WEBVTT - AI Is Worse Than The Dot Com Bubble: Part Two 0:00:02.800 --> 0:00:17.880 Zone Media. I'm d Zetron and this is better offline. 0:00:19.880 --> 0:00:21.720 That's right, folks, We're back for the second part of 0:00:21.760 --> 0:00:23.799 our series in the dot Com Bubble and why I 0:00:23.800 --> 0:00:27.320 believe the AI Bubble could be much, much worse. While 0:00:27.360 --> 0:00:29.440 the dot com bubble was a mixture of dodgy venture 0:00:29.480 --> 0:00:32.080 capital deals and websites that could never turn a profit, 0:00:32.120 --> 0:00:35.400 combined with a global mania around the interconnectivity of high 0:00:35.400 --> 0:00:38.680 speed Internet companies, the AI bubble is one company selling 0:00:38.800 --> 0:00:42.320 expensive aigpus, a bunch of companies building data sentence to 0:00:42.320 --> 0:00:44.360 put them in, and a bunch of companies building shit 0:00:44.400 --> 0:00:46.640 that runs on GPUs that only loses money and that 0:00:46.680 --> 0:00:49.720 customers kind of fucking hate. I should also note that 0:00:49.760 --> 0:00:52.000 a very important part of the story is venture capital's 0:00:52.080 --> 0:00:54.640 lack of returns in the last few years, something I 0:00:54.680 --> 0:00:57.960 covered in the Hitter financial Crisis last week, specifically in 0:00:58.000 --> 0:01:02.360 parts two and three. In simple terms, AI startups now 0:01:02.400 --> 0:01:04.880 make up more than half a venture investment, and I 0:01:04.920 --> 0:01:07.040 believe that most of these stops will die because of 0:01:07.040 --> 0:01:10.760 their horrible margins, no part to profitability, and products that 0:01:10.760 --> 0:01:13.080 people really don't want to pay for its scale, and 0:01:13.120 --> 0:01:15.679 when they die, they will leave venture capitalists stranded with 0:01:15.760 --> 0:01:18.240 tons of dead equity at a time when they already 0:01:18.319 --> 0:01:21.600 have trouble generating returns and thus raising money from their 0:01:21.640 --> 0:01:25.839 limited partners. The result, I worry, will be gruesome. Venture 0:01:25.840 --> 0:01:28.319 capitalists make their money through the fees they generate, which 0:01:28.360 --> 0:01:30.240 are based on the value of their investments and the 0:01:30.280 --> 0:01:34.280 returns they give their investors, which don't seem to be happening. 0:01:34.680 --> 0:01:36.920 What do you think happens when they can't generate any 0:01:36.959 --> 0:01:40.360 returns and their investments aren't worth anything. The answer is simple, 0:01:40.640 --> 0:01:42.800 they won't have any way of raising more capital as 0:01:42.840 --> 0:01:45.600 their limited partners won't fucking trust them. And to be clear, 0:01:45.880 --> 0:01:48.920 these arswapes have cocked it up many years at a time. 0:01:49.240 --> 0:01:53.480 Look at crypto, look at NFTs, look at ar VR, metaverse, 0:01:53.600 --> 0:01:56.600 all of that, and all of this ridiculousness has happened 0:01:56.600 --> 0:02:00.320 because of the ridiculous, ridiculous myths of AI, like just 0:02:00.320 --> 0:02:03.480 the insatiable demand for AI compute and the made up 0:02:03.520 --> 0:02:06.640 decline in the price of intelligence, which, by the way, 0:02:07.640 --> 0:02:10.520 that last one I've talked about it, I swear to god, 0:02:10.560 --> 0:02:13.120 I mentioned it in the Guardia, mentioned it everywhere. It 0:02:13.160 --> 0:02:15.760 may be cheaper to pay for the tokens, but you 0:02:15.960 --> 0:02:18.960 use more of them, so it's more expensive in the aggregate. 