1 00:00:02,800 --> 00:00:17,880 Speaker 1: Zone Media. I'm d Zetron and this is better offline. 2 00:00:19,880 --> 00:00:21,720 Speaker 1: That's right, folks, We're back for the second part of 3 00:00:21,760 --> 00:00:23,799 Speaker 1: our series in the dot Com Bubble and why I 4 00:00:23,800 --> 00:00:27,320 Speaker 1: believe the AI Bubble could be much, much worse. While 5 00:00:27,360 --> 00:00:29,440 Speaker 1: the dot com bubble was a mixture of dodgy venture 6 00:00:29,480 --> 00:00:32,080 Speaker 1: capital deals and websites that could never turn a profit, 7 00:00:32,120 --> 00:00:35,400 Speaker 1: combined with a global mania around the interconnectivity of high 8 00:00:35,400 --> 00:00:38,680 Speaker 1: speed Internet companies, the AI bubble is one company selling 9 00:00:38,800 --> 00:00:42,320 Speaker 1: expensive aigpus, a bunch of companies building data sentence to 10 00:00:42,320 --> 00:00:44,360 Speaker 1: put them in, and a bunch of companies building shit 11 00:00:44,400 --> 00:00:46,640 Speaker 1: that runs on GPUs that only loses money and that 12 00:00:46,680 --> 00:00:49,720 Speaker 1: customers kind of fucking hate. I should also note that 13 00:00:49,760 --> 00:00:52,000 Speaker 1: a very important part of the story is venture capital's 14 00:00:52,080 --> 00:00:54,640 Speaker 1: lack of returns in the last few years, something I 15 00:00:54,680 --> 00:00:57,960 Speaker 1: covered in the Hitter financial Crisis last week, specifically in 16 00:00:58,000 --> 00:01:02,360 Speaker 1: parts two and three. In simple terms, AI startups now 17 00:01:02,400 --> 00:01:04,880 Speaker 1: make up more than half a venture investment, and I 18 00:01:04,920 --> 00:01:07,040 Speaker 1: believe that most of these stops will die because of 19 00:01:07,040 --> 00:01:10,760 Speaker 1: their horrible margins, no part to profitability, and products that 20 00:01:10,760 --> 00:01:13,080 Speaker 1: people really don't want to pay for its scale, and 21 00:01:13,120 --> 00:01:15,679 Speaker 1: when they die, they will leave venture capitalists stranded with 22 00:01:15,760 --> 00:01:18,240 Speaker 1: tons of dead equity at a time when they already 23 00:01:18,319 --> 00:01:21,600 Speaker 1: have trouble generating returns and thus raising money from their 24 00:01:21,640 --> 00:01:25,839 Speaker 1: limited partners. The result, I worry, will be gruesome. Venture 25 00:01:25,840 --> 00:01:28,319 Speaker 1: capitalists make their money through the fees they generate, which 26 00:01:28,360 --> 00:01:30,240 Speaker 1: are based on the value of their investments and the 27 00:01:30,280 --> 00:01:34,280 Speaker 1: returns they give their investors, which don't seem to be happening. 28 00:01:34,680 --> 00:01:36,920 Speaker 1: What do you think happens when they can't generate any 29 00:01:36,959 --> 00:01:40,360 Speaker 1: returns and their investments aren't worth anything. The answer is simple, 30 00:01:40,640 --> 00:01:42,800 Speaker 1: they won't have any way of raising more capital as 31 00:01:42,840 --> 00:01:45,600 Speaker 1: their limited partners won't fucking trust them. And to be clear, 32 00:01:45,880 --> 00:01:48,920 Speaker 1: these arswapes have cocked it up many years at a time. 33 00:01:49,240 --> 00:01:53,480 Speaker 1: Look at crypto, look at NFTs, look at ar VR, metaverse, 34 00:01:53,600 --> 00:01:56,600 Speaker 1: all of that, and all of this ridiculousness has happened 35 00:01:56,600 --> 00:02:00,320 Speaker 1: because of the ridiculous, ridiculous myths of AI, like just 36 00:02:00,320 --> 00:02:03,480 Speaker 1: the insatiable demand for AI compute and the made up 37 00:02:03,520 --> 00:02:06,640 Speaker 1: decline in the price of intelligence, which, by the way, 38 00:02:07,640 --> 00:02:10,520 Speaker 1: that last one I've talked about it, I swear to god, 39 00:02:10,560 --> 00:02:13,120 Speaker 1: I mentioned it in the Guardia, mentioned it everywhere. It 40 00:02:13,160 --> 00:02:15,760 Speaker 1: may be cheaper to pay for the tokens, but you 41 00:02:15,960 --> 00:02:18,960 Speaker 1: use more of them, so it's more expensive in the aggregate. 