1 00:00:00,040 --> 00:00:03,240 OK, picture this. You have successfully gathered 2 00:00:03,240 --> 00:00:06,080 the smartest minds in the history of the universe into one 3 00:00:06,080 --> 00:00:08,320 single room. You have Einstein, you have 4 00:00:08,320 --> 00:00:09,960 Curie, you have Turing, you have Hawking. 5 00:00:10,520 --> 00:00:13,680 Millions of them. Actually, in the context of the 6 00:00:13,680 --> 00:00:16,120 current tech landscape, these are your GPU's. 7 00:00:16,400 --> 00:00:19,400 They are the silicon brains that are ready to solve cancer, 8 00:00:20,080 --> 00:00:24,600 crack, fusion energy, and, I don't know, write the next Great 9 00:00:24,600 --> 00:00:26,440 American novel all at the same time. 10 00:00:26,440 --> 00:00:30,000 A room full of potential. Right, but there is a catch. 11 00:00:30,000 --> 00:00:32,080 There is always a catch. And it's a massive 1. 12 00:00:32,200 --> 00:00:35,040 The room is the size of a football stadium, and it is 13 00:00:35,040 --> 00:00:37,360 deafeningly loud. Everyone is shouting. 14 00:00:37,680 --> 00:00:40,600 And because sound travels relatively slowly, by the time 15 00:00:40,600 --> 00:00:43,120 Einstein hears what Turing shouted from the other side of 16 00:00:43,120 --> 00:00:46,640 the room, the man has passed. The calculation is stale. 17 00:00:47,320 --> 00:00:49,320 So instead of a super intelligence, you just have a 18 00:00:49,320 --> 00:00:52,280 very expensive, very loud cafeteria. 19 00:00:52,320 --> 00:00:56,040 That is a terrifyingly accurate metaphor for the single biggest 20 00:00:56,040 --> 00:00:58,640 bottleneck in artificial intelligence right now. 21 00:00:58,880 --> 00:01:01,040 We are, you know, we're collectively obsessed with the 22 00:01:01,040 --> 00:01:04,360 brains, the chips, the NVIDIA H1, hundreds, the black whales. 23 00:01:04,360 --> 00:01:08,120 The shiny objects. Exactly the shiny objects, but 24 00:01:08,120 --> 00:01:11,120 we have completely neglected the ears and the mouths, the parts 25 00:01:11,120 --> 00:01:12,440 that let them all talk to each other. 26 00:01:12,440 --> 00:01:14,120 Exactly. And that is where we are 27 00:01:14,120 --> 00:01:16,760 starting today. We are ignoring the processors 28 00:01:16,760 --> 00:01:19,640 for a moment to look at the hidden nervous system of the AI 29 00:01:19,640 --> 00:01:21,520 boom. We are going to talk about a 30 00:01:21,520 --> 00:01:25,000 piece of hardware that acts as the universal translator in that 31 00:01:25,000 --> 00:01:27,720 room of geniuses. And this little piece of 32 00:01:27,720 --> 00:01:30,120 hardware is called an optical transceiver. 33 00:01:30,200 --> 00:01:33,200 It sounds like a prop from Star Trek, but it is actually the 34 00:01:33,200 --> 00:01:36,720 specific component that might determine whether AI scales up 35 00:01:36,720 --> 00:01:40,400 to the level of AGI or stalls out because of well. 36 00:01:40,560 --> 00:01:42,320 Physics. It really is that critical. 37 00:01:42,320 --> 00:01:45,360 It's the plumbing. And when the plumbing fails, it 38 00:01:45,360 --> 00:01:47,280 doesn't matter how smart the people in the building are. 39 00:01:47,640 --> 00:01:50,200 To guide us through this, we are pulling from a fascinating 40 00:01:50,200 --> 00:01:54,080 TechCrunch article. It's dated February 17th, 2026. 41 00:01:54,440 --> 00:01:59,120 The title is SpaceX Vets Raise $50 Layer Series A for Data 42 00:01:59,120 --> 00:02:01,680 Centrelinks and it's written by Tim Fernholtz. 43 00:02:01,840 --> 00:02:03,080 This story really has everything. 44 00:02:03,080 --> 00:02:05,880 I mean, it has rocket scientists coming back to Earth to solve a 45 00:02:05,880 --> 00:02:09,280 terrestrial problem. It has a $50 million bet from a 46 00:02:09,280 --> 00:02:12,840 major VC firm, Drive Capital. And it has a manufacturing 47 00:02:12,840 --> 00:02:15,800 challenge that touches on everything from global 48 00:02:15,800 --> 00:02:18,600 geopolitics to the future of American industry. 49 00:02:18,920 --> 00:02:21,880 It's a huge story packed into one little startup. 50 00:02:22,280 --> 00:02:25,400 The protagonists here are a startup called Mesh Optical 51 00:02:25,400 --> 00:02:28,760 Technologies, and as the headline suggests, they aren't 52 00:02:28,760 --> 00:02:31,080 your typical Silicon Valley software kids. 53 00:02:31,080 --> 00:02:34,200 No, not at all. They are former SpaceX engineers 54 00:02:34,200 --> 00:02:36,600 who cut their teeth working on Starlink. 55 00:02:36,600 --> 00:02:38,880 Which is crucial context, right? Right. 56 00:02:38,920 --> 00:02:41,600 They spent years figuring out how to make satellites talk to 57 00:02:41,600 --> 00:02:43,960 each other in the vacuum of space using lasers. 58 00:02:44,080 --> 00:02:45,600 I mean, think about that environment. 59 00:02:45,800 --> 00:02:48,880 You can't send a repairman. It has to work, and it has to 60 00:02:48,880 --> 00:02:52,480 work perfectly for years. So they're trying to apply that 61 00:02:52,480 --> 00:02:54,960 same rigor, that same kind of, you know, bulletproof 62 00:02:54,960 --> 00:02:57,000 engineering to the data center. Exactly. 63 00:02:57,280 --> 00:02:59,800 So here is our mission for this deep dive. 64 00:03:00,000 --> 00:03:03,720 We are going to explain what an optical transceiver actually is, 65 00:03:03,880 --> 00:03:07,280 using analogies that even a business executive who has never 66 00:03:07,280 --> 00:03:09,280 touched a soldering iron can understand. 67 00:03:09,360 --> 00:03:12,080 And we'll breakdown why the founders background at SpaceX 68 00:03:12,080 --> 00:03:15,240 makes them uniquely qualified for this specific challenge. 69 00:03:15,280 --> 00:03:19,920 And then we are going to explore the massive, and I mean massive 70 00:03:19,920 --> 00:03:23,800 challenge of bringing lights out automated manufacturing back to 71 00:03:23,800 --> 00:03:26,720 the US. It is a story about light speed 72 00:03:27,200 --> 00:03:30,200 and the sheer difficulty of building physical things in a 73 00:03:30,200 --> 00:03:32,960 digital age. It's atoms versus bits. 74 00:03:33,160 --> 00:03:35,360 I love it. Let's jump right in Part 1, the 75 00:03:35,360 --> 00:03:38,600 tech because optical transceiver sounds complicated. 76 00:03:38,600 --> 00:03:41,200 It sounds intimidating. It does, but the concept is 77 00:03:41,200 --> 00:03:43,920 actually very elegant if we strip away the jargon, I 78 00:03:43,920 --> 00:03:45,880 promise. OK, so let's do that for the 79 00:03:45,880 --> 00:03:48,640 business executive listening, who knows strategy, who knows 80 00:03:48,640 --> 00:03:51,440 panel statements, but maybe skipped electrical engineering 81 00:03:51,440 --> 00:03:54,000 in college. Help us visualize this. 82 00:03:54,000 --> 00:03:56,160 Let's go back to our room of geniuses, right? 