0:02:19.200 --> 0:02:21.400 On top of that, it doesn't mean that the price 0:02:21.440 --> 0:02:25.120 of intelligence for the model providers is going down. Jesus Christ, 0:02:25.160 --> 0:02:30.200 I'm so tired of making this points. But in both cases, 0:02:30.240 --> 0:02:33.800 these assumptions are convincing investors that it's time to invest 0:02:33.800 --> 0:02:36.239 in data centers that only lose money because you assume 0:02:36.320 --> 0:02:38.680 that the demand will be there, or in AI startups 0:02:38.680 --> 0:02:41.400 to only lose money because you'll assume they magically stop 0:02:41.480 --> 0:02:44.720 losing money somehow. And it turns out we do actually 0:02:44.760 --> 0:02:48.200 have a historical comparison. The mania of the dot com 0:02:48.200 --> 0:02:50.480 bubble was based on a misunderstanding of the scale of 0:02:50.480 --> 0:02:53.640 the Internet at the time, rather than its actual potential. 0:02:54.160 --> 0:02:57.239 Hundreds of billions of dollars were invested based on flimsy logic. 0:02:57.800 --> 0:03:02.200 To quote researcher Justin Kohler. This continental rewiring was also 0:03:02.320 --> 0:03:05.520 justified by another powerful myth that Internet traffic was doubling 0:03:05.520 --> 0:03:10.320 every ninety days. This claim spread through analyst reports, burning 0:03:10.360 --> 0:03:15.200 course and investor presentations like a particularly virulent meme. If true, 0:03:15.480 --> 0:03:18.280 it meant the demand was growing exponentially, far out pacing 0:03:18.280 --> 0:03:21.320 any conceivable supply, and that every new trench of fiber 0:03:21.360 --> 0:03:23.960 would soon pay for itself many times over. And I 0:03:24.040 --> 0:03:27.200 pause here to go, oh god, they're doing it again. 0:03:27.960 --> 0:03:31.640 Back to the quote. But the mathematics were fiction. Network 0:03:31.680 --> 0:03:35.000 researchers like Andrew Adlisco at AT and T, looking at 0:03:35.040 --> 0:03:38.720 actual traffic data, found that US backbone traffic was doubling 0:03:38.840 --> 0:03:42.440 roughly once a year, rapid growth, certainly, but nowhere near 0:03:42.520 --> 0:03:46.680 the purported ninety day cycle. Meanwhile, advances in fiber technology 0:03:46.720 --> 0:03:51.240 were making each strand exponentially more powerful. Dense wavelength division 0:03:51.320 --> 0:03:54.760 multiplexing allowed dozens of signals to travel simultaneously down the 0:03:54.800 --> 0:03:58.560 same line a different wavelengths of light, like multiple conversations 0:03:58.560 --> 0:04:02.760 happening in different colours. While demand doubled annually, supply expanded 0:04:02.800 --> 0:04:06.880 tenfold or more. Carriers buried the discrepancy under layers of 0:04:06.920 --> 0:04:10.640 creative accounting that would have impressed medieval alchemists. First of all, 0:04:10.760 --> 0:04:13.120 justin if you hear this, I fucking love that. That 0:04:13.240 --> 0:04:16.400 was very fun. Second of all, to quote twin peaks, 0:04:17.240 --> 0:04:21.920 it's happening again. Mania had taken hold based on very 0:04:21.920 --> 0:04:25.760 flimsy logic. Economically speaking, this meant the telecoms companies, serve 0:04:25.760 --> 0:04:30.440 a hardware companies, ISPs, construction firms, optical cable providers, wireless 0:04:30.440 --> 0:04:33.560 technology companies, and basically anybody related to the business of 0:04:33.600 --> 0:04:36.279 providing Internet access in any way saw a massive influx 0:04:36.320 --> 0:04:39.080 of business to build capacity that didn't need to be 0:04:39.080 --> 0:04:42.200 built yet. If you're in the business of selling services 0:04:42.200 --> 0:04:44.120 to get people online, you were high on the hug, 0:04:44.480 --> 0:04:49.040 you know, kind of like selling high bandwidth rem Similarly, 0:04:49.080 --> 0:04:51.200 one could get a startup funded if you had a website, 0:04:51.279 --> 0:04:53.400 or even take in public. One could raise debt to 0:04:53.400 --> 0:04:56.039 build a nascent ISSP or a fiber network. The money 0:04:56.080 --> 0:04:59.840 was flowing because people people weren't really really being thoughtful 0:05:00.000 --> 0:05:02.400 about it. And as far as the dot com part 0:05:02.520 --> 0:05:05.800 of the dot com bubble, the unsustainable