 603 00:37:47,860 --> 00:37:51,700 It's very, very, very hard job to do. And it's not 604 00:37:51,700 --> 00:37:55,140 like it's not like a new reim, you know, 605 00:37:55,220 --> 00:37:59,020 like what are minings and all things that you can hedge it, right? 606 00:37:59,020 --> 00:38:02,580 Like there's a way to hatch the future curve. But like it's not like they 607 00:38:02,580 --> 00:38:06,260 can forecast, forecast and do a swap on that because 608 00:38:06,260 --> 00:38:10,060 the market is so concentrated and then very 609 00:38:10,060 --> 00:38:13,500 binary and a huge size. Who's taking the other side? 610 00:38:13,660 --> 00:38:17,500 I don't know. It's very hard over the concentrate to 611 00:38:17,500 --> 00:38:21,180 do so for them is to get clarity supply demand curve in 10 612 00:38:21,180 --> 00:38:24,800 years. I mean they do also edge computing chips as well, not just data 613 00:38:24,800 --> 00:38:28,640 center chips. Right? But how do they think through that? I think that's 614 00:38:28,640 --> 00:38:32,240 really challenging. I think will be really challenging for me 615 00:38:32,240 --> 00:38:36,080 for sure. I'm sure they have way smarter people there to think through those problems. 616 00:38:36,560 --> 00:38:39,360 But yeah, it's an interesting problem to have. 617 00:38:40,640 --> 00:38:44,360 That's why TSMC and I, for example, they sell to 618 00:38:44,360 --> 00:38:47,360 their clients who are in the vds of the world. So they have that kind 619 00:38:47,360 --> 00:38:50,610 of transparency. But what they don't have, which 620 00:38:51,010 --> 00:38:54,770 may be a different indicator for the supply demand curve in 621 00:38:54,770 --> 00:38:56,930 10 years is end users 622 00:38:58,690 --> 00:39:02,530 pricing volatility. Right? And then you know, okay, so if 623 00:39:03,010 --> 00:39:06,810 every single chip, every single chip I produced, right, Data center 624 00:39:06,810 --> 00:39:10,610 quality chips, one dying price, right. 625 00:39:12,050 --> 00:39:15,850 Is the indicator for supply demand shifting. Maybe it 626 00:39:15,850 --> 00:39:19,620 is Maybe it's not right. At least you have some, some data points which your 627 00:39:19,620 --> 00:39:23,340 immediate sales and revenues which is T0 628 00:39:23,340 --> 00:39:27,100 won't give you because then a few degrees removed from 629 00:39:27,100 --> 00:39:30,660 end user experiences you give Nvidia and Nvidia packages it to 630 00:39:30,660 --> 00:39:33,820 AWS and GCP and end users and you and me. Right. 631 00:39:34,700 --> 00:39:37,340 So that's something that for them to think through as well. 632 00:39:38,060 --> 00:39:41,700 Interesting. One of the stories I heard and I 633 00:39:41,700 --> 00:39:45,350 wonder if it's true, was that part of the 634 00:39:45,350 --> 00:39:49,150 reason why there was part of the reason 635 00:39:49,550 --> 00:39:53,110 why Nvidia was able to really capitalize on this. There's a lot of 636 00:39:53,110 --> 00:39:56,550 reasons, but one of them was the fact that in the 637 00:39:56,550 --> 00:40:00,270 crypto craze, the run up to get chips for that Nvidia 638 00:40:00,270 --> 00:40:03,630 had purchased. Now what you said makes a lot more sense now. Nvidia had purchased 639 00:40:03,630 --> 00:40:07,310 the. They basically purchased a certain amount of capacity at TSMC 640 00:40:08,030 --> 00:40:11,830 for like three to four years, something like that. And then that happened to 641 00:40:11,830 --> 00:40:15,490 coincide with the AI boom. Is that, is that true? And 642 00:40:15,570 --> 00:40:18,210 that. I guess that's a market too, right? Like you know, like hey, 643 00:40:19,890 --> 00:40:23,650 so I wasn't. I know so 7 so I'm not following all ASICS 644 00:40:23,810 --> 00:40:27,450 so they have a specific for. For the, for. For the mining 645 00:40:27,450 --> 00:40:30,530 chips. That could be true. So I think 646 00:40:30,930 --> 