83 00:03:56,480 --> 00:03:59,400 The loud cafeteria inside a computer chip. 84 00:03:59,400 --> 00:04:02,960 So inside that geniuses brain information travels as 85 00:04:02,960 --> 00:04:05,560 electricity. OK, it's electrons moving 86 00:04:05,560 --> 00:04:08,760 through tiny copper or traces, and that works great for very, 87 00:04:08,760 --> 00:04:11,800 very short distances, like millimeters inside a chip or 88 00:04:11,800 --> 00:04:13,280 maybe centimeters on a motherboard. 89 00:04:13,600 --> 00:04:16,200 It's super fast and efficient for that microscopic scale. 90 00:04:16,240 --> 00:04:18,320 OK, so electrons are the local dialect. 91 00:04:18,320 --> 00:04:20,720 They speak electricity. That's the internal monologue of 92 00:04:20,720 --> 00:04:22,640 the genius. That is a perfect way to put it. 93 00:04:22,720 --> 00:04:24,640 It's their internal thought process. 94 00:04:25,200 --> 00:04:28,720 But electricity has a physical limitation as soon as you try to 95 00:04:28,720 --> 00:04:31,760 push it fast over to long distance, and in modern 96 00:04:31,760 --> 00:04:35,600 computing, long distance can just be across a server rack 97 00:04:35,800 --> 00:04:37,840 barely a few feet. Just a few feet. 98 00:04:37,840 --> 00:04:40,600 Yeah, we're not talking miles. As soon as you do that, it meets 99 00:04:40,600 --> 00:04:43,440 resistance. The copper wire literally fights 100 00:04:43,440 --> 00:04:45,360 back. It fights back it. 101 00:04:45,760 --> 00:04:48,480 Gets hot, the signal degrades. It becomes noisy. 102 00:04:48,480 --> 00:04:50,600 It's like trying to shout across that football field. 103 00:04:51,120 --> 00:04:53,840 You lose your voice and the person on the other end only 104 00:04:53,840 --> 00:04:56,400 hears mumbling. The message gets garbled. 105 00:04:56,600 --> 00:04:59,160 So the local dialect doesn't travel well. 106 00:04:59,160 --> 00:05:01,520 You can't shout electricity across the room and expect 107 00:05:01,520 --> 00:05:03,120 anyone to understand you. Precisely. 108 00:05:03,800 --> 00:05:06,640 You lose both speed and clarity, but there is a universal 109 00:05:06,640 --> 00:05:10,120 language that travels perfectly over long distances with almost 110 00:05:10,120 --> 00:05:11,560 no loss. Light. 111 00:05:11,840 --> 00:05:14,880 Light photons. This stuff of fiber optics. 112 00:05:15,160 --> 00:05:17,440 Right. Light is incredibly fast. 113 00:05:17,440 --> 00:05:19,160 I mean, it's the speed limit of the universe. 114 00:05:19,360 --> 00:05:22,240 It generates almost no heat compared to electricity during 115 00:05:22,240 --> 00:05:26,120 transmission, and it can carry a massive amount of data without 116 00:05:26,120 --> 00:05:29,160 losing integrity. You can send a beam of light 117 00:05:29,440 --> 00:05:32,840 thousands of miles under the ocean and the message arrives 118 00:05:32,840 --> 00:05:35,240 perfectly clear. OK, so now we have a problem. 119 00:05:35,240 --> 00:05:38,320 The chips speak electricity, but the best way to connect them is 120 00:05:38,320 --> 00:05:39,880 with the light. They're speaking two different 121 00:05:39,880 --> 00:05:40,840 languages. Exactly. 122 00:05:40,840 --> 00:05:43,080 So you need a translator. And that's the transceiver. 123 00:05:43,080 --> 00:05:45,920 That is a transceiver. It is a tiny, incredibly 124 00:05:45,920 --> 00:05:48,600 sophisticated little box that sits at the edge of the 125 00:05:48,600 --> 00:05:50,760 computer. It takes the electrical signal 126 00:05:50,760 --> 00:05:54,200 from the GPU, the local dialect translates it into a pulse of 127 00:05:54,200 --> 00:05:57,960 laser light, zips it across the fiber optic cable to another 128 00:05:57,960 --> 00:06:01,160 computer, and then another transceiver at the other end 129 00:06:01,320 --> 00:06:04,880 catches that light pulse and translates it back into 130 00:06:04,880 --> 00:06:07,200 electricity so the receiving chip can understand it. 131 00:06:07,200 --> 00:06:09,200 So it's literally a translator box. 132 00:06:09,200 --> 00:06:11,960 Electricity in, light out, and on the other side, light in, 133 00:06:11,960 --> 00:06:14,400 electricity out. That is exactly what it does, 134 00:06:14,720 --> 00:06:17,800 and it has to do this billions, even trillions of times per 135 00:06:17,800 --> 00:06:19,760 second with near perfect accuracy. 136 00:06:19,920 --> 00:06:21,560 It's an amazing piece of engineering. 137 00:06:21,920 --> 00:06:24,640 For a long time we didn't really need this for short distances, 138 00:06:24,640 --> 00:06:26,760 right? We used something else, right? 139 00:06:27,120 --> 00:06:31,600 For a long time, we relied on radio frequencies, or RF, to 140 00:06:31,600 --> 00:06:35,320 move data over copper cables for these shorter distances. 141 00:06:35,760 --> 00:06:37,440 Think about your Ethernet cable at home. 142 00:06:38,280 --> 00:06:42,080 But Travis Brashears, the CEO of Mesh Optical, put it really well 143 00:06:42,080 --> 00:06:45,000 in the source material he said. I'm quoting here. 144 00:06:45,320 --> 00:06:48,640 The world has primarily focused on radio frequencies for a long 145 00:06:48,640 --> 00:06:50,640 time. We want to be at the precipice 146 00:06:50,640 --> 00:06:53,000 of transition from RF to photonics. 147 00:06:53,000 --> 00:06:55,840 Precipice of transition? That sounds dramatic. 148 00:06:56,040 --> 00:06:58,080 It is, though. He's talking about a fundamental 149 00:06:58,080 --> 00:07:00,240 shift in the physics of how we build computers, right? 150 00:07:00,320 --> 00:07:02,160 We're hitting the wall with copper and RF. 151 00:07:02,400 --> 00:07:05,520 You just can't push electrons any faster or denser without the 152 00:07:06,000 --> 00:07:08,400 the components. So the future is moving from 153 00:07:08,400 --> 00:07:11,160 pushing electrons through copper to shooting photons through 154 00:07:11,160 --> 00:07:14,400 glass. Yes, and not just for the long 155 00:07:14,400 --> 00:07:17,400 haul cables under the ocean, which we've done for decades, 156 00:07:17,680 --> 00:07:20,960 but for the short, incredibly fast connections between 157 00:07:20,960 --> 00:07:23,880 computers in the same room. The connections that form the 158 00:07:23,880 --> 00:07:26,760 brain of an AI. So if I'm that executive, I'm 159 00:07:26,760 --> 00:07:31,040 thinking, OK, I understand the box, it translates, but why does 160 00:07:31,040 --> 00:07:33,760 this matter now? Why is this suddenly a crisis? 161 00:07:33,920 --> 00:07:36,680 Because the sheer volume of conversation has exploded. 162 00:07:36,880 --> 00:07:39,360 It's not one genius talking to another anymore. 