42 00:02:19,200 --> 00:02:21,400 Speaker 1: On top of that, it doesn't mean that the price 43 00:02:21,440 --> 00:02:25,120 Speaker 1: of intelligence for the model providers is going down. Jesus Christ, 44 00:02:25,160 --> 00:02:30,200 Speaker 1: I'm so tired of making this points. But in both cases, 45 00:02:30,240 --> 00:02:33,800 Speaker 1: these assumptions are convincing investors that it's time to invest 46 00:02:33,800 --> 00:02:36,239 Speaker 1: in data centers that only lose money because you assume 47 00:02:36,320 --> 00:02:38,680 Speaker 1: that the demand will be there, or in AI startups 48 00:02:38,680 --> 00:02:41,400 Speaker 1: to only lose money because you'll assume they magically stop 49 00:02:41,480 --> 00:02:44,720 Speaker 1: losing money somehow. And it turns out we do actually 50 00:02:44,760 --> 00:02:48,200 Speaker 1: have a historical comparison. The mania of the dot com 51 00:02:48,200 --> 00:02:50,480 Speaker 1: bubble was based on a misunderstanding of the scale of 52 00:02:50,480 --> 00:02:53,640 Speaker 1: the Internet at the time, rather than its actual potential. 53 00:02:54,160 --> 00:02:57,239 Speaker 1: Hundreds of billions of dollars were invested based on flimsy logic. 54 00:02:57,800 --> 00:03:02,200 Speaker 1: To quote researcher Justin Kohler. This continental rewiring was also 55 00:03:02,320 --> 00:03:05,520 Speaker 1: justified by another powerful myth that Internet traffic was doubling 56 00:03:05,520 --> 00:03:10,320 Speaker 1: every ninety days. This claim spread through analyst reports, burning 57 00:03:10,360 --> 00:03:15,200 Speaker 1: course and investor presentations like a particularly virulent meme. If true, 58 00:03:15,480 --> 00:03:18,280 Speaker 1: it meant the demand was growing exponentially, far out pacing 59 00:03:18,280 --> 00:03:21,320 Speaker 1: any conceivable supply, and that every new trench of fiber 60 00:03:21,360 --> 00:03:23,960 Speaker 1: would soon pay for itself many times over. And I 61 00:03:24,040 --> 00:03:27,200 Speaker 1: pause here to go, oh god, they're doing it again. 62 00:03:27,960 --> 00:03:31,640 Speaker 1: Back to the quote. But the mathematics were fiction. Network 63 00:03:31,680 --> 00:03:35,000 Speaker 1: researchers like Andrew Adlisco at AT and T, looking at 64 00:03:35,040 --> 00:03:38,720 Speaker 1: actual traffic data, found that US backbone traffic was doubling 65 00:03:38,840 --> 00:03:42,440 Speaker 1: roughly once a year, rapid growth, certainly, but nowhere near 66 00:03:42,520 --> 00:03:46,680 Speaker 1: the purported ninety day cycle. Meanwhile, advances in fiber technology 67 00:03:46,720 --> 00:03:51,240 Speaker 1: were making each strand exponentially more powerful. Dense wavelength division 68 00:03:51,320 --> 00:03:54,760 Speaker 1: multiplexing allowed dozens of signals to travel simultaneously down the 69 00:03:54,800 --> 00:03:58,560 Speaker 1: same line a different wavelengths of light, like multiple conversations 70 00:03:58,560 --> 00:04:02,760 Speaker 1: happening in different colours. While demand doubled annually, supply expanded 71 00:04:02,800 --> 00:04:06,880 Speaker 1: tenfold or more. Carriers buried the discrepancy under layers of 72 00:04:06,920 --> 00:04:10,640 Speaker 1: creative accounting that would have impressed medieval alchemists. First of all, 73 00:04:10,760 --> 00:04:13,120 Speaker 1: justin if you hear this, I fucking love that. That 74 00:04:13,240 --> 00:04:16,400 Speaker 1: was very fun. Second of all, to quote twin peaks, 75 00:04:17,240 --> 00:04:21,920 Speaker 1: it's happening again. Mania had taken hold based on very 76 00:04:21,920 --> 00:04:25,760 Speaker 1: flimsy logic. Economically speaking, this meant the telecoms companies, serve 77 00:04:25,760 --> 00:04:30,440 Speaker 1: a hardware companies, ISPs, construction firms, optical cable providers, wireless 78 00:04:30,440 --> 00:04:33,560 Speaker 1: technology companies, and basically anybody related to the business of 79 00:04:33,600 --> 00:04:36,279 Speaker 1: providing Internet access in any way saw a massive influx 80 00:04:36,320 --> 00:04:39,080 Speaker 1: of business to build capacity that didn't need to be 81 00:04:39,080 --> 00:04:42,200 Speaker 1: built yet. If you're in the business of selling services 82 00:04:42,200 --> 00:04:44,120 Speaker 1: to get people online, you were high on the hug, 83 00:04:44,480 --> 00:04:49,040 Speaker 1: you know, kind of like selling high bandwidth rem Similarly, 84 00:04:49,080 --> 00:04:51,200 Speaker 1: one could get a startup funded if you had a website, 85 00:04:51,279 --> 00:04:53,400 Speaker 1: or even take in public. One could raise debt to 86 00:04:53,400 --> 00:04:56,039 Speaker 1: build a nascent ISSP or a fiber network. The money 87 00:04:56,080 --> 00:04:59,840 Speaker 1: was flowing because people people weren't really really being thoughtful 88 00:05:00,000 --> 00:05:02,400 Speaker 1: about it. And as far as the dot com part 89 00:05:02,520 --> 00:05:05,800 Speaker 1: of the dot com bubble, the unsustainable websites, the problem 90 00:05:05,839 --> 00:05:08,680 Speaker 1: wasn't so much the industry but the businesses themselves, which 91 00:05:08,720 --> 00:05:11,479 