websites, the problem 0:05:05.839 --> 0:05:08.680 wasn't so much the industry but the businesses themselves, which 0:05:08.720 --> 0:05:11.479 were hyped and dumped onto the public markets with little 0:05:11.480 --> 0:05:14.080 regard for their long term health. The problem here is 0:05:14.120 --> 0:05:17.000 relatively simple. These were bad companies that people ignored the 0:05:17.040 --> 0:05:19.080 issues with because of and they quote the power of 0:05:19.120 --> 0:05:22.480 the Internet and how it would somehow save them, which 0:05:22.640 --> 0:05:26.599 which it didn't obviously had they been kept private. And 0:05:26.640 --> 0:05:28.680 died in the dark. I don't think these companies would 0:05:28.720 --> 0:05:30.800 have had the same reputation. I don't think we give 0:05:30.839 --> 0:05:34.559 a fuck about pets dot com. I also don't think 0:05:34.600 --> 0:05:37.680 they were really in accurate comparison to anything happening today. 0:05:38.839 --> 0:05:42.320 While the valuations were ridiculous, the Globe's market cap was 0:05:42.320 --> 0:05:45.039 at one point eight hundred and forty million dollars, the 0:05:45.080 --> 0:05:48.880 scale of destruction caused by dot com startups was significantly smaller, 0:05:49.120 --> 0:05:52.440 even in today's money. The economics were bad, but not 0:05:52.480 --> 0:05:55.400 anywhere near as bad. For the first nine months of 0:05:55.480 --> 0:05:57.760 nineteen ninety eight, the Globe made two point seven million 0:05:57.800 --> 0:06:00.680 dollars in revenue and lost eleven point five million dollars, 0:06:00.760 --> 0:06:03.800 largely due to trying to move into multiple different business 0:06:03.800 --> 0:06:06.320 lines at once like voice over IP. And to be clear, 0:06:06.400 --> 0:06:09.839 this company made money selling ads and also buying random companies. 0:06:09.880 --> 0:06:13.080 Buying random companies was the thing that happened during the 0:06:13.120 --> 0:06:16.640 dot com boom. Everybody fucking loved buying companies. You just 0:06:16.680 --> 0:06:21.920 bought random. There was I think like Excite, bought at Home, 0:06:22.120 --> 0:06:25.960 like there was the AT and T sort, the Aol 0:06:26.040 --> 0:06:31.160 time Warner merger. So many stupid mergers, so little time. Nevertheless, 0:06:31.160 --> 0:06:33.599 the economics of this shit show were quite complex. You 0:06:33.640 --> 0:06:36.000 had companies raising money to do any website they could 0:06:36.000 --> 0:06:39.120 think of, companies raising money to lay fiber, companies raising 0:06:39.160 --> 0:06:42.400 money to found ISPs, all of which had multifaested layers 0:06:42.400 --> 0:06:47.000 of physical and digital infrastructure that were quite ugh unbuilt. 0:06:47.040 --> 0:06:50.440 I think is the term. It's tempting yet incorrect to 0:06:50.440 --> 0:06:54.000 say the thing about AI. The similarity everybody points to 0:06:54.080 --> 0:06:55.919 is that people doubted the Internet at the time, and 0:06:55.960 --> 0:06:59.000 people really need to remember their fucking history. In two 0:06:59.000 --> 0:07:01.160 thousand and only fifty two percent of Americans who were 0:07:01.240 --> 0:07:03.480 using the Internet. By two thousand and three, the number 0:07:03.520 --> 0:07:06.640 had only increased the sixty one percent per the World Bank. 0:07:06.680 --> 0:07:09.240 In two thousand and five, only sixteen percent of the 0:07:09.240 --> 0:07:11.600 world used the Internet, and in twenty twenty four the 0:07:11.760 --> 0:07:14.640 number had increased to seventy one percent. Yet the real 0:07:14.680 --> 0:07:17.680 difference is the access to high speed Internet. When the 0:07:17.680 --> 0:07:20.320 Internet was connected via a fifth to six K modem, 0:07:20.440 --> 0:07:22.920 access was at times charged by the minute, and even 0:07:22.920 --> 0:07:25.520 if it was unlimited, it was always much much slower. 