163 00:07:39,600 --> 00:07:42,160 It's a million geniuses all trying to talk to each other at 164 00:07:42,160 --> 00:07:44,480 the exact same time. The vision is interconnecting 165 00:07:44,480 --> 00:07:46,760 everything, right? I think Brashears mentions that. 166 00:07:46,760 --> 00:07:49,240 He does. He explicitly says we want to 167 00:07:49,240 --> 00:07:51,960 interconnect everything, and not just computers. 168 00:07:51,960 --> 00:07:54,840 But that's where we're starting. This is step one in a much 169 00:07:54,840 --> 00:07:57,000 larger vision. OK, so that's the what? 170 00:07:57,160 --> 00:08:00,080 It's a high speed translator for the internet's internal traffic. 171 00:08:00,280 --> 00:08:03,360 Now let's talk about the why. Because usually plumbing is 172 00:08:03,360 --> 00:08:04,800 boring. You only notice it when it 173 00:08:04,800 --> 00:08:07,840 breaks. Why is Thrive Capital dumping 174 00:08:07,840 --> 00:08:11,560 $50 million into this right now? Because the plumbing is about to 175 00:08:11,560 --> 00:08:13,840 burst. In fact, it's already leaking. 176 00:08:13,880 --> 00:08:17,280 That sounds ominous. It's the scale of AI. 177 00:08:17,720 --> 00:08:19,680 We have to wrap our heads around the numbers here because they 178 00:08:19,680 --> 00:08:22,640 are just staggering. When we talk about scaling a web 179 00:08:22,640 --> 00:08:25,600 app, maybe you add a few more servers to handle more users. 180 00:08:25,640 --> 00:08:28,080 Easy enough. Yeah, When you scale an AI 181 00:08:28,080 --> 00:08:31,560 model, you are building a supercomputer that acts as one 182 00:08:31,560 --> 00:08:34,400 cohesive brain. All the parts have to work in 183 00:08:34,400 --> 00:08:37,600 perfect concert. The source material specifically 184 00:08:37,600 --> 00:08:42,440 mentions a 1,000,000 GPU cluster, which just a few years 185 00:08:42,440 --> 00:08:44,120 ago would have sounded like science fiction. 186 00:08:44,120 --> 00:08:47,320 Completely. And here is the key stat from 187 00:08:47,320 --> 00:08:51,560 the article that blew my mind. Breshear says someone will brag 188 00:08:51,560 --> 00:08:55,880 about a million GPU cluster. You have to multiply by 4 to 5 189 00:08:56,000 --> 00:08:58,200 for the number of transceivers in that cluster. 190 00:08:58,360 --> 00:09:01,160 Wait, hold on, let's do that math. 1,000,000 GPU's? 191 00:09:01,160 --> 00:09:05,400 You're telling me you need 4 to 5 million optical transceivers? 192 00:09:05,400 --> 00:09:09,720 Correct. 4 to 5 million of these high tech translator boxes for 193 00:09:09,720 --> 00:09:12,440 one single installation. Why? 194 00:09:12,600 --> 00:09:15,600 That seems counterintuitive? Why does one chip need 5 195 00:09:15,600 --> 00:09:17,640 translators? Because of how deep learning 196 00:09:17,640 --> 00:09:21,520 works, these models, like the ones powering the chat bots we 197 00:09:21,520 --> 00:09:24,120 all use, aren't trained on a single computer. 198 00:09:24,280 --> 00:09:27,120 They're too big. They are trained by slicing the 199 00:09:27,120 --> 00:09:30,120 brain into thousands of pieces and spreading it across 200 00:09:30,120 --> 00:09:32,640 thousands of chips. But those chips have to 201 00:09:32,640 --> 00:09:35,440 constantly talk to each other. It's a constant, furious 202 00:09:35,440 --> 00:09:37,880 conversation. Hey, I learned this pattern. 203 00:09:37,880 --> 00:09:39,400 What did you see? I updated this weight. 204 00:09:39,400 --> 00:09:41,920 You need to adjust that. So the chatter is constant. 205 00:09:41,920 --> 00:09:45,680 It's not just one chip sending another a file to download, it's 206 00:09:46,280 --> 00:09:50,040 active real time collaboration. It's relentless. 207 00:09:50,280 --> 00:09:52,000 It's a many to many conversation. 208 00:09:52,360 --> 00:09:55,960 Every GPU needs a high speed link to many other GPU's. 209 00:09:56,160 --> 00:09:59,040 If those links are slow, if the translators can't keep up, the 210 00:09:59,040 --> 00:10:01,280 whole system stalls. You have a million Ferraris 211 00:10:01,280 --> 00:10:03,440 stuck in traffic. And at that point, you spent 212 00:10:03,560 --> 00:10:06,280 billions on GPU's that are just sitting there waiting. 213 00:10:06,280 --> 00:10:08,360 Exactly. You're just burning electricity. 214 00:10:08,360 --> 00:10:09,720 That's why the market is exploding. 215 00:10:09,720 --> 00:10:12,600 Companies are desperate for faster, more efficient plumbing. 216 00:10:12,840 --> 00:10:15,600 And it is a massive market. The article brings up a really 217 00:10:15,600 --> 00:10:18,560 concrete example. It mentions A supplier called 218 00:10:18,760 --> 00:10:21,400 AOI. Yes, Applied Optoelectronics, an 219 00:10:21,400 --> 00:10:23,960 established US supplier. A big player. 220 00:10:24,040 --> 00:10:28,560 And they won a contract just last year worth $4 billion with 221 00:10:28,600 --> 00:10:32,920 AB, with AB $4 billion. And that was just for AWS data 222 00:10:32,920 --> 00:10:37,120 centers. 1 customer, one contract. 223 00:10:37,320 --> 00:10:39,640 Wow, that gives you a sense of the pot of gold at the end of 224 00:10:39,640 --> 00:10:43,000 this rainbow. We are talking about a 225 00:10:43,000 --> 00:10:46,000 foundational component for the entire AI industry. 226 00:10:46,040 --> 00:10:49,240 OK, so the money is there, the demand is there, everyone needs 227 00:10:49,240 --> 00:10:52,000 this. But what is Mesh Optical doing 228 00:10:52,000 --> 00:10:54,800 differently? Why not just buy from AOI or the 229 00:10:54,800 --> 00:10:56,960 other existing guys? There must be a catch. 230 00:10:57,080 --> 00:10:59,280 There is. The existing solutions are 231 00:10:59,280 --> 00:11:01,800 hitting a wall in efficiency. This is where we get into the 232 00:11:01,800 --> 00:11:03,720 weeds of the engineering a bit, but stay with me, it's 233 00:11:03,720 --> 00:11:04,840 important. Let's do it. 234 00:11:05,080 --> 00:11:08,600 Mesh is core innovation. Their secret sauce is a new 235 00:11:08,600 --> 00:11:12,120 design that removes a specific, commonly used component from the 236 00:11:12,120 --> 00:11:14,640 transceiver. The source calls it a power 237 00:11:14,640 --> 00:11:16,320 hungry component. Right, they're referring to 238 00:11:16,320 --> 00:11:19,640 something called a DSP, a digital signal processor or 239 00:11:19,680 --> 00:11:21,520 similar components called retimers. 240 00:11:22,000 --> 00:11:24,320 Think of these as little amplifiers and error checkers 241 00:11:24,320 --> 00:11:26,040 for the signal. They cleaned up the mumbling I 242 00:11:26,040 --> 00:11:28,000 mentioned earlier. OK, that sounds useful. 