Speaker 1: were hyped and dumped onto the public markets with little 92 00:05:11,480 --> 00:05:14,080 Speaker 1: regard for their long term health. The problem here is 93 00:05:14,120 --> 00:05:17,000 Speaker 1: relatively simple. These were bad companies that people ignored the 94 00:05:17,040 --> 00:05:19,080 Speaker 1: issues with because of and they quote the power of 95 00:05:19,120 --> 00:05:22,480 Speaker 1: the Internet and how it would somehow save them, which 96 00:05:22,640 --> 00:05:26,599 Speaker 1: which it didn't obviously had they been kept private. And 97 00:05:26,640 --> 00:05:28,680 Speaker 1: died in the dark. I don't think these companies would 98 00:05:28,720 --> 00:05:30,800 Speaker 1: have had the same reputation. I don't think we give 99 00:05:30,839 --> 00:05:34,559 Speaker 1: a fuck about pets dot com. I also don't think 100 00:05:34,600 --> 00:05:37,680 Speaker 1: they were really in accurate comparison to anything happening today. 101 00:05:38,839 --> 00:05:42,320 Speaker 1: While the valuations were ridiculous, the Globe's market cap was 102 00:05:42,320 --> 00:05:45,039 Speaker 1: at one point eight hundred and forty million dollars, the 103 00:05:45,080 --> 00:05:48,880 Speaker 1: scale of destruction caused by dot com startups was significantly smaller, 104 00:05:49,120 --> 00:05:52,440 Speaker 1: even in today's money. The economics were bad, but not 105 00:05:52,480 --> 00:05:55,400 Speaker 1: anywhere near as bad. For the first nine months of 106 00:05:55,480 --> 00:05:57,760 Speaker 1: nineteen ninety eight, the Globe made two point seven million 107 00:05:57,800 --> 00:06:00,680 Speaker 1: dollars in revenue and lost eleven point five million dollars, 108 00:06:00,760 --> 00:06:03,800 Speaker 1: largely due to trying to move into multiple different business 109 00:06:03,800 --> 00:06:06,320 Speaker 1: lines at once like voice over IP. And to be clear, 110 00:06:06,400 --> 00:06:09,839 Speaker 1: this company made money selling ads and also buying random companies. 111 00:06:09,880 --> 00:06:13,080 Speaker 1: Buying random companies was the thing that happened during the 112 00:06:13,120 --> 00:06:16,640 Speaker 1: dot com boom. Everybody fucking loved buying companies. You just 113 00:06:16,680 --> 00:06:21,920 Speaker 1: bought random. There was I think like Excite, bought at Home, 114 00:06:22,120 --> 00:06:25,960 Speaker 1: like there was the AT and T sort, the Aol 115 00:06:26,040 --> 00:06:31,160 Speaker 1: time Warner merger. So many stupid mergers, so little time. Nevertheless, 116 00:06:31,160 --> 00:06:33,599 Speaker 1: the economics of this shit show were quite complex. You 117 00:06:33,640 --> 00:06:36,000 Speaker 1: had companies raising money to do any website they could 118 00:06:36,000 --> 00:06:39,120 Speaker 1: think of, companies raising money to lay fiber, companies raising 119 00:06:39,160 --> 00:06:42,400 Speaker 1: money to found ISPs, all of which had multifaested layers 120 00:06:42,400 --> 00:06:47,000 Speaker 1: of physical and digital infrastructure that were quite ugh unbuilt. 121 00:06:47,040 --> 00:06:50,440 Speaker 1: I think is the term. It's tempting yet incorrect to 122 00:06:50,440 --> 00:06:54,000 Speaker 1: say the thing about AI. The similarity everybody points to 123 00:06:54,080 --> 00:06:55,919 Speaker 1: is that people doubted the Internet at the time, and 124 00:06:55,960 --> 00:06:59,000 Speaker 1: people really need to remember their fucking history. In two 125 00:06:59,000 --> 00:07:01,160 Speaker 1: thousand and only fifty two percent of Americans who were 126 00:07:01,240 --> 00:07:03,480 Speaker 1: using the Internet. By two thousand and three, the number 127 00:07:03,520 --> 00:07:06,640 Speaker 1: had only increased the sixty one percent per the World Bank. 128 00:07:06,680 --> 00:07:09,240 Speaker 1: In two thousand and five, only sixteen percent of the 129 00:07:09,240 --> 00:07:11,600 Speaker 1: world used the Internet, and in twenty twenty four the 130 00:07:11,760 --> 00:07:14,640 Speaker 1: number had increased to seventy one percent. Yet the real 131 00:07:14,680 --> 00:07:17,680 Speaker 1: difference is the access to high speed Internet. When the 132 00:07:17,680 --> 00:07:20,320 Speaker 1: Internet was connected via a fifth to six K modem, 133 00:07:20,440 --> 00:07:22,920 Speaker 1: access was at times charged by the minute, and even 134 00:07:22,920 --> 00:07:25,520 Speaker 1: if it was unlimited, it was always much much slower. 135 00:07:26,000 --> 00:07:28,120 Speaker 1: While we're used to connecting its speeds that make using 136 00:07:28,200 --> 00:07:31,320 Speaker 1: web based app near indistinguishable from using one on a computer. 