1 00:00:03,160 --> 00:00:07,360 Speaker 1: Media. Hello and welcome to bear off Line. I'm your 2 00:00:07,360 --> 00:00:22,160 Speaker 1: host ed Zetron. Today is the final part of Bubble 3 00:00:22,200 --> 00:00:25,279 Speaker 1: Week and our four part dot com Bubble special. I 4 00:00:25,320 --> 00:00:28,160 Speaker 1: am worryingly invested in this show. I've put so many 5 00:00:28,160 --> 00:00:30,080 Speaker 1: hours into it. This is the most I've done in 6 00:00:30,120 --> 00:00:32,760 Speaker 1: the January. I'm having the time of my fucking life. 7 00:00:33,320 --> 00:00:36,120 Speaker 1: But this four part has been rough, and it's not 8 00:00:36,159 --> 00:00:38,639 Speaker 1: been rough for the recording. It's rough because every time 9 00:00:38,680 --> 00:00:42,159 Speaker 1: I go over these details, they fill me full of anxiety. 10 00:00:42,280 --> 00:00:46,520 Speaker 1: I don't enjoy being right here. Sure I get the satisfaction, 11 00:00:46,680 --> 00:00:48,800 Speaker 1: Sure I get the clicks and all that, but like 12 00:00:51,080 --> 00:00:55,480 Speaker 1: it sucks, It sucks. I'm really fucking worried. I'm not 13 00:00:55,600 --> 00:00:58,280 Speaker 1: doing this to be contrarian. I'm seeing these things and 14 00:00:58,320 --> 00:01:00,400 Speaker 1: I'm worried more people aren't seeing them, and I'm worried 15 00:01:00,400 --> 00:01:04,120 Speaker 1: that the after effects will be so much worse than 16 00:01:04,120 --> 00:01:08,040 Speaker 1: the dot com bubble. So this Foe part was somewhat 17 00:01:08,120 --> 00:01:09,920 Speaker 1: drawn from a premium piece I did a few weeks 18 00:01:09,959 --> 00:01:12,959 Speaker 1: ago called Inventively. This is worse than the dot com bubble. 19 00:01:13,000 --> 00:01:15,880 Speaker 1: I really suggest you subscribe to my premium newsletter or 20 00:01:15,959 --> 00:01:17,880 Speaker 1: be in there. But if you don't. I've had a 21 00:01:17,880 --> 00:01:20,560 Speaker 1: little mustard on it and truncated it for this because 22 00:01:20,920 --> 00:01:24,679 Speaker 1: you needed to be different in voice. Nevertheless, you should subscribe. 23 00:01:24,920 --> 00:01:27,600 Speaker 1: But when I was writing that, I actually went into 24 00:01:27,640 --> 00:01:31,039 Speaker 1: it trying to prove myself wrong. I spent hours and 25 00:01:31,120 --> 00:01:34,480 Speaker 1: hours reading old articles through broken paywalls and archive dot 26 00:01:34,560 --> 00:01:36,920 Speaker 1: org links, in the hopes that I'd find some sort 27 00:01:36,959 --> 00:01:39,640 Speaker 1: of positive spin, a way that the AI bubble wasn't 28 00:01:39,680 --> 00:01:42,399 Speaker 1: as bad, some sliver or nugget of joy, some shread 29 00:01:42,440 --> 00:01:45,160 Speaker 1: of hope. And everything I've read has made me more worried, 30 00:01:45,280 --> 00:01:49,120 Speaker 1: not less. The real problem is that there are similarities, 31 00:01:49,160 --> 00:01:52,280 Speaker 1: but only with the worst parts of both bubbles. We've 32 00:01:52,280 --> 00:01:55,200 Speaker 1: got stupid amounts of money being sunk into stuff. We've 33 00:01:55,200 --> 00:01:58,680 Speaker 1: got stupid companies that want to ypo. We've got hopeless 34 00:01:58,680 --> 00:02:02,360 Speaker 1: startups duced up by edulous venture capitalists and marketed by 35 00:02:02,400 --> 00:02:05,640 Speaker 1: even more credulous journalists. Except this time we have the 36 00:02:05,640 --> 00:02:09,520 Speaker 1: worst margins this side of a vineyard planted in Las Vegas, Nevada, 37 00:02:09,560 --> 00:02:11,760 