0:07:26.000 --> 0:07:28.120 While we're used to connecting its speeds that make using 0:07:28.200 --> 0:07:31.320 web based app near indistinguishable from using one on a computer. 0:07:31.680 --> 0:07:33.760 Back in two thousand, two thousand and one or two 0:07:33.800 --> 0:07:36.360 thousand and two, the average US Internet speed was at 0:07:36.360 --> 0:07:39.400 best four hundred kilobits per second, or roughly fifty killer 0:07:39.480 --> 0:07:42.560 bytes per second, compared to the average US Internet speed 0:07:42.560 --> 0:07:45.040 today of other two hundred megabits per second or twenty 0:07:45.040 --> 0:07:49.520 five megabytes per second. In simpler terms than the younger 0:07:49.600 --> 0:07:52.640 members of the audience won't understand this, A website took 0:07:52.640 --> 0:07:54.800 time to load in a way that feels almost impossible 0:07:54.800 --> 0:07:56.960 to conceive if you didn't experience it at the time, 0:07:57.480 --> 0:07:59.720 you had to make a commitment to God all website. 0:07:59.760 --> 0:08:02.880 It was like you've browsed multiple tabs and fuck around 0:08:02.880 --> 0:08:05.560 the different windows. You sat there and you waited a 0:08:05.600 --> 0:08:08.080 little bit. Sometimes it came up quick and then another. 0:08:08.240 --> 0:08:11.360 In fact, websites like Google were quite popular because they 0:08:11.400 --> 0:08:14.200 were very clean, and the reason that them being clean 0:08:14.280 --> 0:08:16.920 wasn't usability. It was the fact it loaded quickly, which 0:08:16.920 --> 0:08:21.000 I guess would be usability either way. We've also had 0:08:21.080 --> 0:08:24.320 dramatic improvements in web design and accessibility. The advent of 0:08:24.320 --> 0:08:27.960 mobile browsing and the proliferation of widespread mobile and desktop 0:08:28.000 --> 0:08:31.000 Internet access in the two thousands, we were at the 0:08:31.120 --> 0:08:33.439 very early days of e commerce, and the weird irony 0:08:33.440 --> 0:08:36.000 of the dot com bubble is that it was actually 0:08:36.000 --> 0:08:39.320 pretty useful to lay millions of miles of fiber optic cable. 0:08:40.240 --> 0:08:44.240 This is in no way, shape or form remotely comparable 0:08:44.240 --> 0:08:48.840 to large language models GPUs or any nebulous VC spunk 0:08:48.880 --> 0:09:07.160 around Generative AI. Global Internet access has never been higher 0:09:07.280 --> 0:09:09.720 or cheaper, and for the most part, billions of people 0:09:09.800 --> 0:09:12.480 can access the connection fast enough to use Generative AI. 0:09:13.040 --> 0:09:16.120 There is very little stopping anyone from using an LM, 0:09:16.360 --> 0:09:19.960 Chat GBT is free, chat GPT's cheaper ghost subscription has 0:09:20.000 --> 0:09:22.960 now spread to the entire world. When I originally wrote 0:09:22.960 --> 0:09:26.160 this section, it was originally just in the global South, 0:09:26.200 --> 0:09:29.480 but now it's everywhere. Gemini is free, Perplexit is three, 0:09:29.480 --> 0:09:32.160 and metas LM is free. Whether dot com bubble was 0:09:32.160 --> 0:09:34.679 made up of stupid businesses and the lack of fundamental 0:09:34.679 --> 0:09:37.400 infrastructure to give most people the opportunity to access a 0:09:37.440 --> 0:09:42.000 reliable Internet experience, Basically anybody can get reliable access to 0:09:42.040 --> 0:09:45.680 Generative AI. Anyone claiming this is just like the early 0:09:45.760 --> 0:09:48.040 days of the Internet is a fucking liar or a 0:09:48.080 --> 0:09:51.280 fucking moron. Llms have now spread to every nook and 0:09:51.280 --> 0:09:54.559 cranny of the Internet. Anybody can use one, anybody can 0:09:54.640 --> 0:09:58.160 experience the so called power of AI. Users are not 0:09:58.240 --> 0:10:01.160 sitting frothing at the mouth unable to access chet GPT 0:10:01.360 --> 0:10:04.760 due to a lack of infrastructure. Nor is anybody saying, amen, 