00:40:34,730 not because I'm straight, I mean and a girl can dream. I'm 647 00:40:34,730 --> 00:40:38,130 strapped to be like, you know, to really 648 00:40:38,290 --> 00:40:41,980 help the industry and then be, you know, like 649 00:40:42,220 --> 00:40:45,980 the company the team hopefully can propel the industry 650 00:40:45,980 --> 00:40:49,260 move forward. Right. I'm strive to point zero over percent people 651 00:40:50,220 --> 00:40:54,020 and then competency is very important. Obviously execution, your 652 00:40:54,020 --> 00:40:57,580 hard work is important. Not a big piece is you have to be 653 00:40:57,660 --> 00:41:01,340 really, really lucky. That is also everyone's control. 654 00:41:01,340 --> 00:41:04,980 And then Nvidia puts so much time effort into everything they do. You can argue 655 00:41:04,980 --> 00:41:08,560 they were really great company even before the AI boom and 656 00:41:08,560 --> 00:41:12,360 everything. But the lock piece and how do you control that? How do 657 00:41:12,360 --> 00:41:15,640 you. How do you know quota gonna be like the piece 658 00:41:15,880 --> 00:41:19,600 that's needed? Right. Well, some. 659 00:41:19,600 --> 00:41:23,280 Someone said that, you know, Jensen Wang is like the epitome of, 660 00:41:23,280 --> 00:41:26,280 you know, the better you, the harder you work, the more luck you have. 661 00:41:27,160 --> 00:41:30,840 True. Like there's a lot to that and I know it's 662 00:41:30,840 --> 00:41:34,560 complicated but like I'm just, I just. It's interesting how the crypto kind 663 00:41:34,560 --> 00:41:38,030 of boom and bust really kind of also 664 00:41:38,030 --> 00:41:41,550 propel us into the AI. Not, not all by 665 00:41:41,550 --> 00:41:45,030 itself but it definitely I think gave. There was some momentum where 666 00:41:46,710 --> 00:41:50,550 no momentum was expected, if that makes sense. Right. Yeah, I agree, 667 00:41:50,550 --> 00:41:53,670 I agree Timing is so interesting, but 668 00:41:54,150 --> 00:41:57,150 we just have to two point like the heart of your world. You have to 669 00:41:57,150 --> 00:42:00,950 do everything you can with the environment. Right? That's 670 00:42:00,950 --> 00:42:03,960 cool. That's cool. All data. So we'll see happens what 671 00:42:05,320 --> 00:42:08,480 I mean. That'S the importance of data. Right. Like, you know, people don't realize that. 672 00:42:08,480 --> 00:42:12,280 And I go calling back to Bloomberg. So I'm referring to Michael Bloomberg, 673 00:42:12,360 --> 00:42:15,480 former mayor of New York. But before he was mayor he 674 00:42:15,960 --> 00:42:19,240 basically started a company called Bloomberg. And 675 00:42:19,640 --> 00:42:23,480 he was not the only factor but like 676 00:42:23,480 --> 00:42:27,160 a big part of, you know, people getting into, you know, his 677 00:42:27,240 --> 00:42:30,830 philosophy. As I understood it, if there's a good, if there's a good biography book 678 00:42:30,830 --> 00:42:34,630 on him, I totally would want to listen to it. But basically getting 679 00:42:34,630 --> 00:42:38,230 the traders access to data gave them an advantage. Right. And it was 680 00:42:38,230 --> 00:42:41,750 really, he was really early on in the idea of that data is 681 00:42:42,150 --> 00:42:45,710 not just something that's created as a byproduct of 682 00:42:45,710 --> 00:42:49,230 transactions, but can actually be, you know, monetized 683 00:42:49,230 --> 00:42:51,750 and arguably weaponized. Right. Like so. 684 00:42:52,870 --> 00:42:56,640 And you know, Bloomberg terminals 685 00:42:56,640 --> 00:43:00,240 before, you know, it was interesting because he basically sold these custom terminals so you'd 686 00:43:00,240 --> 00:43:04,040 not to rely on like local ID who were still struggling with like, you 687 00:43:04,040 --> 00:43:07,720 know, just keeping the network up and running, you know, these separate 688 00:43:07,720 --> 00:43:11,400 