1 00:00:02,160 --> 00:00:07,039 Speaker 1: Zone Media. Hello and welcome to this week's Better Offline Monologue. 2 00:00:07,040 --> 00:00:17,400 Speaker 1: I'm your host ed ze Trunk. In August of last year, 3 00:00:17,440 --> 00:00:20,240 Speaker 1: I put out a podcast called How the AI Bubble Bursts, 4 00:00:20,239 --> 00:00:21,759 Speaker 1: where I ran down what I thought would be the 5 00:00:21,760 --> 00:00:24,440 Speaker 1: signs that things were collapsing, my pale horses of the 6 00:00:24,480 --> 00:00:27,280 Speaker 1: AI Apocalypse, if you will, which is something that reads 7 00:00:27,320 --> 00:00:29,159 Speaker 1: really well, but when you say it out loud, not 8 00:00:29,280 --> 00:00:31,440 Speaker 1: so good. The reason I put these together is that 9 00:00:31,480 --> 00:00:33,920 Speaker 1: the bubble is unlikely to have one specific moment where 10 00:00:33,960 --> 00:00:36,960 Speaker 1: things explode. Things like burst earns Is collapse during the 11 00:00:36,960 --> 00:00:40,559 Speaker 1: Great Financial Crisis for such significant moments, in part because 12 00:00:40,720 --> 00:00:43,600 Speaker 1: that was a public company. Mortgages were a massive part 13 00:00:43,640 --> 00:00:45,800 Speaker 1: and are a massive part of the economy, and indeed 14 00:00:45,800 --> 00:00:48,240 Speaker 1: there was billions and indeed I think trillions of dollars 15 00:00:48,240 --> 00:00:50,159 Speaker 1: resting on them, and I don't really think there's a 16 00:00:50,159 --> 00:00:52,920 Speaker 1: situation like that, nor is there one where a public 17 00:00:52,960 --> 00:00:56,280 Speaker 1: company collapses as a result of the AI bubble bursting. 18 00:00:56,840 --> 00:00:59,320 Speaker 1: This bubble is different because it's so thoroughly based on 19 00:00:59,760 --> 00:01:02,400 Speaker 1: that games. I'll end up at some point doing a 20 00:01:02,400 --> 00:01:04,480 Speaker 1: longer episode about this, But the long and short of 21 00:01:04,480 --> 00:01:06,840 Speaker 1: it is that the actual core of the system. The 22 00:01:06,880 --> 00:01:10,560 Speaker 1: real weak point is actually in video. In Vidia's continued 23 00:01:10,600 --> 00:01:12,880 Speaker 1: success comes from the reliable quarter of a quarter or 24 00:01:12,880 --> 00:01:15,080 Speaker 1: a year of a year growth from selling GPUs, the 25 00:01:15,120 --> 00:01:18,240 Speaker 1: graphics processing units that power generally of AI and that 26 00:01:18,280 --> 00:01:20,480 Speaker 1: people take and put inside of servers so that they 27 00:01:20,520 --> 00:01:24,000 Speaker 1: can immediately start losing money. More than forty percent of 28 00:01:24,040 --> 00:01:26,640 Speaker 1: their GPU revenue comes from the other six companies in 29 00:01:26,680 --> 00:01:31,080 Speaker 1: the Magnificent Seven, Microsoft, Google, Meta, Tesla, Apple, and Amazon, 30 00:01:31,400 --> 00:01:34,240 Speaker 1: and then VideA. Deeply depends on their continued and growing 31 00:01:34,360 --> 00:01:37,200 Speaker 1: hunger for GPUs. It isn't just enough that they buy 32 00:01:37,240 --> 00:01:40,520 Speaker 1: the same amounts, they must buy more. The rest of 33 00:01:40,560 --> 00:01:43,319 Speaker 1: the mag seven's valuations are dependent on continued growth too, 34 00:01:43,360 --> 00:01:45,640 Speaker 1: which is what AI is meant to provide them. But 35 00:01:45,760 --> 00:01:50,080 Speaker 1: it's not really providing growth at all. Indeed, their growth 36 00:01:50,120 --> 00:01:53,320 Speaker 1: stories are not really based on numbers but vibes. They're 37 00:01:53,360 --> 00:01:57,240 Speaker 1: based on feelings and the general warm, fuzzy feeling the 38 00:01:57,280 --> 00:01:59,720 Speaker 1: dipgits get when they claim that AI is the future. 