0:10:05.160 --> 0:10:07.360 I can't access CLAW because I don't have a local 0:10:07.440 --> 0:10:09.760 data center. They might be doing it because there's a 0:10:09.760 --> 0:10:12.920 fucking rate limit, because Anthropic can't afford to run their services, 0:10:12.920 --> 0:10:18.760 but that's not what this is about. Edward experiences are 0:10:18.760 --> 0:10:21.440 not worse because these companies have a lack of access 0:10:21.440 --> 0:10:25.079 to infrastructure or capital. They're worse because the underlying technology 0:10:25.080 --> 0:10:28.480 of transformer based models is inherently limited, and in turn, 0:10:28.559 --> 0:10:32.200 any company connected to these models is limited along with them. 0:10:32.720 --> 0:10:34.839 Then we get to the economics of the AI bubble 0:10:34.880 --> 0:10:38.240 and things begin to get more worrying. While the dot 0:10:38.240 --> 0:10:41.400 com bubble rested on the back of companies like luciend, Cisco, 0:10:41.679 --> 0:10:45.679 Nortel WorldCom Enron, and others, the AI bubble rests fundamentally 0:10:45.760 --> 0:10:49.200 on one company in Video, and to a lesser extent, 0:10:49.240 --> 0:10:55.240 the valuations of the remainder of the Magnificent seven Microsoft Amazon, Google, Meta, Apple, 0:10:55.320 --> 0:10:59.120 and Tesla. Four of those companies Amazon, Microsoft, Google, and 0:10:59.200 --> 0:11:02.840 Meta through to midiaries I'll get to in a future episode. 0:11:02.840 --> 0:11:05.680 I think. Actually i'll explain in a second have spent 0:11:05.800 --> 0:11:08.359 hundreds of billions of dollars on GPUs and their associated 0:11:08.360 --> 0:11:11.360 infrastructure for reasons that none of them can seem to explain. 0:11:12.080 --> 0:11:14.960 As in a side by the way. I will get 0:11:15.000 --> 0:11:16.840 to this in an episode. I had to cut it 0:11:16.840 --> 0:11:19.280 from the script just for length. It's already quite long. 0:11:20.240 --> 0:11:23.600 There is a weird thing going on where Microsoft, Google, Meta, Amazon, 0:11:23.640 --> 0:11:27.080 They don't buy their GPUs directly from Nvidia. They get 0:11:27.080 --> 0:11:32.120 them through various Taiwanese holding companies like Fox Cotton Holding 0:11:32.120 --> 0:11:35.920 companies is the wrong world. Manufacturers of server hardware and 0:11:36.000 --> 0:11:40.959 search called like Honhai, who is Fox con, Wistron, Quantum Computing, 0:11:41.000 --> 0:11:43.360 they order through Taiwan and then those sels are put 0:11:43.360 --> 0:11:46.200 together and shipped to their data centers. This allows them 0:11:46.240 --> 0:11:48.960 to hide how many GPUs they're buying from their investors 0:11:49.000 --> 0:11:52.880 because guess what is quite a long anyway, when this 0:11:52.920 --> 0:11:55.480 all collapses, we're also going to see a market contagion 0:11:55.520 --> 0:11:59.080 that goes to Taiwan because all those Taiwanese companies are 0:11:59.080 --> 0:12:03.640 booking revenue from selling these fucking servers anyway. Lots of 0:12:03.679 --> 0:12:07.640 fun there, but let's keep going now. In Vidious revenue 0:12:07.679 --> 0:12:11.240 is also predominantly eighty eight percent in its data center segment, 0:12:11.280 --> 0:12:13.240 and it's customers of those who can afford at the 0:12:13.320 --> 0:12:16.200 very least fifty to one hundred GPUs retailing at four 0:12:16.280 --> 0:12:18.680 hundred grand or more for a pot of eight of them, 0:12:19.080 --> 0:12:21.840 and you've require tens of thousands of dollars a networking 0:12:21.880 --> 0:12:23.600 gear to go with them to make them turn on. 0:12:24.360 --> 0:12:27.040 The Customers of those renting those GPUs are either AI 0:12:27.160 --> 0:12:30.280 labs training or running inference for models, and their customers 0:12:30.320 --> 0:12:33.440 are AI startups. The problem isn't so much that nobody 0:12:33.480 --> 0:12:35.839 can afford a GPU, but that you can't