Speaker 1: and despicable boosters that will go to war with you 38 00:02:11,840 --> 00:02:14,440 Speaker 1: to try and improve You're wrong. Using ghost stories, fairy tales, 39 00:02:14,440 --> 00:02:17,880 Speaker 1: and Bixie dust, I realize I'm being a little cavalier 40 00:02:17,880 --> 00:02:22,400 Speaker 1: and colloquial about this, and that's because I feel emotionally drained, 41 00:02:22,680 --> 00:02:25,359 Speaker 1: is the way i'd put in. Things are so much 42 00:02:25,400 --> 00:02:28,720 Speaker 1: worse than I thought. I really thought i'd find something good. 43 00:02:28,760 --> 00:02:30,920 Speaker 1: I'd dig into the past and find that there were direct, 44 00:02:31,280 --> 00:02:35,720 Speaker 1: obvious similarities that would say, Okay, these GPUs will be useful. 45 00:02:35,760 --> 00:02:40,000 Speaker 1: There will be something we've built here. I can't find it. 46 00:02:40,080 --> 00:02:42,239 Speaker 1: I'm actually now convinced that things are going to be 47 00:02:42,320 --> 00:02:45,600 Speaker 1: much worse. The dot com bubble was relatively distributed and 48 00:02:45,639 --> 00:02:49,080 Speaker 1: created useful infrastructure. The ideas of some of the most 49 00:02:49,120 --> 00:02:52,520 Speaker 1: prominent dot com startups were valid, but based on stupid 50 00:02:52,600 --> 00:02:55,200 Speaker 1: hypergrowth business models that didn't make sense and were only 51 00:02:55,240 --> 00:02:59,959 Speaker 1: sustained through mass hysteria and ignorance. By comparison, AI startups 52 00:03:00,120 --> 00:03:03,880 Speaker 1: fundamentally illogical or selling unreliable software that's best known for 53 00:03:03,919 --> 00:03:06,919 Speaker 1: its consistent mistakes and active harms, backed by the corrosive 54 00:03:07,000 --> 00:03:10,760 Speaker 1: and ever worsening infrastructural costs of GPUs, and they're the 55 00:03:10,800 --> 00:03:13,440 Speaker 1: customers of the largest company on the stock market that 56 00:03:13,520 --> 00:03:16,520 Speaker 1: can only keep its valuation on which the markets depend 57 00:03:16,800 --> 00:03:18,919 Speaker 1: by selling more of it. An expensive to buy and 58 00:03:19,000 --> 00:03:21,720 Speaker 1: more expensive to install item that only appears to lose 59 00:03:21,760 --> 00:03:25,760 Speaker 1: its customers money. There is no squaring in Vidia's circle. 60 00:03:26,200 --> 00:03:29,600 Speaker 1: It is inevitable even if AI somehow worked out that 61 00:03:29,720 --> 00:03:32,840 Speaker 1: companies will have to stop buying as many GPUs, and 62 00:03:32,840 --> 00:03:35,880 Speaker 1: in Vidia cannot afford for them to slow down at all. 63 00:03:36,000 --> 00:03:38,600 Speaker 1: They must speed up, and they must do so forever. 64 00:03:39,640 --> 00:03:43,440 Speaker 1: There's also no other business model for GPUs that scales none. 65 00:03:43,480 --> 00:03:46,520 Speaker 1: There is no use case. This is not useful infrastructure. 66 00:03:46,920 --> 00:03:49,680 Speaker 1: Once the AI bubble bursts, these data centers cannot be 67 00:03:49,720 --> 00:03:51,800 Speaker 1: used for other things, at least not things that are 68 00:03:51,840 --> 00:03:53,800 Speaker 1: going to make more than a couple of single digit 69 00:03:53,880 --> 00:03:56,560 Speaker 1: millions in total. They will have to be emptied out 70 00:03:56,600 --> 00:03:59,040 Speaker 1: and replaced with useful gear. They will be those that 71 00:03:59,080 --> 00:04:00,760 Speaker 1: are finished, and I think going to be a bunch 72 00:04:00,800 --> 00:04:04,080 Speaker 1: of undone construction. They're going to be powered shells, which 73 00:04:04,080 --> 00:04:07,080 Speaker 1: are just empty buildings with power attached. And really, I 74 00:04:07,120 --> 00:04:10,080 Speaker 1: don't think that's gonna be. That's gonna be across the board. 