39 00:02:00,080 --> 00:02:02,360 Speaker 1: Let me give you an example. Metas stock jump because 40 00:02:02,360 --> 00:02:05,440 Speaker 1: they've announced the theoretical data center the size of Manhattan 41 00:02:05,880 --> 00:02:08,760 Speaker 1: now important detail here as well. Meta is making exactly 42 00:02:08,840 --> 00:02:13,799 Speaker 1: minus dollars on generative AI. They're just putting generative AI everywhere, 43 00:02:14,280 --> 00:02:16,880 Speaker 1: and yeah, giving people a little bit of psychosis. I 44 00:02:16,880 --> 00:02:20,160 Speaker 1: imagine filling feeds full of slop. It's very bad. And 45 00:02:20,200 --> 00:02:23,760 Speaker 1: you know what, it also isn't making them money. And 46 00:02:24,040 --> 00:02:26,600 Speaker 1: right now this is really good for the Magnificent seven 47 00:02:26,639 --> 00:02:28,840 Speaker 1: because they're not actually having to prove that AI is 48 00:02:28,840 --> 00:02:31,280 Speaker 1: a big deal. People are basically believing it is because 49 00:02:31,280 --> 00:02:33,880 Speaker 1: the AI trade, which means the value of these stocks, 50 00:02:33,880 --> 00:02:37,320 Speaker 1: remains positive, and honestly, it is currently a positive trade, 51 00:02:37,360 --> 00:02:41,240 Speaker 1: even though it's based on effectively nothing. Seriously, Amazon may 52 00:02:41,280 --> 00:02:44,760 Speaker 1: only make five billion dollars in revenue this year on AI, 53 00:02:44,840 --> 00:02:48,120 Speaker 1: and I mean revenue, not profit. Googled maybe three to 54 00:02:48,200 --> 00:02:51,640 Speaker 1: seven billion, Microsoft thirteen billion dollars, and ten billion dollars 55 00:02:51,680 --> 00:02:54,840 Speaker 1: of that is open AI's compute. These are pathetic numbers. 56 00:02:54,880 --> 00:02:57,400 Speaker 1: But these companies are let off because the AI trade 57 00:02:57,440 --> 00:03:01,519 Speaker 1: still works. The possibilities are still theres that don't exist 58 00:03:01,560 --> 00:03:04,160 Speaker 1: today and don't appear to have any path to them either. 59 00:03:04,360 --> 00:03:08,560 Speaker 1: They're coming, we swear, if we say agents enough time, 60 00:03:08,680 --> 00:03:11,600 Speaker 1: maybe it'll work. The long and short of it is 61 00:03:11,639 --> 00:03:14,560 Speaker 1: that everything about the AI industry is based on some level, 62 00:03:14,560 --> 00:03:17,400 Speaker 1: on lies, on lies both subtle and avert, and the 63 00:03:17,440 --> 00:03:20,000 Speaker 1: general sense that generative AI is both the future and 64 00:03:20,040 --> 00:03:22,440 Speaker 1: the future of the economy. There are a few real 65 00:03:22,520 --> 00:03:25,160 Speaker 1: tangible things to point at other than chat GPT's five 66 00:03:25,240 --> 00:03:27,480 Speaker 1: hundred million weekly users and the tens of billions of 67 00:03:27,520 --> 00:03:31,720 Speaker 1: dollars invested in AI startups, or maybe the weird acquisition 68 00:03:31,840 --> 00:03:34,400 Speaker 1: deals like the two point four billion dollars that Google 69 00:03:34,520 --> 00:03:38,920 Speaker 1: just gave kind of AI code editor Windsurf, except Google 70 00:03:39,040 --> 00:03:42,240 Speaker 1: took the staff, some of the staff the C suite, 71 00:03:42,440 --> 00:03:44,600 Speaker 1: the R and D team for the two point four billion, 72 00:03:44,640 --> 00:03:48,440 Speaker 1: and then Cognition, who makes the Devin coding pop took 73 00:03:48,480 --> 00:03:51,960 Speaker 1: the rest for an undisclosed sum. It's all really, really 74 00:03:51,960 --> 00:03:55,040 Speaker 1: fucking weird. It's all very strange, and yet when you 75 00:03:55,080 --> 00:03:58,720 Speaker 1: look beneath the surface, there really are few returns. Windsurfs 76 00:03:58,720 --> 00:04:01,960 Speaker 1: monthly revenue was six point eight brillion, three million dollars 77 00:04:02,000 --> 00:04:04,560 Speaker 1: a month, a ridiculously small amount, working out to about 78 00:04:04,560 --> 00:04:07,520 Speaker 1: eighty two million dollars of annualized revenue. Yet based on 79 00:04:07,560 --> 00:04:10,440 Speaker 1: the Information's Generative AI data base, Windsurf was actually one 80 00:04:10,480 --> 00:04:13,240 Speaker 1: of the highest earning AI startups out there no really 81 00:04:13,280 --> 00:04:16,800 Speaker 1: based on annualized revenue meaning month multiplied by twelve, most 82 00:04:16,800 --> 00:04:20,599 Speaker 1: Ai companies make about one hundred million dollars annualized or less. 83 00:04:20,720 --> 00:04:23,159 Speaker 1: Not really, that's