get very 0:12:35.840 --> 0:12:37.680 far with just one. You have to buy so many 0:12:37.720 --> 0:12:39.280 of them. You need to build a big data center 0:12:39.320 --> 0:12:41.840 around them, You need to get power to that data center, 0:12:42.480 --> 0:12:45.400 and then you have the massive environmental concerns of well 0:12:45.480 --> 0:12:49.240 running all that power. This naturally means that there are 0:12:49.240 --> 0:12:52.040 really only two customers who can afford these chips at scale, 0:12:52.440 --> 0:12:55.320 the Magnificent seven, who have all now begun to take 0:12:55.360 --> 0:12:58.920 on debt after previously financing their GPU purchases with cash 0:12:58.920 --> 0:13:02.520 flow and companies that raise them with companies meaning anybody 0:13:02.559 --> 0:13:04.880 who wants to build a data center, an oracle who 0:13:04.960 --> 0:13:07.800 had negative thirteen billion dollars in cash flow last quarter 0:13:08.000 --> 0:13:10.200 and is steeped in debt to the point that bondholders 0:13:10.240 --> 0:13:13.560 are suing them. We also have no idea if the 0:13:13.559 --> 0:13:16.960 economics of renting GPUs actually makes sense, and based on 0:13:17.000 --> 0:13:19.800 everything I've found, I'm not sure anybody renting them can 0:13:19.800 --> 0:13:22.040 ever make a profit due to either or both the 0:13:22.120 --> 0:13:24.600 upfront cost and debt necessary to pay it, and the 0:13:24.640 --> 0:13:28.599 power intensive nature of providing AI compute. It is fundamentally 0:13:28.640 --> 0:13:31.760 insane that we don't know for sure. It's so crazy. 0:13:32.000 --> 0:13:34.160 How do we not know? How the fuck do we 0:13:34.200 --> 0:13:41.520 not know that this is crazy? It's What we do 0:13:41.679 --> 0:13:43.840 know is that the only company making any kind of 0:13:43.840 --> 0:13:46.199 profit during the AI bubble appears to be in video 0:13:46.600 --> 0:13:50.600 or Companies selling RAM, Microsoft, Google, Meta, and Amazon refuse 0:13:50.640 --> 0:13:53.080 to share their actual AI revenues, and because people have 0:13:53.160 --> 0:13:55.960 the brains of dogs, they have conflated revenue growth from 0:13:56.040 --> 0:14:00.480 hyperscalis already existing segments like software and advertising, with growth 0:14:00.520 --> 0:14:04.000 created by AI. In the dot com bubble, one could, 0:14:04.000 --> 0:14:06.040 at the very least point to where a company was 0:14:06.080 --> 0:14:08.520 making revenue, even if the answer was handing a dollar 0:14:08.559 --> 0:14:11.280 to somebody and getting handed at a dollar back. People 0:14:11.280 --> 0:14:14.600 bought and installed physical infrastructure, and that infrastructure was, albeit 0:14:14.640 --> 0:14:17.400 a much lewer scale than the build out anticipated paid 0:14:17.400 --> 0:14:21.040 for by the associated services. Companies got greedy, rushed to 0:14:21.080 --> 0:14:23.880 expand in the way that was unnecessary, took on ruinous debt, 0:14:24.000 --> 0:14:27.600 and suffered the consequences. This isn't what's happening in the 0:14:27.680 --> 0:14:31.080 AI bubble. Consumers have no problem getting exposure to AI. 0:14:31.320 --> 0:14:34.680 In fact, AI is breaking into every single device and 0:14:34.720 --> 0:14:37.800 app that we have like an angry pervert with a knife. 0:14:38.400 --> 0:14:41.680 While WorldCom wannabes like open Ai and Anthropic are whining 0:14:41.680 --> 0:14:44.560 about not having enough compute, it's very clear they've got 0:14:44.560 --> 0:14:46.520 more than enough to burp out a new model every 0:14:46.760 --> 0:14:50.160 few months or drop copyright infringement machines on millions of 0:14:50.160 --> 0:14:53.560 people at a moment's notice. The post bubble over build 0:14:53.600 --> 0:14:56.040 of fiber left thousands of miles of dark I e 0:14:56.280 --> 0:14:59.360 not connected to anything, cabling that took years to light. 