243 00:11:28,000 --> 00:11:31,120 Why would you want to remove it? Because it uses a ton of power 244 00:11:31,360 --> 00:11:34,680 and generates a ton of heat, it's a necessary evil in older 245 00:11:34,680 --> 00:11:37,720 designs. Mesh believes their architecture 246 00:11:37,720 --> 00:11:40,880 is so clean and efficient that they don't need it, or at least 247 00:11:40,880 --> 00:11:43,160 they can use a much simpler, lower power version. 248 00:11:43,280 --> 00:11:45,200 So by stripping that out, what's the impact? 249 00:11:45,280 --> 00:11:48,400 They claim they can reduce the power usage of a GPU cluster by 250 00:11:48,400 --> 00:11:50,760 3% to 5%. You know, I can hear the 251 00:11:50,760 --> 00:11:52,800 listener saying 3%. That's it. 252 00:11:52,800 --> 00:11:55,240 I tip more than that for coffee. It sounds tiny. 253 00:11:55,240 --> 00:11:58,240 I know, right? In the consumer world, 3% is a 254 00:11:58,240 --> 00:12:00,000 rounding error. You wouldn't even notice it on 255 00:12:00,000 --> 00:12:01,880 your phone's battery life. Exactly. 256 00:12:01,920 --> 00:12:04,640 But in the world of hyperscalers, Amazon, Google, 257 00:12:04,640 --> 00:12:08,360 Microsoft, 3% is a fortune. An absolute fortune. 258 00:12:08,360 --> 00:12:11,400 Because these data centers use as much power as a a small city. 259 00:12:11,640 --> 00:12:13,440 A medium sized city in some cases. 260 00:12:13,440 --> 00:12:15,200 We were talking about gigawatts of electricity. 261 00:12:15,400 --> 00:12:20,000 A 3% of 5% reduction across a massive cluster saves 10s of 262 00:12:20,000 --> 00:12:21,960 millions of dollars a year in electricity bills. 263 00:12:22,200 --> 00:12:24,680 But honestly, the money is almost the secondary benefit. 264 00:12:24,680 --> 00:12:26,200 What's the primary one? Heat. 265 00:12:26,320 --> 00:12:29,040 Heat is the enemy. Heat is the ultimate enemy of 266 00:12:29,040 --> 00:12:32,000 computing. Every single Watt of power you 267 00:12:32,000 --> 00:12:35,040 use generates heat. If you generate heat, you have 268 00:12:35,040 --> 00:12:38,240 to run massive air conditioning systems to cool it down, which 269 00:12:38,240 --> 00:12:42,120 takes more power. It's this vicious cycle I see. 270 00:12:42,200 --> 00:12:45,760 If mesh can reduce the heat generated by the plumbing by 5%, 271 00:12:46,000 --> 00:12:49,240 that means you can pack the GPU's closer together, or you 272 00:12:49,240 --> 00:12:51,520 can run them harder and faster without the melting. 273 00:12:51,840 --> 00:12:54,440 It's a win for density and a win for performance. 274 00:12:54,440 --> 00:12:56,600 Philip Clark, the partner at Thrive Capital. 275 00:12:56,600 --> 00:12:58,280 He pointed this out too in the article. 276 00:12:58,320 --> 00:13:00,240 He did. He said mesh is solving the 277 00:13:00,240 --> 00:13:04,160 immediate term need for better interconnects to keep scaling 278 00:13:04,240 --> 00:13:06,440 AI. This isn't a science project. 279 00:13:06,440 --> 00:13:09,040 For 10 years from now. The hyper scalers have Rd. maps, 280 00:13:09,200 --> 00:13:11,120 they know the power and heat limits that are about to hit 281 00:13:11,120 --> 00:13:13,640 next year and they are desperate for solutions. 282 00:13:13,640 --> 00:13:15,600 They need this hardware yesterday, literally. 283 00:13:15,600 --> 00:13:18,840 So we have a massive demand, a bottleneck in physics, and a 284 00:13:18,840 --> 00:13:21,000 solution that saves critical power and heat. 285 00:13:21,680 --> 00:13:24,400 Now let's look at the team, because this isn't just three 286 00:13:24,400 --> 00:13:28,080 grads from a coding boot camp. This is the SpaceX DNA. 287 00:13:28,360 --> 00:13:30,480 This is one of the most compelling parts of the story 288 00:13:30,480 --> 00:13:33,360 for me. The founders Travis Brashears, 289 00:13:33,360 --> 00:13:37,200 the CEO, Chairman Ramos, the President, and Serena Grown 290 00:13:37,200 --> 00:13:40,080 Heberly, the VP of product. They all work together at 291 00:13:40,080 --> 00:13:42,760 SpaceX, not just at the same company, but on the same 292 00:13:42,760 --> 00:13:45,760 projects. Yes, specifically they worked on 293 00:13:45,760 --> 00:13:48,520 optical communications links for Starlink. 294 00:13:48,720 --> 00:13:51,800 Starlink is the satellite Internet constellation, right? 295 00:13:51,800 --> 00:13:54,160 The one with thousands of satellites. 296 00:13:54,160 --> 00:13:56,480 That's the one. Thousands of satellites orbiting 297 00:13:56,480 --> 00:13:58,240 Earth beaming Internet down to us. 298 00:13:58,960 --> 00:14:02,440 But crucially, those satellites also have to talk to each other 299 00:14:02,440 --> 00:14:04,280 in space. How do they do that? 300 00:14:04,440 --> 00:14:06,640 With lasers, that is an optical link. 301 00:14:07,000 --> 00:14:09,840 They are shooting beams of light from one satellite to another 302 00:14:10,040 --> 00:14:13,800 across hundreds of miles of empty space to relay data around 303 00:14:13,800 --> 00:14:15,520 the globe. So they were building laser 304 00:14:15,520 --> 00:14:17,400 plumbing for space. Precisely. 305 00:14:17,800 --> 00:14:19,560 And let's just think about space for a moment. 306 00:14:19,880 --> 00:14:22,600 It is the harshest engineering environment imaginable. 307 00:14:23,000 --> 00:14:26,080 You have extreme temperature swings, boiling hot in the sun, 308 00:14:26,200 --> 00:14:27,960 freezing cold in the earths shadow. 309 00:14:28,200 --> 00:14:31,120 You have constant radiation bombarding the electronics. 310 00:14:31,120 --> 00:14:35,160 And most importantly, you cannot send a repair technician to low 311 00:14:35,160 --> 00:14:37,720 Earth orbit. If the Wi-Fi breaks in space, 312 00:14:38,120 --> 00:14:40,120 nobody can come and reboot the router. 313 00:14:40,120 --> 00:14:41,960 Exact. You can't just jiggle the cable. 314 00:14:42,360 --> 00:14:46,000 So the engineering culture at SpaceX is all about extreme 315 00:14:46,000 --> 00:14:49,080 reliability, vertical integration, and aggressive 316 00:14:49,080 --> 00:14:51,560 scaling. You don't just build 1 perfect 317 00:14:51,560 --> 00:14:54,640 satellite, you build thousands of them cheaply and quickly. 318 00:14:54,880 --> 00:14:58,360 And that mindset is what they're bringing to this new problem. 319 00:14:58,400 --> 00:15:02,520 The article mentions a specific aha moment for them. 320 00:15:02,720 --> 00:15:04,880 It wasn't just hey let's go do a startup. 321 00:15:05,120 --> 00:15:07,520 It came out of a problem they were trying to solve at SpaceX. 322 00:15:07,520 --> 00:15:09,720 It did. This is a classic scratch your 323 00:15:09,720 --> 00:15:12,040 own itch story. They were designing a new 324 00:15:12,040 --> 00:15:16,680 generation of space F satellites that were very compute hungry. 