137 00:07:31,680 --> 00:07:33,760 Speaker 1: Back in two thousand, two thousand and one or two 138 00:07:33,800 --> 00:07:36,360 Speaker 1: thousand and two, the average US Internet speed was at 139 00:07:36,360 --> 00:07:39,400 Speaker 1: best four hundred kilobits per second, or roughly fifty killer 140 00:07:39,480 --> 00:07:42,560 Speaker 1: bytes per second, compared to the average US Internet speed 141 00:07:42,560 --> 00:07:45,040 Speaker 1: today of other two hundred megabits per second or twenty 142 00:07:45,040 --> 00:07:49,520 Speaker 1: five megabytes per second. In simpler terms than the younger 143 00:07:49,600 --> 00:07:52,640 Speaker 1: members of the audience won't understand this, A website took 144 00:07:52,640 --> 00:07:54,800 Speaker 1: time to load in a way that feels almost impossible 145 00:07:54,800 --> 00:07:56,960 Speaker 1: to conceive if you didn't experience it at the time, 146 00:07:57,480 --> 00:07:59,720 Speaker 1: you had to make a commitment to God all website. 147 00:07:59,760 --> 00:08:02,880 Speaker 1: It was like you've browsed multiple tabs and fuck around 148 00:08:02,880 --> 00:08:05,560 Speaker 1: the different windows. You sat there and you waited a 149 00:08:05,600 --> 00:08:08,080 Speaker 1: little bit. Sometimes it came up quick and then another. 150 00:08:08,240 --> 00:08:11,360 Speaker 1: In fact, websites like Google were quite popular because they 151 00:08:11,400 --> 00:08:14,200 Speaker 1: were very clean, and the reason that them being clean 152 00:08:14,280 --> 00:08:16,920 Speaker 1: wasn't usability. It was the fact it loaded quickly, which 153 00:08:16,920 --> 00:08:21,000 Speaker 1: I guess would be usability either way. We've also had 154 00:08:21,080 --> 00:08:24,320 Speaker 1: dramatic improvements in web design and accessibility. The advent of 155 00:08:24,320 --> 00:08:27,960 Speaker 1: mobile browsing and the proliferation of widespread mobile and desktop 156 00:08:28,000 --> 00:08:31,000 Speaker 1: Internet access in the two thousands, we were at the 157 00:08:31,120 --> 00:08:33,439 Speaker 1: very early days of e commerce, and the weird irony 158 00:08:33,440 --> 00:08:36,000 Speaker 1: of the dot com bubble is that it was actually 159 00:08:36,000 --> 00:08:39,320 Speaker 1: pretty useful to lay millions of miles of fiber optic cable. 160 00:08:40,240 --> 00:08:44,240 Speaker 1: This is in no way, shape or form remotely comparable 161 00:08:44,240 --> 00:08:48,840 Speaker 1: to large language models GPUs or any nebulous VC spunk 162 00:08:48,880 --> 00:09:07,160 Speaker 1: around Generative AI. Global Internet access has never been higher 163 00:09:07,280 --> 00:09:09,720 Speaker 1: or cheaper, and for the most part, billions of people 164 00:09:09,800 --> 00:09:12,480 Speaker 1: can access the connection fast enough to use Generative AI. 165 00:09:13,040 --> 00:09:16,120 Speaker 1: There is very little stopping anyone from using an LM, 166 00:09:16,360 --> 00:09:19,960 Speaker 1: Chat GBT is free, chat GPT's cheaper ghost subscription has 167 00:09:20,000 --> 00:09:22,960 Speaker 1: now spread to the entire world. When I originally wrote 168 00:09:22,960 --> 00:09:26,160 Speaker 1: this section, it was originally just in the global South, 169 00:09:26,200 --> 00:09:29,480 Speaker 1: but now it's everywhere. Gemini is free, Perplexit is three, 170 00:09:29,480 --> 00:09:32,160 Speaker 1: and metas LM is free. Whether dot com bubble was 171 00:09:32,160 --> 00:09:34,679 Speaker 1: made up of stupid businesses and the lack of fundamental 172 00:09:34,679 --> 00:09:37,400 Speaker 1: infrastructure to give most people the opportunity to access a 173 00:09:37,440 --> 00:09:42,000 Speaker 1: reliable Internet experience, Basically anybody can get reliable access to 174 00:09:42,040 --> 00:09:45,680 Speaker 1: Generative AI. Anyone claiming this is just like the early 175 00:09:45,760 --> 00:09:48,040 Speaker 1: days of the Internet is a fucking liar or a 176 00:09:48,080 --> 00:09:51,280 Speaker 1: fucking moron. Llms have now spread to every nook and 177 00:09:51,280 --> 00:09:54,559 Speaker 1: cranny of the Internet. Anybody can use one, anybody can 178 00:09:54,640 --> 00:09:58,160 Speaker 1: experience the so called power of AI. Users are not 179 00:09:58,240 --> 00:10:01,160 Speaker 1: sitting frothing at the mouth unable to access chet GPT 180 00:10:01,360 --> 00:10:04,760 Speaker 1: due to a lack of infrastructure. Nor is anybody saying, amen, 181 00:10:05,160 --> 00:10:07,360 Speaker 1: I can't access CLAW because I don't have a local 182 00:10:07,440 --> 00:10:09,760 Speaker 