devices that became status symbols. And ultimately he, that's become like 689 00:43:11,400 --> 00:43:15,120 this media empire that, you know, I can watch Bloomberg on my 690 00:43:15,120 --> 00:43:18,600 tv, I can listen to it, you know, whether it's a satellite radio or the 691 00:43:18,600 --> 00:43:22,110 app or you know, FM or AM radio 692 00:43:22,110 --> 00:43:25,750 stations. You know, I think it's in San Francisco, New York and 693 00:43:25,750 --> 00:43:29,270 D.C. they have a big office in D.C. they always have an 694 00:43:29,270 --> 00:43:32,910 interesting show called Political Capital. I think that plays 695 00:43:33,870 --> 00:43:37,230 at 5pm every day. I listen to it because it's kind of the 696 00:43:37,630 --> 00:43:40,830 policy side of finance and kind of what's going on in the world around. 697 00:43:41,390 --> 00:43:45,150 And AI has come up a lot digital sovereignty. So it's interesting 698 00:43:45,150 --> 00:43:48,780 how all of these worlds, I like your thoughts on this, 699 00:43:48,780 --> 00:43:52,420 right. The worlds of finance, the worlds of tech and the worlds of policy, 700 00:43:52,580 --> 00:43:56,420 politics and dare I say war. Right. They're all kind of like 701 00:43:56,420 --> 00:44:00,220 crashing together in this giant thing. And 702 00:44:00,220 --> 00:44:02,260 it's kind of cool, kind of scary. 703 00:44:04,500 --> 00:44:07,860 I think it can be. I mean, sometimes I'm scared I was like, 704 00:44:08,180 --> 00:44:10,900 you know, because you see a few things, it's like, whoa. 705 00:44:12,180 --> 00:44:15,860 There's a lot I feel like for people born post Covid, 706 00:44:16,020 --> 00:44:19,460 not born, but grew up post Covid, I would call Jen the second 707 00:44:19,540 --> 00:44:23,380 Gen Z Gen Alpha. Yes. I think Gen 708 00:44:23,380 --> 00:44:27,180 Z's apparently now like I'm all confused. But for 709 00:44:27,180 --> 00:44:30,820 them it's like, of course they should. They should. My AI should be my 710 00:44:30,820 --> 00:44:34,180 boyfriend, girlfriend. Right. Like whatever. And then for me it's like, 711 00:44:34,500 --> 00:44:38,020 this is not comfortable at all. Weird. 712 00:44:38,020 --> 00:44:41,700 Yeah, yeah, yeah. For me it's not. I have no idea what's going on. 713 00:44:41,700 --> 00:44:44,740 Like, I just so creeped out by this. But for lot of people it's like, 714 00:44:44,740 --> 00:44:47,620 of course you do that. Of course you tell AI all your secrets. 715 00:44:48,740 --> 00:44:52,420 Of course they can. My phone can record my conversation. Of course 716 00:44:52,420 --> 00:44:55,860 you can train, you know, your AI model. My 717 00:44:56,020 --> 00:44:59,620 model use my all my Gmail content information. 718 00:45:00,100 --> 00:45:03,460 All edge computing. I have my own AI model. Of course you can wear, 719 00:45:04,020 --> 00:45:07,780 you know, glasses and then record everything you and me talk 720 00:45:07,780 --> 00:45:11,310 about. And how secure is everything 721 00:45:11,630 --> 00:45:15,390 right now? Right. 722 00:45:15,390 --> 00:45:19,150 The hardware level encryption 723 00:45:19,150 --> 00:45:22,990 is only available on a very specific few chips. 724 00:45:23,310 --> 00:45:26,270 TPU can do that. You rely on software encryption. 725 00:45:28,510 --> 00:45:32,310 No, it's true. And software encryption that is vulnerable to a quantum 726 00:45:32,310 --> 00:45:36,080 attack which is not that far away. We are not the 727 00:45:36,080 --> 00:45:39,800 software and use cases moving so quickly. The hardware hasn't been able to cut 728 00:45:39,800 --> 00:45:43,600 up. And it's expensive to do hardware encryption. It takes 729 00:45:43,600 --> 00:45:47,080 longer and it's more expensive. That's why sometimes the hyperscaler charging 730 00:45:47,160 --> 00:45:50,680 higher premium for that reason. Right. Are you willing to spend a 731 00:45:50,680 --> 00:45:54,480 token and time and effort to do so? Some use cases, you can argue. 