325 00:15:17,080 --> 00:15:20,560 They needed to move a lot of data inside the satellite itself 326 00:15:20,720 --> 00:15:24,200 between different processors. So naturally they did what any 327 00:15:24,200 --> 00:15:26,320 engineer would do. They look at the existing market 328 00:15:26,320 --> 00:15:29,080 for optical transceivers to see if they could just buy the parts 329 00:15:29,080 --> 00:15:31,360 off the shelf, and they saw limitations. 330 00:15:31,800 --> 00:15:34,480 That's the very polite engineering way of saying this 331 00:15:34,480 --> 00:15:36,200 stuff isn't good enough. It's too slow. 332 00:15:36,200 --> 00:15:39,720 It uses too much power and it's not reliable enough for what we 333 00:15:39,720 --> 00:15:41,960 need. I love that it's the ultimate 334 00:15:41,960 --> 00:15:45,280 engineer's move. We checked the store, didn't 335 00:15:45,280 --> 00:15:47,320 like the merchandise, so we decided to build our own 336 00:15:47,320 --> 00:15:49,440 factory. It's classic innovation by 337 00:15:49,440 --> 00:15:51,360 necessity. They couldn't find what they 338 00:15:51,360 --> 00:15:54,200 needed for the rigor of space, so they decided to build it 339 00:15:54,200 --> 00:15:56,000 themselves. And then they had the bigger 340 00:15:56,000 --> 00:15:58,320 realization. Wait a minute, the data centers 341 00:15:58,320 --> 00:16:01,520 down on Earth have the exact same problem, just at a much, 342 00:16:01,520 --> 00:16:04,600 much bigger scale. They are taking that space grade 343 00:16:04,600 --> 00:16:07,720 mindset where failure is not an option and efficiency is 344 00:16:07,720 --> 00:16:09,840 everything, and applying it to the data center floor. 345 00:16:09,960 --> 00:16:13,520 And that SpaceX DNA informs their entire strategy. 346 00:16:13,960 --> 00:16:16,560 It's not just about the design of the product, it's about how 347 00:16:16,560 --> 00:16:19,240 they plan to build it. Which brings us to the biggest 348 00:16:19,240 --> 00:16:22,760 hurdle in this entire story, the real crux of the challenge. 349 00:16:22,800 --> 00:16:24,880 The manufacturing. The manufacturing this. 350 00:16:24,880 --> 00:16:28,480 Is Part 4 the challenge? And frankly, this is the part 351 00:16:28,480 --> 00:16:31,560 that makes me nervous for them, because designing a cool gadget 352 00:16:31,560 --> 00:16:34,600 on a computer is one thing. Building millions of them is a 353 00:16:34,600 --> 00:16:37,120 whole other ball game. And building millions of them in 354 00:16:37,120 --> 00:16:40,160 the United States is a whole other level of difficulty on top 355 00:16:40,160 --> 00:16:41,920 of that. It's like playing the game on 356 00:16:41,920 --> 00:16:43,960 ultra hard mode. Let's look at their goals from 357 00:16:43,960 --> 00:16:45,800 the article. They want to be manufacturing 358 00:16:45,800 --> 00:16:47,720 1000 units per day within the year. 359 00:16:48,120 --> 00:16:51,000 Is an absolute Sprint. And then the real goal is to 360 00:16:51,000 --> 00:16:54,400 qualify for bulk orders in 2027 and 2028. 361 00:16:54,600 --> 00:16:57,880 That means ramping up to 10s of thousands a day to feed the 362 00:16:57,880 --> 00:17:00,040 hyperscalers. And the founders say the main 363 00:17:00,040 --> 00:17:03,200 challenge, the thing that keeps them up at night, is executing 364 00:17:03,280 --> 00:17:06,000 lights out automated manufacturing techniques. 365 00:17:06,280 --> 00:17:08,280 Let's unpack that term. Lights out. 366 00:17:08,480 --> 00:17:10,160 It's a very evocative phrase. It is. 367 00:17:10,160 --> 00:17:12,599 Ideally it means you can literally turn the lights off in 368 00:17:12,599 --> 00:17:15,480 the factory and leave because there are no humans working on 369 00:17:15,480 --> 00:17:17,560 the assembly line. It's just robots building 370 00:17:17,560 --> 00:17:19,440 things, 2047. Exactly. 371 00:17:19,440 --> 00:17:24,119 It implies total automation, high precision, high speed, and 372 00:17:24,280 --> 00:17:29,160 theoretically 0 human error. And here is the brutal reality. 373 00:17:29,560 --> 00:17:32,720 The United States, as an industrial base, has largely 374 00:17:32,720 --> 00:17:35,200 forgotten how to do this for optoelectronics. 375 00:17:35,320 --> 00:17:36,920 The source is pretty blunt about this. 376 00:17:36,920 --> 00:17:39,880 It says this expertise is heavily concentrated in China. 377 00:17:39,960 --> 00:17:44,240 It is the China problem, as it's often called over the last 20-30 378 00:17:44,240 --> 00:17:46,040 years. The entire supply chain, the 379 00:17:46,080 --> 00:17:51,560 engineering talent, the specific institutional know how for mass 380 00:17:51,560 --> 00:17:55,040 producing these tiny complex optical components. 381 00:17:56,080 --> 00:17:58,640 It all moved to Asia and primarily to China. 382 00:17:58,800 --> 00:18:01,640 There is an anecdote in the article that really, really 383 00:18:01,640 --> 00:18:04,400 drives this home for me. It's about a European equipment 384 00:18:04,400 --> 00:18:07,120 supplier. The German form story. 385 00:18:07,400 --> 00:18:09,040 This stuck with me too. It's so telling. 386 00:18:09,040 --> 00:18:11,360 Tell us about that, because it's a perfect illustration of the 387 00:18:11,360 --> 00:18:14,760 problem. O Mesh Otical is in the process 388 00:18:14,760 --> 00:18:18,680 of buying the complex machinery they need to set U their factory 389 00:18:18,680 --> 00:18:21,480 in Los Angeles. They go to a German firm that 390 00:18:21,480 --> 00:18:23,720 makes the best in class manufacturing gear. 391 00:18:24,040 --> 00:18:25,720 The robots that build the transceivers. 392 00:18:26,240 --> 00:18:28,960 The standard intake form for that German company, the one you 393 00:18:28,960 --> 00:18:32,480 have to fill out to become a customer, asks for a Chinese 394 00:18:32,480 --> 00:18:35,360 company registration number. Wow, not just company 395 00:18:35,360 --> 00:18:37,400 registration number, specifically a Chinese one. 396 00:18:37,680 --> 00:18:39,560 Just think about the baked in assumption there. 397 00:18:40,280 --> 00:18:44,960 The form assumes by default that if you are a company buying this 398 00:18:44,960 --> 00:18:48,360 highly specialized equipment, you must be a Chinese company, 399 00:18:48,960 --> 00:18:51,440 because who else in the world would be buying it at scale? 400 00:18:51,520 --> 00:18:55,440 That is a sobering detail. It shows just how entrenched 401 00:18:55,440 --> 00:18:57,760 that global supply chain dominance is. 402 00:18:57,760 --> 00:19:00,840 It's not just a talking point. It's built into the paperwork. 