1: data center. They might be doing it because there's a 183 00:10:09,760 --> 00:10:12,920 Speaker 1: fucking rate limit, because Anthropic can't afford to run their services, 184 00:10:12,920 --> 00:10:18,760 Speaker 1: but that's not what this is about. Edward experiences are 185 00:10:18,760 --> 00:10:21,440 Speaker 1: not worse because these companies have a lack of access 186 00:10:21,440 --> 00:10:25,079 Speaker 1: to infrastructure or capital. They're worse because the underlying technology 187 00:10:25,080 --> 00:10:28,480 Speaker 1: of transformer based models is inherently limited, and in turn, 188 00:10:28,559 --> 00:10:32,200 Speaker 1: any company connected to these models is limited along with them. 189 00:10:32,720 --> 00:10:34,839 Speaker 1: Then we get to the economics of the AI bubble 190 00:10:34,880 --> 00:10:38,240 Speaker 1: and things begin to get more worrying. While the dot 191 00:10:38,240 --> 00:10:41,400 Speaker 1: com bubble rested on the back of companies like luciend, Cisco, 192 00:10:41,679 --> 00:10:45,679 Speaker 1: Nortel WorldCom Enron, and others, the AI bubble rests fundamentally 193 00:10:45,760 --> 00:10:49,200 Speaker 1: on one company in Video, and to a lesser extent, 194 00:10:49,240 --> 00:10:55,240 Speaker 1: the valuations of the remainder of the Magnificent seven Microsoft Amazon, Google, Meta, Apple, 195 00:10:55,320 --> 00:10:59,120 Speaker 1: and Tesla. Four of those companies Amazon, Microsoft, Google, and 196 00:10:59,200 --> 00:11:02,840 Speaker 1: Meta through to midiaries I'll get to in a future episode. 197 00:11:02,840 --> 00:11:05,680 Speaker 1: I think. Actually i'll explain in a second have spent 198 00:11:05,800 --> 00:11:08,359 Speaker 1: hundreds of billions of dollars on GPUs and their associated 199 00:11:08,360 --> 00:11:11,360 Speaker 1: infrastructure for reasons that none of them can seem to explain. 200 00:11:12,080 --> 00:11:14,960 Speaker 1: As in a side by the way. I will get 201 00:11:15,000 --> 00:11:16,840 Speaker 1: to this in an episode. I had to cut it 202 00:11:16,840 --> 00:11:19,280 Speaker 1: from the script just for length. It's already quite long. 203 00:11:20,240 --> 00:11:23,600 Speaker 1: There is a weird thing going on where Microsoft, Google, Meta, Amazon, 204 00:11:23,640 --> 00:11:27,080 Speaker 1: They don't buy their GPUs directly from Nvidia. They get 205 00:11:27,080 --> 00:11:32,120 Speaker 1: them through various Taiwanese holding companies like Fox Cotton Holding 206 00:11:32,120 --> 00:11:35,920 Speaker 1: companies is the wrong world. Manufacturers of server hardware and 207 00:11:36,000 --> 00:11:40,959 Speaker 1: search called like Honhai, who is Fox con, Wistron, Quantum Computing, 208 00:11:41,000 --> 00:11:43,360 Speaker 1: they order through Taiwan and then those sels are put 209 00:11:43,360 --> 00:11:46,200 Speaker 1: together and shipped to their data centers. This allows them 210 00:11:46,240 --> 00:11:48,960 Speaker 1: to hide how many GPUs they're buying from their investors 211 00:11:49,000 --> 00:11:52,880 Speaker 1: because guess what is quite a long anyway, when this 212 00:11:52,920 --> 00:11:55,480 Speaker 1: all collapses, we're also going to see a market contagion 213 00:11:55,520 --> 00:11:59,080 Speaker 1: that goes to Taiwan because all those Taiwanese companies are 214 00:11:59,080 --> 00:12:03,640 Speaker 1: booking revenue from selling these fucking servers anyway. Lots of 215 00:12:03,679 --> 00:12:07,640 Speaker 1: fun there, but let's keep going now. In Vidious revenue 216 00:12:07,679 --> 00:12:11,240 Speaker 1: is also predominantly eighty eight percent in its data center segment, 217 00:12:11,280 --> 00:12:13,240 Speaker 1: and it's customers of those who can afford at the 218 00:12:13,320 --> 00:12:16,200 Speaker 1: very least fifty to one hundred GPUs retailing at four 219 00:12:16,280 --> 00:12:18,680 Speaker 1: hundred grand or more for a pot of eight of them, 220 00:12:19,080 --> 00:12:21,840 Speaker 1: and you've require tens of thousands of dollars a networking 221 00:12:21,880 --> 00:12:23,600 Speaker 1: gear to go with them to make them turn on. 222 00:12:24,360 --> 00:12:27,040 Speaker 1: The Customers of those renting those GPUs are either AI 223 00:12:27,160 --> 00:12:30,280 Speaker 1: labs training or running inference for models, and their customers 224 00:12:30,320 --> 00:12:33,440 Speaker 1: are AI startups. The problem isn't so much that nobody 225 00:12:33,480 --> 00:12:35,839 Speaker 1: can afford a GPU, but that you can't get very 226 00:12:35,840 --> 00:12:37,680 Speaker 1: far with just one. You have to buy so many 227 00:12:37,720 --> 00:12:39,280 