75 00:04:10,080 --> 00:04:11,880 Speaker 1: I think we're gonna have a lot of failed construction. 76 00:04:12,520 --> 00:04:14,680 Speaker 1: And throughout all of this bullshit, the rest of them, 77 00:04:14,760 --> 00:04:17,680 Speaker 1: Magnificent seven has seen their stock values saw off the 78 00:04:17,680 --> 00:04:20,559 Speaker 1: back of pure hype because they weren't making a dime 79 00:04:20,600 --> 00:04:24,359 Speaker 1: of profit from selling artificial intelligence services, and they've done 80 00:04:24,400 --> 00:04:27,800 Speaker 1: so by it acquiring hundreds of billions of dollars of 81 00:04:27,880 --> 00:04:31,400 Speaker 1: physical assets. The whole point of these fucking companies was 82 00:04:31,440 --> 00:04:35,159 Speaker 1: that they were asset light and cash heavy. Now they're 83 00:04:35,200 --> 00:04:39,240 Speaker 1: full of fucking GPUs to depreciate. This is a fucking calamity. 84 00:04:39,480 --> 00:04:42,640 Speaker 1: I am fucking terrified because everything I've read about the 85 00:04:42,680 --> 00:04:44,960 Speaker 1: past makes me certain that this will be worse. The 86 00:04:45,040 --> 00:04:47,440 Speaker 1: hands will be longer, and nobody seems to want to 87 00:04:47,480 --> 00:04:51,560 Speaker 1: accept the true logical endpoint of my arguments, So I'm 88 00:04:51,600 --> 00:04:54,560 Speaker 1: gonna kind of lay them out depending on how right 89 00:04:54,640 --> 00:04:57,760 Speaker 1: I am. I'm worrying about the following things. And Nvidia 90 00:04:57,800 --> 00:05:00,080 Speaker 1: will lose anywhere between forty percent and night besore and 91 00:05:00,160 --> 00:05:02,280 Speaker 1: of its revenue in a relatively short amount of time, 92 00:05:02,360 --> 00:05:05,239 Speaker 1: a couple of quarters maybe, which will annihilate the value 93 00:05:05,240 --> 00:05:07,200 Speaker 1: of the S and P five hundred because the stock 94 00:05:07,240 --> 00:05:11,360 Speaker 1: will go down because the market will go Ah, Microsoft, 95 00:05:11,400 --> 00:05:15,000 Speaker 1: Google and Amazon will face massive multi billion dollar impairments 96 00:05:15,080 --> 00:05:17,240 Speaker 1: within the next few years. For God to say, meta, 97 00:05:17,279 --> 00:05:20,760 Speaker 1: but they are included too. Now, this one is funny 98 00:05:21,120 --> 00:05:24,880 Speaker 1: but will still be bad. Oracle will potentially face bankruptcy 99 00:05:24,960 --> 00:05:27,479 Speaker 1: or at least a severe horrendous series of layoffs and 100 00:05:27,520 --> 00:05:31,080 Speaker 1: restructuring efforts to try and write the shift after agreeing 101 00:05:31,080 --> 00:05:33,240 Speaker 1: to two hundred and forty eight billion dollars in lease 102 00:05:33,279 --> 00:05:37,000 Speaker 1: obligations and fifty six billion dollars in debt primarily to 103 00:05:37,040 --> 00:05:39,120 Speaker 1: build data centers to get the revenue from its three 104 00:05:39,200 --> 00:05:42,240 Speaker 1: hundred billion dollar five year long deal with open Ai, 105 00:05:42,560 --> 00:05:44,560 Speaker 1: a company that's going to run out of money in 106 00:05:44,600 --> 00:05:47,800 Speaker 1: the next year. Venture Capital is going to take such 107 00:05:47,800 --> 00:05:50,320 Speaker 1: a severe haircut from the AI bubble