403 00:19:00,840 --> 00:19:03,440 It really is. It's not just that China has the 404 00:19:03,440 --> 00:19:06,440 factories. The entire global ecosystem of 405 00:19:06,440 --> 00:19:09,960 suppliers and experts assumes China is the only place this 406 00:19:09,960 --> 00:19:12,560 kind of manufacturing happens. But Mesh is trying to change 407 00:19:12,560 --> 00:19:13,480 that. And this is where the 408 00:19:13,480 --> 00:19:16,760 geopolitical angle comes roaring in Thrive Capital. 409 00:19:16,760 --> 00:19:19,080 Philip Clark again. He didn't mince words about the 410 00:19:19,080 --> 00:19:20,800 national security aspect of this. 411 00:19:20,800 --> 00:19:23,000 No, he didn't. And just to be clear for our 412 00:19:23,000 --> 00:19:25,680 listeners, this is the perspective of the investor in 413 00:19:25,680 --> 00:19:28,480 the source text he wrote to TechCrunch, and I'm 414 00:19:28,480 --> 00:19:32,080 paraphrasing, but the gist was if AI is the most important 415 00:19:32,080 --> 00:19:35,480 technology of our generation and they believe it is, then having 416 00:19:35,480 --> 00:19:38,480 the critical CapEx parts, the foundational hardware running 417 00:19:38,480 --> 00:19:42,080 through misaligned competitive countries is a massive strategic 418 00:19:42,080 --> 00:19:45,520 problem. Misaligned countries that is VC 419 00:19:45,520 --> 00:19:48,960 speak for. We don't want our entire AI 420 00:19:48,960 --> 00:19:52,160 backbone dependent on China. That's exactly what it means. 421 00:19:52,160 --> 00:19:54,960 It's a supply chain risk. The article notes that trade 422 00:19:54,960 --> 00:19:58,000 restrictions haven't hit this specific market yet, but 423 00:19:58,000 --> 00:19:59,760 everyone sees the writing on the wall. 424 00:20:00,480 --> 00:20:04,240 If geopolitical tensions rise, you don't want your access to 425 00:20:04,240 --> 00:20:06,360 optical transceivers suddenly cut off. 426 00:20:06,920 --> 00:20:10,040 That would strangle your domestic AI progress overnight. 427 00:20:10,120 --> 00:20:13,120 So Mesh sees a strategic advantage in building a supply 428 00:20:13,120 --> 00:20:16,320 chain outside of China. They're positioning themselves 429 00:20:16,320 --> 00:20:18,960 as the secure domestic alternative. 430 00:20:18,960 --> 00:20:20,800 They do. They are trying to get ahead of 431 00:20:20,800 --> 00:20:24,080 the dilemma to offer a solution before it becomes a full blown 432 00:20:24,080 --> 00:20:27,160 crisis for the US tech industry. And they are doing it by Co 433 00:20:27,160 --> 00:20:29,240 locating their design and production. 434 00:20:29,240 --> 00:20:30,560 Right. Their design team is in Los 435 00:20:30,560 --> 00:20:32,080 Angeles. Their production line is being 436 00:20:32,080 --> 00:20:34,000 built in Los Angeles. They're in the same building. 437 00:20:34,000 --> 00:20:37,520 Why does that matter in a world of Zoom and global logistics? 438 00:20:37,520 --> 00:20:40,920 Why not design in LA and manufacture in, I don't know, 439 00:20:40,920 --> 00:20:43,920 Vietnam or Mexico? Because when you are trying to 440 00:20:43,920 --> 00:20:46,960 invent a completely new manufacturing process, this 441 00:20:46,960 --> 00:20:50,040 lights out automation. You need the design engineer 442 00:20:50,040 --> 00:20:52,120 standing next to the robots on the factory floor. 443 00:20:53,200 --> 00:20:57,480 If there is a tiny 3% efficiency gain to be found, you find it by 444 00:20:57,480 --> 00:21:01,160 watching the robot, quinking the code and iterating instantly. 445 00:21:01,560 --> 00:21:04,440 You don't find it by emailing a factory halfway across the world 446 00:21:04,440 --> 00:21:07,160 and waiting two weeks for a new prototype to ship back. 447 00:21:07,680 --> 00:21:09,840 The feedback loop has to be instantaneous. 448 00:21:09,920 --> 00:21:12,840 So it's about the speed of iteration, the speed of learning 449 00:21:12,840 --> 00:21:13,480 it's. Speed. 450 00:21:13,560 --> 00:21:16,080 And they believe ultimately it's about cost. 451 00:21:16,760 --> 00:21:19,320 They're making a bet that if they can crack the code on 452 00:21:19,320 --> 00:21:22,320 automation, they can actually produce these things cheaper in 453 00:21:22,320 --> 00:21:25,400 the US than they could by outsourcing, simply because they 454 00:21:25,400 --> 00:21:28,600 eliminate the shipping, the delays, the tariffs, and the 455 00:21:28,600 --> 00:21:31,160 communication overhead. That is the Holy Grail of re 456 00:21:31,160 --> 00:21:33,520 shoring, isn't it? Bringing manufacturing back not 457 00:21:33,520 --> 00:21:36,320 out of some sense of patriotism, but because it's actually Better 458 00:21:36,320 --> 00:21:38,480 Business it. It the only way it works long 459 00:21:38,480 --> 00:21:40,640 term. Patriotism is great for a press 460 00:21:40,640 --> 00:21:43,080 release, but unit economics rule the world. 461 00:21:43,280 --> 00:21:45,680 I want to pause on the lights out aspect for a second. 462 00:21:45,880 --> 00:21:49,960 We say robots building robots, but what does that actually look 463 00:21:49,960 --> 00:21:52,480 like for something this small and delicate? 464 00:21:52,800 --> 00:21:55,800 That's a great question because we aren't talking about the 465 00:21:55,800 --> 00:21:59,000 giant robotic arms you see welding car frames in a Tesla 466 00:21:59,000 --> 00:22:01,520 factory. These transceivers are small, 467 00:22:01,840 --> 00:22:03,440 maybe the size of a stick of gum. 468 00:22:04,000 --> 00:22:06,320 The components inside are microscopic. 469 00:22:06,480 --> 00:22:08,680 So we're talking about incredible precision. 470 00:22:08,680 --> 00:22:11,920 Extreme precision. Remember, one of the key steps 471 00:22:12,120 --> 00:22:15,600 is to align a tiny laser with a fiber optic cable. 472 00:22:16,440 --> 00:22:19,960 The core of that fiber optic cable is about the width of a 473 00:22:19,960 --> 00:22:22,440 human hair. You have to shoot a beam of 474 00:22:22,440 --> 00:22:25,280 light perfectly into that hair width target, and you have to do 475 00:22:25,280 --> 00:22:28,480 it while everything is hot and vibrating slightly, and then you 476 00:22:28,480 --> 00:22:31,440 have to lock it in place with a microscopic dab of epoxy. 477 00:22:31,600 --> 00:22:33,320 And doing that by hand is slow and. 478 00:22:33,320 --> 00:22:35,000 Difficult. It's incredibly slow, and it's 479 00:22:35,360 --> 00:22:37,840 prone to air. If a human technician has a 480 00:22:37,840 --> 00:22:41,280 shaky hand after their morning coffee, the laser misses the 481 00:22:41,280 --> 00:22:44,760 target and the part is garbage. Automation allows for what's 482 00:22:44,760 --> 00:22:47,400 called active alignment. It's where the robot holds the 483 00:22:47,400 --> 00:22:49,240 laser. A sensor measures the light 484 00:22:49,240 --> 00:22:52,360 coming out the other end of the fiber, and the robot software 485 00:22:52,360 --> 00:22:56,400 uses that feedback to adjust the lasers position by nanometers 486 00:22:56,760 --> 00:22:59,640 until the signal is perfect, then it locks it in. 487 00:22:59,800 --> 00:23:01,520 And it does this in a fraction of a second. 