Speaker 1: of them. You need to build a big data center 228 00:12:39,320 --> 00:12:41,840 Speaker 1: around them, You need to get power to that data center, 229 00:12:42,480 --> 00:12:45,400 Speaker 1: and then you have the massive environmental concerns of well 230 00:12:45,480 --> 00:12:49,240 Speaker 1: running all that power. This naturally means that there are 231 00:12:49,240 --> 00:12:52,040 Speaker 1: really only two customers who can afford these chips at scale, 232 00:12:52,440 --> 00:12:55,320 Speaker 1: the Magnificent seven, who have all now begun to take 233 00:12:55,360 --> 00:12:58,920 Speaker 1: on debt after previously financing their GPU purchases with cash 234 00:12:58,920 --> 00:13:02,520 Speaker 1: flow and companies that raise them with companies meaning anybody 235 00:13:02,559 --> 00:13:04,880 Speaker 1: who wants to build a data center, an oracle who 236 00:13:04,960 --> 00:13:07,800 Speaker 1: had negative thirteen billion dollars in cash flow last quarter 237 00:13:08,000 --> 00:13:10,200 Speaker 1: and is steeped in debt to the point that bondholders 238 00:13:10,240 --> 00:13:13,560 Speaker 1: are suing them. We also have no idea if the 239 00:13:13,559 --> 00:13:16,960 Speaker 1: economics of renting GPUs actually makes sense, and based on 240 00:13:17,000 --> 00:13:19,800 Speaker 1: everything I've found, I'm not sure anybody renting them can 241 00:13:19,800 --> 00:13:22,040 Speaker 1: ever make a profit due to either or both the 242 00:13:22,120 --> 00:13:24,600 Speaker 1: upfront cost and debt necessary to pay it, and the 243 00:13:24,640 --> 00:13:28,599 Speaker 1: power intensive nature of providing AI compute. It is fundamentally 244 00:13:28,640 --> 00:13:31,760 Speaker 1: insane that we don't know for sure. It's so crazy. 245 00:13:32,000 --> 00:13:34,160 Speaker 1: How do we not know? How the fuck do we 246 00:13:34,200 --> 00:13:41,520 Speaker 1: not know that this is crazy? It's What we do 247 00:13:41,679 --> 00:13:43,840 Speaker 1: know is that the only company making any kind of 248 00:13:43,840 --> 00:13:46,199 Speaker 1: profit during the AI bubble appears to be in video 249 00:13:46,600 --> 00:13:50,600 Speaker 1: or Companies selling RAM, Microsoft, Google, Meta, and Amazon refuse 250 00:13:50,640 --> 00:13:53,080 Speaker 1: to share their actual AI revenues, and because people have 251 00:13:53,160 --> 00:13:55,960 Speaker 1: the brains of dogs, they have conflated revenue growth from 252 00:13:56,040 --> 00:14:00,480 Speaker 1: hyperscalis already existing segments like software and advertising, with growth 253 00:14:00,520 --> 00:14:04,000 Speaker 1: created by AI. In the dot com bubble, one could, 254 00:14:04,000 --> 00:14:06,040 Speaker 1: at the very least point to where a company was 255 00:14:06,080 --> 00:14:08,520 Speaker 1: making revenue, even if the answer was handing a dollar 256 00:14:08,559 --> 00:14:11,280 Speaker 1: to somebody and getting handed at a dollar back. People 257 00:14:11,280 --> 00:14:14,600 Speaker 1: bought and installed physical infrastructure, and that infrastructure was, albeit 258 00:14:14,640 --> 00:14:17,400 Speaker 1: a much lewer scale than the build out anticipated paid 259 00:14:17,400 --> 00:14:21,040 Speaker 1: for by the associated services. Companies got greedy, rushed to 260 00:14:21,080 --> 00:14:23,880 Speaker 1: expand in the way that was unnecessary, took on ruinous debt, 261 00:14:24,000 --> 00:14:27,600 Speaker 1: and suffered the consequences. This isn't what's happening in the 262 00:14:27,680 --> 00:14:31,080 Speaker 1: AI bubble. Consumers have no problem getting exposure to AI. 263 00:14:31,320 --> 00:14:34,680 Speaker 1: In fact, AI is breaking into every single device and 264 00:14:34,720 --> 00:14:37,800 Speaker 1: app that we have like an angry pervert with a knife. 265 00:14:38,400 --> 00:14:41,680 Speaker 1: While WorldCom wannabes like open Ai and Anthropic are whining 266 00:14:41,680 --> 00:14:44,560 Speaker 1: about not having enough compute, it's very clear they've got 267 00:14:44,560 --> 00:14:46,520 Speaker 1: more than enough to burp out a new model every 268 00:14:46,760 --> 00:14:50,160 Speaker 1: few months or drop copyright infringement machines on millions of 269 00:14:50,160 --> 00:14:53,560 Speaker 1: people at a moment's notice. The post bubble over build 270 00:14:53,600 --> 00:14:56,040 Speaker 1: of fiber left thousands of miles of dark I e 271 00:14:56,280 --> 00:14:59,360 Speaker 1: not connected to anything, cabling that took years to light. 