that it will 108 00:05:50,320 --> 00:05:52,920 Speaker 1: effectively kill any chances of raising a round off the 109 00:05:52,960 --> 00:05:56,880 Speaker 1: Series B for many, many years. I'm really worried about 110 00:05:56,880 --> 00:05:59,160 Speaker 1: the reported one hundred and seventy eight point five billion 111 00:05:59,200 --> 00:06:03,400 Speaker 1: dollars in US date center deals going underwater. I don't 112 00:06:03,440 --> 00:06:04,800 Speaker 1: think they're going to be paid, and it's going to 113 00:06:04,920 --> 00:06:07,320 Speaker 1: lead to massive losses with private equity and banks, which 114 00:06:07,360 --> 00:06:10,040 Speaker 1: will lead to a chill in effectively all debt markets 115 00:06:10,279 --> 00:06:13,720 Speaker 1: and massive job losses across the many many construction projects 116 00:06:13,760 --> 00:06:15,880 Speaker 1: that have started as a result of the AI bubble, 117 00:06:16,440 --> 00:06:18,680 Speaker 1: and if open Ai and Anthropic can actually manage to 118 00:06:18,680 --> 00:06:21,560 Speaker 1: go public. I think every investor is going to lose 119 00:06:21,560 --> 00:06:24,200 Speaker 1: out as they collapse under the weight of their bullshit assumptions. 120 00:06:24,600 --> 00:06:28,560 Speaker 1: Every partner, every company relying on them, every infrastructure partner, 121 00:06:28,800 --> 00:06:31,360 Speaker 1: every AI starter built on their tech will be burned. 122 00:06:32,560 --> 00:06:35,320 Speaker 1: I also believe that we as a society will face 123 00:06:35,360 --> 00:06:38,080 Speaker 1: a violent paradigm shift as the result of the AI 124 00:06:38,120 --> 00:06:40,920 Speaker 1: bubble bursting, because it revealed how many people just don't 125 00:06:41,000 --> 00:06:45,479 Speaker 1: understand stuff. The amount of bosses, influencers, movie stars, politicians, 126 00:06:45,480 --> 00:06:48,120 Speaker 1: and other individuals of note that have fallen behind generative 127 00:06:48,160 --> 00:06:50,280 Speaker 1: AI and claimed again and again that it can do 128 00:06:50,320 --> 00:06:52,680 Speaker 1: things that it can't have effatuated myths that will lead 129 00:06:52,680 --> 00:06:55,000 Speaker 1: to massive ciple harms both day and in the future, 130 00:06:55,160 --> 00:07:12,520 Speaker 1: have revealed themselves to be fucking frauds. It's been obvious 131 00:07:12,560 --> 00:07:15,240 Speaker 1: since twenty twenty three that generative AI flat out does 132 00:07:15,280 --> 00:07:17,640 Speaker 1: not do what many people say it does or will do, 133 00:07:18,000 --> 00:07:21,000 Speaker 1: and obvious since twenty twenty four that the economics didn't work. 134 00:07:21,200 --> 00:07:24,360 Speaker 1: Yet still, every single goddamn day, we are surrounded by 135 00:07:24,360 --> 00:07:27,200 Speaker 1: people screaming in our faces the AI is the future, 136 00:07:27,400 --> 00:07:29,920 Speaker 1: that AI will replace us, that AI is the panacea 137 00:07:29,960 --> 00:07:32,320 Speaker 1: to every fucking problem, even though it only seems to 138 00:07:32,320 --> 00:07:36,040 Speaker 1: create new ones. I also believe that AI psychosis is 139 00:07:36,080 --> 00:07:38,280 Speaker 1: a prevalent thing on a scale that few people want 140 00:07:38,320 --> 00:07:41,160 Speaker 1: to discuss. There's something about the process of using a 141 00:07:41,200 --> 00:07:44,000 Speaker 1: large language model that convinces people that AI is powerful 142 00:07:44,120 --> 00:07:46,240 Speaker 1: because they managed to bunk it on the head enough 