488 00:23:01,720 --> 00:23:05,720 Right, to hit that 1000 units a day target and eventually 10s of 489 00:23:05,720 --> 00:23:09,640 thousands, those robots have to be a blur of motion performing 490 00:23:09,640 --> 00:23:11,840 these microscopic surgeries over and over again. 491 00:23:12,000 --> 00:23:14,720 Florida State. So when they say lights out, 492 00:23:14,920 --> 00:23:17,480 it's not just about saving money on electricity for the room 493 00:23:17,480 --> 00:23:21,360 lights, it's about creating a process that humans physically 494 00:23:21,360 --> 00:23:24,600 cannot perform at the required speed and quality. 495 00:23:24,680 --> 00:23:27,160 Exactly. Humans become the bottom neck in 496 00:23:27,160 --> 00:23:29,880 the manufacturing process. Just like copper is the 497 00:23:29,880 --> 00:23:32,000 bottleneck in the data transmission process. 498 00:23:32,560 --> 00:23:35,720 Mesh is trying to remove the human element from the assembly 499 00:23:35,720 --> 00:23:38,560 line to reach the necessary scale and precision. 500 00:23:38,680 --> 00:23:40,240 It's removing friction everywhere. 501 00:23:40,320 --> 00:23:42,400 Friction in the wire, Friction in the factory. 502 00:23:42,400 --> 00:23:45,120 That is the theme of this whole story, reducing friction to 503 00:23:45,120 --> 00:23:47,160 increase speed. Let's synthesize this. 504 00:23:47,160 --> 00:23:49,920 We are moving into part 5 here, connecting all these dots. 505 00:23:50,080 --> 00:23:52,520 Let's look at the narrative arc we've uncovered, because it's a 506 00:23:52,520 --> 00:23:55,480 really powerful one. We started with a fundamental 507 00:23:55,480 --> 00:23:58,440 bottleneck in physics, the AI geniuses. 508 00:23:58,600 --> 00:24:02,720 The GPU's are so fast that the current plumbing can't keep up. 509 00:24:03,120 --> 00:24:05,880 Right, The copper wires and the old radio frequency tech are 510 00:24:05,880 --> 00:24:08,480 hitting a wall. The conversation in the loud 511 00:24:08,480 --> 00:24:12,600 cafeteria is breaking down. Enter the solution, a team of 512 00:24:12,600 --> 00:24:16,000 engineers from SpaceX who spent years solving this exact 513 00:24:16,000 --> 00:24:18,640 problem, just in the much harsher environment of the 514 00:24:18,640 --> 00:24:21,840 vacuum of space. They bring a mindset of extreme 515 00:24:21,840 --> 00:24:26,120 reliability and a clever new design that saves critical power 516 00:24:26,120 --> 00:24:29,360 and reduces HEAT, which is the number one enemy of performance. 517 00:24:29,360 --> 00:24:32,720 But they run head first into a wall, a geopolitical and 518 00:24:32,720 --> 00:24:35,960 industrial wall. The manufacturing capability to 519 00:24:35,960 --> 00:24:38,560 build this thing at scale doesn't really exist in the West 520 00:24:38,560 --> 00:24:40,040 anymore. So they have to build that too. 521 00:24:40,040 --> 00:24:42,040 They have to reinvent the factory floor while they 522 00:24:42,040 --> 00:24:43,920 reinvent the component that's built on it. 523 00:24:44,000 --> 00:24:47,040 It's a double challenge. It's product innovation plus 524 00:24:47,040 --> 00:24:49,480 process innovation happening at the same time. 525 00:24:49,640 --> 00:24:52,360 Which is incredibly risky. Most startups would only tackle 526 00:24:52,360 --> 00:24:55,600 one of those, but if it works, the payoff is absolutely 527 00:24:55,600 --> 00:24:57,720 enormous. You don't just own a product, 528 00:24:57,880 --> 00:25:00,080 you own the entire means of production. 529 00:25:00,480 --> 00:25:03,120 Let's go back to that quote from Travis Brochures one more time. 530 00:25:03,120 --> 00:25:04,560 We want to interconnect everything. 531 00:25:04,720 --> 00:25:08,560 That is the bigger picture here. We are fixated on AI right now 532 00:25:08,560 --> 00:25:12,320 because that is the immediate term demand that's driving the 533 00:25:12,320 --> 00:25:15,560 $50 million investment. But think about what happens 534 00:25:15,560 --> 00:25:18,840 when this kind of optical communication becomes cheap, 535 00:25:19,320 --> 00:25:23,880 power efficient, and ubiquitous. It's not just about faster chat 536 00:25:23,880 --> 00:25:27,480 bots or better image generators. No, it's about a fundamental 537 00:25:27,480 --> 00:25:29,520 shift in how all machines communicate. 538 00:25:29,720 --> 00:25:32,600 We are moving from a world where computers talk to each other via 539 00:25:32,600 --> 00:25:35,000 electricity to a world where they talk to each other via. 540 00:25:35,000 --> 00:25:37,320 Light. Light is the new standard for 541 00:25:37,320 --> 00:25:40,200 internal communication, not just for long distance. 542 00:25:40,440 --> 00:25:43,080 Exactly. Imagine self driving cars, 543 00:25:43,360 --> 00:25:46,600 factory robots, airplanes, even your home appliances. 544 00:25:46,960 --> 00:25:49,200 If they can all start communicating with each other 545 00:25:49,360 --> 00:25:52,720 with the speed and efficiency of light, the latency, the delay 546 00:25:53,160 --> 00:25:56,040 all but disappears. The power consumption drops 547 00:25:56,040 --> 00:25:58,080 dramatically. The bandwidth becomes 548 00:25:58,080 --> 00:26:00,360 effectively infinite for most applications. 549 00:26:00,360 --> 00:26:03,440 It's like upgrading the nervous system of the entire planet from 550 00:26:03,440 --> 00:26:06,400 copper wires to a fiber optics. That is a great way to put it. 551 00:26:06,400 --> 00:26:08,560 It's a planetary scale nervous system upgrade. 552 00:26:09,000 --> 00:26:12,000 So for our listeners, specifically the learners out 553 00:26:12,000 --> 00:26:15,380 there who love to stay ahead of the curve, what is the key take 554 00:26:15,380 --> 00:26:18,200 away from this deep dive? The take away is this. 555 00:26:18,720 --> 00:26:21,640 Don't just look at the smarts and don't just focus on the 556 00:26:21,640 --> 00:26:24,720 software or the LMS. When you read about the next big 557 00:26:24,880 --> 00:26:27,920 AI breakthrough or the next super chip from NVIDIA, ask 558 00:26:27,920 --> 00:26:31,360 yourself the boring question how is the data moving? 559 00:26:31,680 --> 00:26:33,320 Look for the plumbing. Look for the plumbing. 