272 00:14:59,440 --> 00:15:01,960 Speaker 1: But doing so, I had an obvious business use case 273 00:15:02,200 --> 00:15:05,080 Speaker 1: connecting people to the Internet and didn't require an entire 274 00:15:05,120 --> 00:15:07,400 Speaker 1: fucking data center and masses of power to do so. 275 00:15:08,120 --> 00:15:10,760 Speaker 1: To make matters worse, as I've hinted that the depreciation 276 00:15:10,840 --> 00:15:15,880 Speaker 1: of these GPUs is utterly brutal purple Kudrowski and I quote, 277 00:15:16,520 --> 00:15:19,560 Speaker 1: we are in a historically anomalous moment. Regardless of what 278 00:15:19,560 --> 00:15:22,320 Speaker 1: one thinks about the merits of AI or explosive data 279 00:15:22,360 --> 00:15:25,840 Speaker 1: center expansion, the scale and pasive capital deployment into a 280 00:15:25,960 --> 00:15:30,200 Speaker 1: rapidly depreciating technology is remarkable. These are not railroads. We 281 00:15:30,240 --> 00:15:34,480 Speaker 1: aren't building century long infrastructure. AI data centers are short lived, 282 00:15:34,640 --> 00:15:39,560 Speaker 1: aset intensive facilities, riding declining cost technology curves, requiring frequent 283 00:15:39,640 --> 00:15:43,800 Speaker 1: hardware replacement to preserve margins. Let me put it a 284 00:15:43,800 --> 00:15:47,480 Speaker 1: little simpler. Imagine if all of that fiber was useless 285 00:15:47,520 --> 00:15:50,920 Speaker 1: in five or six years at best. What if all 286 00:15:50,920 --> 00:15:53,000 Speaker 1: of that fiber could only be used to access a 287 00:15:53,000 --> 00:15:56,680 Speaker 1: small subset of websites. What if all of that fiber 288 00:15:56,760 --> 00:15:59,680 Speaker 1: required such massive amounts of power that it threatened rolling 289 00:15:59,680 --> 00:16:03,440 Speaker 1: blackouts of the East coast of America. That is the 290 00:16:03,480 --> 00:16:05,960 Speaker 1: scale of the apocalypse I am talking about, and I 291 00:16:06,000 --> 00:16:08,600 Speaker 1: am worried that people are not taking the problem. More seriously, 292 00:16:09,760 --> 00:16:12,480 Speaker 1: the demand for Invidio chips is fueled by hype, and 293 00:16:12,520 --> 00:16:15,080 Speaker 1: that hype has caused this company, and to a lesser extent, 294 00:16:15,080 --> 00:16:17,760 Speaker 1: the Magnificent Seven, to become a load bearing part of 295 00:16:17,800 --> 00:16:22,000 Speaker 1: the American stock market. An analysis from portfolio manager Don 296 00:16:22,040 --> 00:16:25,080 Speaker 1: Kee Jong from January twenty twenty five found that the 297 00:16:25,120 --> 00:16:28,600 Speaker 1: Magnificent Seven stocks accounted for forty seven point eight seven 298 00:16:28,640 --> 00:16:32,040 Speaker 1: percent of the Russell one thousand indexes returns in twenty 299 00:16:32,080 --> 00:16:34,560 Speaker 1: twenty four. And that's an index fund of the thousand 300 00:16:34,640 --> 00:16:37,840 Speaker 1: highest ranked stocks and the foot Sea Russell's Index. In 301 00:16:37,880 --> 00:16:40,760 Speaker 1: really simple terms, without the mostly vibes driven nature of 302 00:16:40,760 --> 00:16:43,200 Speaker 1: the Magnificent Seven's growth as none, if this is based 303 00:16:43,240 --> 00:16:46,120 Speaker 1: on anyone's actual revenues, the US stock market would be 304 00:16:46,200 --> 00:16:50,200 Speaker 1: in incredibly rough shape. Except unlike the dot com bubble, 305 00:16:50,400 --> 00:16:52,880 Speaker 1: most of these companies have taken on incredibly large amounts 306 00:16:52,920 --> 00:16:56,720 Speaker 1: of GPUs debt finance and operating leases and data centers 307 00:16:56,720 --> 00:16:58,720 Speaker 1: full of GPUs that can't be used for really much 308 00:16:58,760 --> 00:17:02,320 Speaker 1: of anything else. And because GPU is a guaranteed to depreciate, 309 00:17:02,480 --> 00:17:04,600 Speaker 1: each and every one of them will without fail have 310 00:17:04,640 --> 00:17:07,359 Speaker 1: to write down the value of upwards of one hundred 311 00:17:07,359 --> 00:17:10,120 Speaker 1: billion dollars of investments in the future, as these things 312 00:17:10,160 --> 00:17:14,120 Speaker 1: are eventually facing the recoverability test, which is when there's 313 00:17:14,160 --> 00:17:16,439 Speaker 1: a huge crash within any market sector and you have 314 00:17:16,480 --> 00:17:18,560 Speaker 1: to look at your assets and say, shit, will these 315 00:17:18,600 --> 00:17:22,240 Speaker 1: actually generate enough money? And this is going to happen 316 00:17:22,280 --> 00:17:24,840 Speaker 1: whether or not the AI bubble bursts. In Video is 317 00:17:24,880 --> 00:17:28,879 Speaker 1: on a yearly cycle of upgrades on their GPUs. Every 318 00:17:29,280 --> 00:17:34,560 