732 00:45:54,480 --> 00:45:58,320 Yes, yes, absolutely. No edge computing 733 00:45:58,320 --> 00:46:01,400 chips can do that kind of hardware level encryption. 734 00:46:02,520 --> 00:46:05,480 And it's happening like now. Right, Right. 735 00:46:06,360 --> 00:46:10,200 I was talking to a startup called Quantum Knight. Nate claimed to have a solution 736 00:46:10,200 --> 00:46:13,520 that is a low, low compute kind of post 737 00:46:13,520 --> 00:46:17,200 quantum ready thing. So I can send you their 738 00:46:17,200 --> 00:46:20,960 link and information. Yeah, we, we track quantum 739 00:46:20,960 --> 00:46:24,680 computing prices as well. Very different than GPU pricing and like, you know, like 740 00:46:24,840 --> 00:46:28,600 a thousand per second per minute pricing versus hourly. Right. This is like 741 00:46:28,720 --> 00:46:32,240 different cycles you run. And then GPU become like error correction component to the whole 742 00:46:32,240 --> 00:46:35,920 thing. But for us it's like, okay, so 743 00:46:36,240 --> 00:46:39,880 computers compute now, GPU and tpu, whatever, pu. And then 744 00:46:39,880 --> 00:46:43,680 it becomes like quantum. How we think through that? I don't 745 00:46:43,680 --> 00:46:47,280 know. My brain just like, you know. Yeah, I know. At some point it just 746 00:46:47,280 --> 00:46:50,400 becomes like. I'm not smart enough 747 00:46:51,200 --> 00:46:55,040 right now to, to. To. To figure that out. I tell you, like I go 748 00:46:55,040 --> 00:46:58,710 through like quantum stuff and like I always joke with Andy, like I'd be like 749 00:46:58,710 --> 00:47:02,390 15 minutes, I get a migraine, which is basically like my brain's version of 750 00:47:02,390 --> 00:47:06,150 blue screening. And like, just like, okay, I can stop. I can get 751 00:47:06,150 --> 00:47:09,150 to about. I can get to about 45 minutes now, which is, you know, an 752 00:47:09,150 --> 00:47:12,149 improvement. But this is actually a good book. 753 00:47:12,950 --> 00:47:16,470 And he was actually a guest recently on the Quantum Computing podcast. 754 00:47:17,270 --> 00:47:21,030 It's a thick book. It's a thick book. But I'll tell you this. 755 00:47:21,030 --> 00:47:24,560 The, the, the, the first three chapters, introducing the concepts 756 00:47:24,640 --> 00:47:28,280 are probably the single best introduction to the 757 00:47:28,280 --> 00:47:31,800 concept I have ever read. Yeah, I will send you the link. Yeah, 758 00:47:31,800 --> 00:47:33,680 yeah. Dancing with Cubits. 759 00:47:35,520 --> 00:47:38,880 Really interesting book. Super nice author too. He's a, he's a trip. 760 00:47:40,000 --> 00:47:43,640 But it, it. No, 761 00:47:43,640 --> 00:47:46,520 you're right. Like, these are. The thing that really worries me is I kind of 762 00:47:46,520 --> 00:47:49,840 think about this like we built our entire economy and we're, we're 763 00:47:50,080 --> 00:47:53,920 on a house of sand. Can we start on this? That's 764 00:47:53,920 --> 00:47:57,680 another thing. We'll have to have you back on the show for 765 00:47:57,680 --> 00:48:01,120 a second one. But like other countries 766 00:48:01,280 --> 00:48:04,800 where they lay off hundreds and thousands of people, not. Not just by American 767 00:48:04,880 --> 00:48:05,680 companies. Right? 