0:14:59.440 --> 0:15:01.960 But doing so, I had an obvious business use case 0:15:02.200 --> 0:15:05.080 connecting people to the Internet and didn't require an entire 0:15:05.120 --> 0:15:07.400 fucking data center and masses of power to do so. 0:15:08.120 --> 0:15:10.760 To make matters worse, as I've hinted that the depreciation 0:15:10.840 --> 0:15:15.880 of these GPUs is utterly brutal purple Kudrowski and I quote, 0:15:16.520 --> 0:15:19.560 we are in a historically anomalous moment. Regardless of what 0:15:19.560 --> 0:15:22.320 one thinks about the merits of AI or explosive data 0:15:22.360 --> 0:15:25.840 center expansion, the scale and pasive capital deployment into a 0:15:25.960 --> 0:15:30.200 rapidly depreciating technology is remarkable. These are not railroads. We 0:15:30.240 --> 0:15:34.480 aren't building century long infrastructure. AI data centers are short lived, 0:15:34.640 --> 0:15:39.560 aset intensive facilities, riding declining cost technology curves, requiring frequent 0:15:39.640 --> 0:15:43.800 hardware replacement to preserve margins. Let me put it a 0:15:43.800 --> 0:15:47.480 little simpler. Imagine if all of that fiber was useless 0:15:47.520 --> 0:15:50.920 in five or six years at best. What if all 0:15:50.920 --> 0:15:53.000 of that fiber could only be used to access a 0:15:53.000 --> 0:15:56.680 small subset of websites. What if all of that fiber 0:15:56.760 --> 0:15:59.680 required such massive amounts of power that it threatened rolling 0:15:59.680 --> 0:16:03.440 blackouts of the East coast of America. That is the 0:16:03.480 --> 0:16:05.960 scale of the apocalypse I am talking about, and I 0:16:06.000 --> 0:16:08.600 am worried that people are not taking the problem. More seriously, 0:16:09.760 --> 0:16:12.480 the demand for Invidio chips is fueled by hype, and 0:16:12.520 --> 0:16:15.080 that hype has caused this company, and to a lesser extent, 0:16:15.080 --> 0:16:17.760 the Magnificent Seven, to become a load bearing part of 0:16:17.800 --> 0:16:22.000 the American stock market. An analysis from portfolio manager Don 0:16:22.040 --> 0:16:25.080 Kee Jong from January twenty twenty five found that the 0:16:25.120 --> 0:16:28.600 Magnificent Seven stocks accounted for forty seven point eight seven 0:16:28.640 --> 0:16:32.040 percent of the Russell one thousand indexes returns in twenty 0:16:32.080 --> 0:16:34.560 twenty four. And that's an index fund of the thousand 0:16:34.640 --> 0:16:37.840 highest ranked stocks and the foot Sea Russell's Index. In 0:16:37.880 --> 0:16:40.760 really simple terms, without the mostly vibes driven nature of 0:16:40.760 --> 0:16:43.200 the Magnificent Seven's growth as none, if this is based 0:16:43.240 --> 0:16:46.120 on anyone's actual revenues, the US stock market would be 0:16:46.200 --> 0:16:50.200 in incredibly rough shape. Except unlike the dot com bubble, 0:16:50.400 --> 0:16:52.880 most of these companies have taken on incredibly large amounts 0:16:52.920 --> 0:16:56.720 of GPUs debt finance and operating leases and data centers 0:16:56.720 --> 0:16:58.720 full of GPUs that can't be used for really much 0:16:58.760 --> 0:17:02.320 of anything else. And because GPU is a guaranteed to depreciate, 0:17:02.480 --> 0:17:04.600 each and every one of them will without fail have 0:17:04.640 --> 0:17:07.359 to write down the value of upwards of one hundred 0:17:07.359 --> 0:17:10.120 billion dollars of investments in the future, as these things 0:17:10.160 --> 0:17:14.120 are eventually facing the recoverability test, which is when there's 0:17:14.160 --> 0:17:16.439 a huge crash within any market sector and you have 0:17:16.480 --> 0:17:18.560 to look at your assets and say, shit, will these 0:17:18.600 --> 0:17:22.240 actually generate enough money? And this is going to happen 0:17:22.280 --> 0:17:24.840 whether or not the AI bubble bursts. In Video is 0:17:24.880 --> 0:17:28.879 on a yearly