Speaker 1: single year, every single GPU investment loses value, and to 319 00:17:34,600 --> 00:17:37,320 Speaker 1: make matts worse, takes fucking years to install these things, 320 00:17:37,359 --> 00:17:39,720 Speaker 1: so by the time they're there, your way in the past. 321 00:17:40,440 --> 00:17:43,199 Speaker 1: Even if the AI bubble doesn't burst, it's gonna The 322 00:17:43,320 --> 00:17:46,640 Speaker 1: US stock market has an unhealthy relationship within video, which 323 00:17:46,680 --> 00:17:48,800 Speaker 1: by this time next year will have to make over 324 00:17:48,920 --> 00:17:51,080 Speaker 1: ninety billion dollars a quarter to keep up with its 325 00:17:51,119 --> 00:17:54,119 Speaker 1: ridiculous fifty percent year every year growth, and by twenty 326 00:17:54,160 --> 00:17:57,160 Speaker 1: twenty eight and video will to keep its ridiculous valuation, 327 00:17:57,520 --> 00:17:59,680 Speaker 1: have to be making more than Apple, which makes about 328 00:17:59,680 --> 00:18:03,080 Speaker 1: foreign and sixteen billion dollars a year in revenue. In fact, 329 00:18:03,119 --> 00:18:05,720 Speaker 1: from my calculations, in video will have to be making 330 00:18:05,760 --> 00:18:08,159 Speaker 1: five hundred to six hundred billion dollars, which puts it 331 00:18:08,160 --> 00:18:12,720 Speaker 1: in the realm of Walmart. It can't happen. It can't happen. 332 00:18:13,160 --> 00:18:15,840 Speaker 1: Can't it can't happen, And to do that in video's 333 00:18:15,840 --> 00:18:18,880 Speaker 1: customers will continue having to be able to afford these GPUs, which, 334 00:18:18,880 --> 00:18:21,920 Speaker 1: as have established, are being paid out of debt because 335 00:18:21,920 --> 00:18:26,480 Speaker 1: AI services do not make a profim Even if AI 336 00:18:26,600 --> 00:18:28,880 Speaker 1: services take off and are useful in a way they've 337 00:18:28,920 --> 00:18:31,919 Speaker 1: never even remotely hinted it being, it is inevitable that 338 00:18:31,960 --> 00:18:34,520 Speaker 1: the debt and cash necessary to keep buying in video 339 00:18:34,600 --> 00:18:37,720 Speaker 1: GPUs runs out, and more than likely the revenues of 340 00:18:37,720 --> 00:18:40,800 Speaker 1: the Magnificent seven will stumble in growth before then, as 341 00:18:40,800 --> 00:18:43,560 Speaker 1: it becomes obvious that those GPUs are not providing any 342 00:18:43,600 --> 00:18:47,119 Speaker 1: meaningful revenue growth. The result, I fear, is that the 343 00:18:47,200 --> 00:18:51,560 Speaker 1: American stock market takes as shit the size of Iowa, 344 00:18:51,600 --> 00:18:54,160 Speaker 1: and due to the unique way that the tech industry functions, 345 00:18:54,200 --> 00:18:57,520 Speaker 1: the contagion will be global. I'll catch you tomorrow for 346 00:18:57,560 --> 00:19:00,000 Speaker 1: part three. I don't have a rosy or funny app 347 00:19:00,280 --> 00:19:02,160 Speaker 1: Every time I think of this stuff, I feel very 348 00:19:02,280 --> 00:19:06,240 Speaker 1: very sad. Well anyway, very optimistic. 349 00:19:06,280 --> 00:19:06,560 Speaker 2: Peace. 350 00:19:06,680 --> 00:19:12,480 Speaker 1: I'll catch you tomorrow. 351 00:19:16,000 --> 00:19:18,439 Speaker 2: Thank you for listening to better Offline The editor and 352 00:19:18,440 --> 00:19:21,600 Speaker 2: composer of the Better Offline theme song is Matasowski. You 353 00:19:21,640 --> 00:19:23,880 Speaker 2: can check out more of his music and audio projects 354 00:19:24,040 --> 00:19:27,560 Speaker 2: at Mattasowski dot com, m A T T O s 355 00:19:27,600 --> 00:19:31,679 Speaker 2: O W s ki dot com. You can email me 356 00:19:31,720 --> 00:19:34,320 Speaker 2: at easy at Better offline dot com or visit Better 357 00:19:34,359 --> 00:19:36,800 Speaker 2: Offline dot com to find more podcast links and of course, 358 00:19:36,840 --> 00:19:39,960 Speaker 2: my newsletter. I also really recommend you go to chat 359 00:19:39,960 --> 00:19:42,640 Speaker 2: dot Where's youreed dot at to visit the discord, and 360 00:19:42,680 --> 00:19:43,720 Speaker 2: go to our slash. 361 00:19:43,520 --> 00:19:45,359 Speaker 1: Better Offline to check out our reddit. 362 00:19:46,160 --> 00:19:49,399 Speaker 2: Thank you so much for listening. Better Offline is a 363 00:19:49,440 --> 00:19:50,800 Speaker 2: production of cool Zone Media. 364 00:19:50,960 --> 00:19:53,800 Speaker 1: For more from cool Zone Media, visit our website cool 365 00:19:53,840 --> 00:19:57,240 Speaker 1: Zonemedia dot com, or check us out on the iHeartRadio app, 366 00:19:57,280 --> 00:20:20,920 Speaker 1: Apple Podcasts, or wherever you get your podcasts.