about eight point three million dollars a 84 00:04:23,240 --> 00:04:27,240 Speaker 1: year a month with since there's the Weka Xai together 85 00:04:27,320 --> 00:04:30,440 Speaker 1: a bridge Glean perplexity, and they're all making about one 86 00:04:30,480 --> 00:04:33,560 Speaker 1: hundred million dollars annualized. And I must be clear about 87 00:04:33,600 --> 00:04:36,240 Speaker 1: how weird this status. Annialize just means the month of 88 00:04:36,279 --> 00:04:41,760 Speaker 1: revenue times twelve. Customers churn. Customers churn, so your arr 89 00:04:41,920 --> 00:04:44,159 Speaker 1: may go down when you have a few people drop out, 90 00:04:44,320 --> 00:04:48,400 Speaker 1: it may spike arbitrarily because you've increased prices, and it's 91 00:04:48,440 --> 00:04:51,040 Speaker 1: just it's a very bad statistic to use. Yet people 92 00:04:51,640 --> 00:04:53,359 Speaker 1: are not really looking at the other thing, which is 93 00:04:54,040 --> 00:04:56,080 Speaker 1: that's not a lot of money. Eight point three million 94 00:04:56,279 --> 00:04:59,360 Speaker 1: dollars a month. You have ad agencies that make several 95 00:04:59,400 --> 00:05:03,239 Speaker 1: times that. Ad agency's bank ads on Facebook and Google. 96 00:05:04,040 --> 00:05:08,159 Speaker 1: These are terrible fucking companies. I mean outside of one 97 00:05:08,240 --> 00:05:11,320 Speaker 1: hundred million ARR side, you've got mid Journey, who are 98 00:05:11,440 --> 00:05:14,760 Speaker 1: inexplicably profitable. We'll look into that, and Neo four j 99 00:05:14,960 --> 00:05:17,479 Speaker 1: and they make two hundred million dollars a month, sorry, 100 00:05:17,560 --> 00:05:20,279 Speaker 1: two one hundred million dollars annualized, so about sixteen point 101 00:05:20,320 --> 00:05:24,320 Speaker 1: six million dollars, which is again not amazing. None of 102 00:05:24,320 --> 00:05:27,800 Speaker 1: these are amazing. These aren't great. These would be okay 103 00:05:27,839 --> 00:05:31,039 Speaker 1: startups if they were profitable, and they're absolutely not. But 104 00:05:31,160 --> 00:05:33,960 Speaker 1: right on the tippy top behind Anthropic and open AI 105 00:05:34,160 --> 00:05:36,400 Speaker 1: is any sphere who makes Cursor and they make five 106 00:05:36,480 --> 00:05:39,160 Speaker 1: hundred million dollars annualized, or about forty one point six 107 00:05:39,240 --> 00:05:43,760 Speaker 1: million dollars a month, which is bad because in June 108 00:05:44,440 --> 00:05:48,000 Speaker 1: all of their prices got increased by Anthropic, and on 109 00:05:48,080 --> 00:05:50,479 Speaker 1: top of that they had to change the pricing and 110 00:05:50,520 --> 00:05:53,080 Speaker 1: now Cursor kind of sucks. Go on their subredd it 111 00:05:53,200 --> 00:05:56,200 Speaker 1: check it out anyway. This is why you don't like 112 00:05:56,200 --> 00:05:58,160 Speaker 1: annualized as a stat by the way, because I reckon 113 00:05:58,200 --> 00:06:01,000 Speaker 1: their July revenue might be a little bit lower than 114 00:06:01,040 --> 00:06:03,160 Speaker 1: forty one point six million dollars, but we'll never know 115 00:06:03,320 --> 00:06:06,240 Speaker 1: because well unless someone reports there. If you've got any 116 00:06:06,360 --> 00:06:09,200 Speaker 1: numbers on any of these companies, my signal is eas 117 00:06:09,279 --> 00:06:14,839 Speaker 1: it tron ezit ron dot seventy six on signal. Please 118 00:06:14,880 --> 00:06:18,440 Speaker 1: send me anything you have. I love numbers anyway, past 119 00:06:18,480 --> 00:06:21,160 Speaker 1: that point, it's only anthropic and open AI. And I've 120 00:06:21,240 --> 00:06:23,480 Speaker 1: left out a few companies like Touring and Scale they 121 00:06:23,480 --> 00:06:26,560 Speaker 1: don't make AI products their consultancies. But the long and 122 00:06:26,560 --> 00:06:29,279 Speaker 1: short of it is, this industry doesn't make much fucking money. 