768 00:48:08,720 --> 00:48:11,640 Yeah. Don't even get me s on that. Well, like, and like, you know, we're 769 00:48:11,640 --> 00:48:14,930 all based on. And, and the other thing, the elephant in the room, right, is 770 00:48:14,930 --> 00:48:18,530 the fact that TC the, the T in 771 00:48:18,530 --> 00:48:22,330 TSMC stands for Taiwan. Right. Kind of. 772 00:48:23,610 --> 00:48:26,250 I know, I know it's very dangerous to talk about this, but, but like. It'S 773 00:48:26,250 --> 00:48:29,650 kind of like, shoot. So I won't say much, but I'll just say it's 774 00:48:29,650 --> 00:48:33,090 contested real estate. How about that? Right. That's a pretty safe way to say it. 775 00:48:33,090 --> 00:48:35,530 Right? It's contested. Right. And you know, 776 00:48:36,730 --> 00:48:40,470 the entire world effectively revolves around the kind of modern 777 00:48:40,470 --> 00:48:43,710 civilization revolves around the manufacturing that happens there. And 778 00:48:44,030 --> 00:48:47,870 God forbid, like, you know, whether it's man made or a tsunami or a bad 779 00:48:47,870 --> 00:48:51,590 earthquake, like, I mean, our world, I mean, we, we get sent back 780 00:48:51,590 --> 00:48:55,070 to the 1700s pretty quickly. You know, 1700 is not, 781 00:48:55,630 --> 00:48:59,430 you know, there are still people, they're still human beings in the 782 00:48:59,430 --> 00:49:02,470 hundreds. That could be worse than that. That's true. It could be way worse than 783 00:49:02,470 --> 00:49:04,990 that. That is a good point. I was trying to keep it. I was trying 784 00:49:04,990 --> 00:49:08,430 to end it on a positive. And I know you're traveling there 785 00:49:08,510 --> 00:49:12,230 like no humans. Well, no, I mean, like, 786 00:49:12,230 --> 00:49:15,590 I mean, there's a lot of ways that the, you know, this apocalypse could go, 787 00:49:15,590 --> 00:49:18,870 so to speak. Right. It could be, you know, but like, it's a very. And 788 00:49:18,870 --> 00:49:22,550 like, just from an infrastructure point of view and supply chain point of view, like, 789 00:49:22,550 --> 00:49:25,430 you know, we, we. We've really championed 790 00:49:25,830 --> 00:49:29,510 globalism and kind of all of these extended supply 791 00:49:29,510 --> 00:49:33,350 chains for, you know, there were reasons there's always reasons, but like 792 00:49:33,350 --> 00:49:37,040 at the cost of resilience. Right, right. That's kind of scary. 793 00:49:37,440 --> 00:49:40,960 I assume you've read Taleb, right? The like anti fragile. 794 00:49:41,760 --> 00:49:45,480 I'm so sorry. No, that's fine. That's fine. But I really appreciate you taking the 795 00:49:45,480 --> 00:49:49,280 time. Where can folks find out more about you? Silicon 796 00:49:49,280 --> 00:49:53,040 Data.com Silicon Data.com awesome. And we'd love to have you back on the show. 797 00:49:53,040 --> 00:49:56,320 And you can tell us what these conferences were like. The. The 798 00:49:56,320 --> 00:49:59,360 ts. Let's see how much I can understand 799 00:49:59,920 --> 00:50:02,970 first. Right, right, right, right, right. That wasn't a good question. 800 00:50:03,530 --> 00:50:06,330 That's why you got to be like the kids today and record all your conversations 801 00:50:06,330 --> 00:50:10,170 so you can talk to the transcript later. All right, 802 00:50:10,170 --> 00:50:13,370 nice seeing you guys. All right, thank you. And we'll let our AI finish the 803 00:50:13,370 --> 00:50:17,210 show. And that wraps up another episode of Data Driven, the podcast 804 00:50:17,210 --> 00:50:20,890 where we ponder the future of AI data and occasionally 805 00:50:20,890 --> 00:50:24,410 the fate of humanity if we don't get GPU pricing under control. 806 00:50:24,970 --> 00:50:28,490 Big thanks to Carmen Lee for joining us and blowing our minds with 807 00:50:28,490 --> 00:50:32,230 compute market mechanics, financial innovation, and just a 808 00:50:32,230 --> 00:50:35,510 touch of economic existentialism. Be sure to check out 809 00:50:35,510 --> 00:50:39,110 silicondata.com to learn more. Just don't try to day trade 810 00:50:39,110 --> 00:50:42,750 H1 hundreds after midnight. If you liked what you heard, 811 00:50:42,910 --> 00:50:46,589 subscribe, leave a review, or send us compute credits. 812 00:50:46,910 --> 00:50:50,510 Until next time, stay curious, stay caffeinated, 813 00:50:50,510 --> 00:50:54,350 and remember, in a world of exponential AI, transparency 814 00:50:54,510 --> 00:50:55,950 might just be the killer app.