cycle of upgrades on their GPUs. Every 0:17:29.280 --> 0:17:34.560 single year, every single GPU investment loses value, and to 0:17:34.600 --> 0:17:37.320 make matts worse, takes fucking years to install these things, 0:17:37.359 --> 0:17:39.720 so by the time they're there, your way in the past. 0:17:40.440 --> 0:17:43.199 Even if the AI bubble doesn't burst, it's gonna The 0:17:43.320 --> 0:17:46.640 US stock market has an unhealthy relationship within video, which 0:17:46.680 --> 0:17:48.800 by this time next year will have to make over 0:17:48.920 --> 0:17:51.080 ninety billion dollars a quarter to keep up with its 0:17:51.119 --> 0:17:54.119 ridiculous fifty percent year every year growth, and by twenty 0:17:54.160 --> 0:17:57.160 twenty eight and video will to keep its ridiculous valuation, 0:17:57.520 --> 0:17:59.680 have to be making more than Apple, which makes about 0:17:59.680 --> 0:18:03.080 foreign and sixteen billion dollars a year in revenue. In fact, 0:18:03.119 --> 0:18:05.720 from my calculations, in video will have to be making 0:18:05.760 --> 0:18:08.159 five hundred to six hundred billion dollars, which puts it 0:18:08.160 --> 0:18:12.720 in the realm of Walmart. It can't happen. It can't happen. 0:18:13.160 --> 0:18:15.840 Can't it can't happen, And to do that in video's 0:18:15.840 --> 0:18:18.880 customers will continue having to be able to afford these GPUs, which, 0:18:18.880 --> 0:18:21.920 as have established, are being paid out of debt because 0:18:21.920 --> 0:18:26.480 AI services do not make a profim Even if AI 0:18:26.600 --> 0:18:28.880 services take off and are useful in a way they've 0:18:28.920 --> 0:18:31.919 never even remotely hinted it being, it is inevitable that 0:18:31.960 --> 0:18:34.520 the debt and cash necessary to keep buying in video 0:18:34.600 --> 0:18:37.720 GPUs runs out, and more than likely the revenues of 0:18:37.720 --> 0:18:40.800 the Magnificent seven will stumble in growth before then, as 0:18:40.800 --> 0:18:43.560 it becomes obvious that those GPUs are not providing any 0:18:43.600 --> 0:18:47.119 meaningful revenue growth. The result, I fear, is that the 0:18:47.200 --> 0:18:51.560 American stock market takes as shit the size of Iowa, 0:18:51.600 --> 0:18:54.160 and due to the unique way that the tech industry functions, 0:18:54.200 --> 0:18:57.520 the contagion will be global. I'll catch you tomorrow for 0:18:57.560 --> 0:19:00.000 part three. I don't have a rosy or funny app 0:19:00.280 --> 0:19:02.160 Every time I think of this stuff, I feel very 0:19:02.280 --> 0:19:06.240 very sad. Well anyway, very optimistic. 0:19:06.280 --> 0:19:06.560 Peace. 0:19:06.680 --> 0:19:12.480 I'll catch you tomorrow. 0:19:16.000 --> 0:19:18.439 Thank you for listening to better Offline The editor and 0:19:18.440 --> 0:19:21.600 composer of the Better Offline theme song is Matasowski. You 0:19:21.640 --> 0:19:23.880 can check out more of his music and audio projects 0:19:24.040 --> 0:19:27.560 at Mattasowski dot com, m A T T O s 0:19:27.600 --> 0:19:31.679 O W s ki dot com. You can email me 0:19:31.720 --> 0:19:34.320 at easy at Better offline dot com or visit Better 0:19:34.359 --> 0:19:36.800 Offline dot com to find more podcast links and of course, 0:19:36.840 --> 0:19:39.960 my newsletter. I also really recommend you go to chat 0:19:39.960 --> 0:19:42.640 dot Where's youreed dot at to visit the discord, and 0:19:42.680 --> 0:19:43.720 go to our slash. 0:19:43.520 --> 0:19:45.359 Better Offline to check out our reddit. 0:19:46.160 --> 0:19:49.399 Thank you so much for listening. Better Offline is a 0:19:49.440 --> 0:19:50.800 production of cool Zone Media. 0:19:50.960 --> 0:19:53.800 For more from cool Zone Media, visit our website cool 0:19:53.840 --> 0:19:57.240 Zonemedia dot com, or check us out on the iHeartRadio app, 0:19:57.280 --> 0:20:20.920 Apple Podcasts, or wherever you get your podcasts.