143 00:07:46,280 --> 00:07:50,240 Speaker 1: times to make something useful come out seemingly intelligent. People 144 00:07:50,240 --> 00:07:52,880 Speaker 1: will spend hours and hours making clawed code burb out 145 00:07:53,040 --> 00:07:56,880 Speaker 1: quasi useful, quasi functional software, claiming that the productivity gains 146 00:07:56,880 --> 00:07:59,320 Speaker 1: are obvious, even when if you sat down and worked 147 00:07:59,360 --> 00:08:01,040 Speaker 1: out how long it took them and how much the 148 00:08:01,040 --> 00:08:03,080 Speaker 1: API rates for the software would be. If they weren't 149 00:08:03,120 --> 00:08:05,400 Speaker 1: being subsidized, they could probably have just learned to code 150 00:08:05,440 --> 00:08:07,440 Speaker 1: in the first place, or paid someone else to do it. 151 00:08:07,760 --> 00:08:10,400 Speaker 1: And no, please stop this thing of oh it's cheaper 152 00:08:10,400 --> 00:08:12,560 Speaker 1: than hiring a developer. Did you go out and ask 153 00:08:12,560 --> 00:08:14,720 Speaker 1: a developer? Did you actually get a quote? Do you 154 00:08:14,760 --> 00:08:16,400 Speaker 1: actually know what you're talking about? Do you know a 155 00:08:16,440 --> 00:08:20,280 Speaker 1: fucking thing about you're talking about? You see, I think 156 00:08:20,440 --> 00:08:23,680 Speaker 1: large language models that are a global gas lighting experiment, 157 00:08:23,960 --> 00:08:26,880 Speaker 1: where the test is to see whether inefficient and questionably 158 00:08:26,920 --> 00:08:29,880 Speaker 1: functional software can convince you that it works because you 159 00:08:29,960 --> 00:08:33,240 Speaker 1: managed to manipulate its handles with the right prompts, explaining 160 00:08:33,280 --> 00:08:36,560 Speaker 1: away failures as you go, and getting increasingly agitated with 161 00:08:36,559 --> 00:08:38,640 Speaker 1: those who don't agree with you that you're living in 162 00:08:38,679 --> 00:08:42,120 Speaker 1: the future. This series forced me to live in the 163 00:08:42,160 --> 00:08:44,440 Speaker 1: past for hours and hours and hours, and all it 164 00:08:44,520 --> 00:08:47,160 Speaker 1: taught me is that people are desperate to compare this 165 00:08:47,240 --> 00:08:49,640 Speaker 1: era to another as a means of explaining away the 166 00:08:49,720 --> 00:08:52,320 Speaker 1: bad habits and specious ideas of a tech industry run 167 00:08:52,360 --> 00:08:56,600 Speaker 1: by fucking influences and management consoles. By comparison, the dot 168 00:08:56,640 --> 00:09:00,000 Speaker 1: com bubble is relatively sensible, built on good ideas execut 169 00:09:00,200 --> 00:09:02,680 Speaker 1: by greedy people rushing to do things based on ideas 170 00:09:02,679 --> 00:09:06,439 Speaker 1: they didn't understand, and analyzes that were utter bullshit. The 171 00:09:06,559 --> 00:09:09,400 Speaker 1: AI bubble is happening in a remarkable era of digital 172 00:09:09,440 --> 00:09:12,840 Speaker 1: information and connectivity that allows us to process and analyze 173 00:09:12,840 --> 00:09:16,120 Speaker 1: details at scale and with speed, And everybody who inflated 174 00:09:16,160 --> 00:09:18,680 Speaker 1: this era should, and, by my fucking sword, will be 175 00:09:18,760 --> 00:09:22,760 Speaker 1: made to experience inextricable, permanent shame for the horrors that 176 00:09:22,800 --> 00:09:25,680 Speaker 1: follow as a result. And believe me, even if I'm 177 00:09:25,800 --> 00:09:28,400 Speaker 1: half right, things will likely be so much worse than 178 