560 00:26:33,440 --> 00:26:35,760 Look at the optical links. That is where the physical 561 00:26:35,760 --> 00:26:38,920 constraints are, and in technology the biggest 562 00:26:38,920 --> 00:26:41,600 constraints are always where the biggest opportunities hide. 563 00:26:41,840 --> 00:26:43,920 That's where the next billion dollar companies are being 564 00:26:43,920 --> 00:26:45,400 built. That is fascinating. 565 00:26:45,400 --> 00:26:48,600 It's a reminder that even in our increasingly digital world, 566 00:26:49,000 --> 00:26:51,400 atoms still matter. You still have to build the 567 00:26:51,400 --> 00:26:54,280 physical thing that moves the photons from A to B. 568 00:26:54,320 --> 00:26:55,840 Absolutely. The cloud is not a cloud. 569 00:26:55,840 --> 00:26:59,000 The cloud is made of metal, silicon and glass running in a 570 00:26:59,000 --> 00:27:01,880 giant power hungry building. I want to challenge one part of 571 00:27:01,880 --> 00:27:04,320 this before we go. We talked a lot about the SpaceX 572 00:27:04,320 --> 00:27:08,000 DNA being a huge asset, but is there a potential downside to 573 00:27:08,000 --> 00:27:09,800 that culture? That's a fair question. 574 00:27:09,800 --> 00:27:13,040 What's the risk? SpaceX is famous for its move 575 00:27:13,040 --> 00:27:16,600 fast and break things ethos. They literally have a highlight 576 00:27:16,600 --> 00:27:17,960 reel of their rockets blowing up. 577 00:27:18,040 --> 00:27:20,240 Rapid unscheduled disassembly. Right. 578 00:27:20,440 --> 00:27:22,960 They test a failure, they build a rocket, launch it, watch it 579 00:27:22,960 --> 00:27:25,880 explode, learn why, and then build another one better in a 580 00:27:25,880 --> 00:27:27,840 few weeks. Can you apply that culture to a 581 00:27:27,840 --> 00:27:30,000 supply chain for Amazon or Google? 582 00:27:30,240 --> 00:27:33,600 If you ship a million bad transceivers to AWS, you don't 583 00:27:33,600 --> 00:27:35,880 get a second chance. Yo, you get sued and your 584 00:27:35,880 --> 00:27:38,000 company dies. Yeah, you're absolutely right. 585 00:27:38,080 --> 00:27:41,920 A rocket blowing up on a remote test stand is valuable data. 586 00:27:42,160 --> 00:27:45,760 A data center in Ohio going dark because your transceivers failed 587 00:27:46,160 --> 00:27:48,520 is an economic catastrophe for your customer. 588 00:27:48,640 --> 00:27:50,400 So that is the tension they have to manage. 589 00:27:50,400 --> 00:27:52,960 They have to keep the innovation, speed and the 590 00:27:52,960 --> 00:27:57,480 agility of SpaceX, but combine it with the Six Sigma 0 defect 591 00:27:57,480 --> 00:28:01,120 reliability of a traditional boring industrial manufacturer. 592 00:28:01,360 --> 00:28:05,280 And that transition from scrappy startup mode to trusted 593 00:28:05,280 --> 00:28:08,760 industrial supplier mode is where so many promising hardware 594 00:28:08,760 --> 00:28:11,440 companies fail. It's known as the Valley of 595 00:28:11,440 --> 00:28:13,680 Death. Scaling from the first working 596 00:28:13,680 --> 00:28:17,040 prototype to the first million units off the assembly line, 597 00:28:17,680 --> 00:28:19,480 that is where the wheels usually fall off. 598 00:28:19,520 --> 00:28:23,280 It is So the $50 million from Thrive Capital, that's not just 599 00:28:23,280 --> 00:28:26,680 for R&D, that's basically the fuel to build the bridge to get 600 00:28:26,680 --> 00:28:28,560 them across that valley before the money runs out. 601 00:28:28,640 --> 00:28:30,960 High stakes. Extremely high stakes, but as we 602 00:28:30,960 --> 00:28:33,560 said, the rewards on the other side are astronomical. 603 00:28:33,640 --> 00:28:35,680 OK, before we sign off, I want to leave the listener with a 604 00:28:35,680 --> 00:28:37,880 final thought, a bit of a provocation to chew on. 605 00:28:37,920 --> 00:28:40,000 Go for it. We talked about this lights out 606 00:28:40,000 --> 00:28:42,400 manufacturing. The idea of a fully automated 607 00:28:42,400 --> 00:28:46,360 factory running in the dark mesh is betting their whole company 608 00:28:46,360 --> 00:28:48,320 that they can pull this off in the USA. 609 00:28:48,320 --> 00:28:52,760 Very bold bet. If they succeed, if these three 610 00:28:52,760 --> 00:28:55,640 engineers from SpaceX can actually figure out how to mass 611 00:28:55,640 --> 00:28:58,960 roduce these incredibly complex recision electronics in Los 612 00:28:58,960 --> 00:29:02,880 Angeles and do it cheaper than the established layers in China, 613 00:29:04,240 --> 00:29:06,000 what does that mean? Does that prove that the whole 614 00:29:06,000 --> 00:29:09,000 narrative about US manufacturing being dead is wrong? 615 00:29:09,080 --> 00:29:12,960 Exactly, Is this the start of a renaissance, a blueprint for how 616 00:29:12,960 --> 00:29:14,920 to bring high tech manufacturing back? 617 00:29:15,360 --> 00:29:16,960 Or is it just a special exception? 618 00:29:16,960 --> 00:29:18,960 A1 off? That's only possible because the 619 00:29:18,960 --> 00:29:22,440 product is so strategically valuable for AI that you can 620 00:29:22,440 --> 00:29:24,840 justify the massive initial investment. 621 00:29:25,040 --> 00:29:26,920 That is the billion dollar question, isn't it? 622 00:29:27,080 --> 00:29:30,120 If they crack the code, maybe we see lights out factories for 623 00:29:30,120 --> 00:29:32,760 other complex electronics. Maybe some consumer electronics 624 00:29:32,760 --> 00:29:34,800 manufacturing comes back. But if they fail. 625 00:29:34,800 --> 00:29:37,480 That it just reinforces the dominant idea that complex 626 00:29:37,480 --> 00:29:40,760 hardware belongs in Asia and software and design belong in 627 00:29:40,760 --> 00:29:42,200 the US. Which is a very dangerous 628 00:29:42,200 --> 00:29:44,680 dichotomy if you care about supply chain security and 629 00:29:44,680 --> 00:29:46,800 national resilience. It absolutely is. 630 00:29:46,920 --> 00:29:50,000 And consider this as a final, final thought. 631 00:29:50,920 --> 00:29:54,280 We are moving toward a world where the speed of light isn't 632 00:29:54,280 --> 00:29:57,800 just a constant in physics textbooks, it is rapidly 633 00:29:57,800 --> 00:30:01,280 becoming the standard speed for internal machine thought. 634 00:30:01,280 --> 00:30:03,880 When the speed of thought literally equals the speed of 635 00:30:03,880 --> 00:30:07,200 light, things are going to get very interesting very quickly. 636 00:30:07,320 --> 00:30:10,080 Indeed they are. And on that note, we are going 637 00:30:10,080 --> 00:30:12,320 to wrap it up for today. It's been a pleasure digging 638 00:30:12,320 --> 00:30:14,560 into this one. Thanks for diving deep with us. 639 00:30:14,560 --> 00:30:17,000 This has been the deep dive. We'll see you on the next one. 640 00:30:17,200 --> 00:30:17,880 Stay curious.