00:09:28,440 --> 00:09:30,840 Speaker 1: when the dot com bubble burst. I don't see a 179 00:09:30,840 --> 00:09:33,480 Speaker 1: scenario where anything survives the collapse of core we've let 180 00:09:33,520 --> 00:09:35,760 Speaker 1: alone the collapse of open AI. I don't see a 181 00:09:35,800 --> 00:09:38,400 Speaker 1: scenario where any more than a few AI starts survive, 182 00:09:38,679 --> 00:09:40,720 Speaker 1: if any and if that happens, there won't be any 183 00:09:40,720 --> 00:09:44,120 Speaker 1: customers for those GPUs, not that anywhere near enough existed 184 00:09:44,120 --> 00:09:46,440 Speaker 1: to fill even a tiny percentile of the demand that 185 00:09:46,480 --> 00:09:50,040 Speaker 1: currently exists, or for the capacity we've already built. On 186 00:09:50,080 --> 00:09:52,680 Speaker 1: some level. I hope I'm wrong. I just don't know 187 00:09:52,720 --> 00:09:56,199 Speaker 1: how anymore. I really don't thank you so much for listening. 188 00:09:56,280 --> 00:09:59,319 Speaker 1: It's been in intense January with this, with CES, with 189 00:09:59,360 --> 00:10:02,760 Speaker 1: the inchittaphone ential crisis. I'm probably going to take the 190 00:10:02,760 --> 00:10:04,920 Speaker 1: week off next week and run a call zone media 191 00:10:05,000 --> 00:10:07,480 Speaker 1: rewind to give my aching brain and lungs a break. 192 00:10:07,800 --> 00:10:11,120 Speaker 1: Thank you so much for listening. Who knows, though, some 193 00:10:11,160 --> 00:10:14,200 Speaker 1: bullshit will probably happen because it's earning season when this runs, 194 00:10:14,559 --> 00:10:16,720 Speaker 1: some crazy shit will happen, or maybe it'll be normal. 195 00:10:16,800 --> 00:10:19,480 Speaker 1: I could just rest, and if that happens, I'll probably 196 00:10:19,520 --> 00:10:23,079 Speaker 1: end up pinking and squawking into the micros. I always though, regardless, 197 00:10:23,120 --> 00:10:25,360 Speaker 1: you'll hear my voice one way or another. Thank you 198 00:10:25,440 --> 00:10:37,600 Speaker 1: so much everyone, This has been bubble week. Thank you 199 00:10:37,640 --> 00:10:39,000 Speaker 1: for listening to Better Offline. 200 00:10:39,120 --> 00:10:41,559 Speaker 2: The editor and composer of the Better Offline theme song 201 00:10:41,640 --> 00:10:44,240 Speaker 2: is Matasowski. You can check out more of his music 202 00:10:44,280 --> 00:10:47,960 Speaker 2: and audio projects at Mattasowski dot com, m A T 203 00:10:47,960 --> 00:10:52,400 Speaker 2: T O S O W s ki dot com. You 204 00:10:52,440 --> 00:10:54,920 Speaker 2: can email me at easy at Better Offline dot com 205 00:10:55,040 --> 00:10:57,319 Speaker 2: or visit Better Offline dot com to find more podcast 206 00:10:57,400 --> 00:11:00,720 Speaker 2: links and of course, my newsletter. I also really recommend 207 00:11:00,760 --> 00:11:02,679 Speaker 2: you go to chat dot Where's Youread dot at to 208 00:11:02,760 --> 00:11:05,160 Speaker 2: visit the discord, and go to our slash. 209 00:11:04,840 --> 00:11:08,000 Speaker 1: Better Offline to check out I'll Reddit. Thank you so 210 00:11:08,080 --> 00:11:11,520 Speaker 1: much for listening. Better Offline is a production of cool 211 00:11:11,600 --> 00:11:12,160 Speaker 1: Zone Media. 212 00:11:12,280 --> 00:11:15,120 Speaker 2: For more from cool Zone Media, visit our website cool 213 00:11:15,200 --> 00:11:18,520 Speaker 1: Zonemedia dot com or check us out on the iHeartRadio app, 214 00:11:18,600 --> 00:11:21,280 Speaker 1: Apple Podcasts, or wherever you get your podcasts.