123 00:06:29,520 --> 00:06:32,000 Speaker 1: The actual returns from it aren't that very good. It's 124 00:06:32,000 --> 00:06:34,800 Speaker 1: not very good at all. But the reason I'm telling 125 00:06:34,839 --> 00:06:36,400 Speaker 1: you all these numbers is that I need you to 126 00:06:36,440 --> 00:06:38,200 Speaker 1: know that this industry is not held up by the 127 00:06:38,279 --> 00:06:42,000 Speaker 1: actual businesses or their revenues. None of these companies are profitable, 128 00:06:42,120 --> 00:06:45,520 Speaker 1: and it certainly isn't held up by the actual products themselves. 129 00:06:46,160 --> 00:06:48,720 Speaker 1: And I'm sure you've got an AI friend who's come 130 00:06:48,760 --> 00:06:50,080 Speaker 1: to you and said, oh, I use it for this. 131 00:06:51,160 --> 00:06:53,479 Speaker 1: See how far you get them in the conversation before 132 00:06:53,480 --> 00:06:55,840 Speaker 1: they just say they're using it like a search engine. 133 00:06:56,080 --> 00:06:58,880 Speaker 1: We had this whole thing with Brian on the podcast 134 00:06:58,920 --> 00:07:01,760 Speaker 1: this week. It's like, yeah, really emphatically talking about the 135 00:07:01,760 --> 00:07:04,039 Speaker 1: fact that you've got a slightly sexier search engine that 136 00:07:04,040 --> 00:07:06,280 Speaker 1: may or may not get things right, that can fool 137 00:07:06,320 --> 00:07:09,279 Speaker 1: you into believing you're correct about something which isn't a 138 00:07:09,320 --> 00:07:12,160 Speaker 1: fucking product. Now, if it was, people will be making money. 139 00:07:12,160 --> 00:07:16,800 Speaker 1: It's ridiculous. But even in areas like coding, coding the 140 00:07:16,840 --> 00:07:19,600 Speaker 1: one place where people can point and go look look 141 00:07:19,760 --> 00:07:23,560 Speaker 1: users like this, look edward, look at them. A study 142 00:07:23,600 --> 00:07:26,560 Speaker 1: from nonprofit group Model Evaluation and Throat Research found that 143 00:07:26,600 --> 00:07:29,800 Speaker 1: despite developers believing AI coding tools sped them up by 144 00:07:29,800 --> 00:07:33,120 Speaker 1: twenty four percent, coding tools actually increase the completion time 145 00:07:33,120 --> 00:07:36,600 Speaker 1: of their tasks by nineteen percent. It actually it slowed 146 00:07:36,640 --> 00:07:40,160 Speaker 1: them down. Oh my god. Every week with this industry, 147 00:07:40,200 --> 00:07:42,920 Speaker 1: I feel like I'm going insane. It's like every week 148 00:07:42,960 --> 00:07:46,160 Speaker 1: there's this goddamn story like this. I still have people 149 00:07:46,160 --> 00:07:48,680 Speaker 1: who tell me it's the future. I feel like I'm 150 00:07:48,720 --> 00:07:52,120 Speaker 1: going insane. But as a result of AI being a 151 00:07:52,160 --> 00:07:55,040 Speaker 1: nearly entirely vibe spaced economy, the pale horses of its 152 00:07:55,080 --> 00:07:58,600 Speaker 1: apocalypse will be vibe shifters rather than defined events that 153 00:07:58,640 --> 00:08:02,000 Speaker 1: would logically change things. Remember Deep Seek, that was a 154 00:08:02,040 --> 00:08:05,200 Speaker 1: clunky story. So you had a Chinese model developer that 155 00:08:05,320 --> 00:08:08,200 Speaker 1: may be cheaper, it was cheaper to train, much cheaper 156 00:08:08,240 --> 00:08:11,600 Speaker 1: to train, but still cheaper to train, and their model 157 00:08:11,680 --> 00:08:14,920 Speaker 1: might be cheaper, and it is comparable. I mean, I've 158 00:08:14,960 --> 00:08:17,160 Speaker 1: certainly made my hay with the episodes and interviews that 159 00:08:17,200 --> 00:08:19,760 Speaker 1: did around it, but I surprise that shocked the market. 160 00:08:19,840 --> 00:08:22,840 Speaker 1: There are bigger, nastiest stories. But if that can do 161 00:08:22,960 --> 00:08:27,560 Speaker 1: it well, basically anything could one thing that really spreads 162 00:08:27,640 --> 00:08:31,200 Speaker 1: like information poison into the heads of business idiot's everywhere 163 00:08:31,200 --> 00:08:34,360 Speaker 1: who are saying AI now please without really understanding what 164 00:08:34,480 --> 00:08:37,160 Speaker 1: it does. It's unlikely to be one big pop but 165 00:08:37,280 --> 00:08:40,600 Speaker 1: several little doots from the bubble deflating. So let's get 166 00:08:40,640 --> 00:08:42,160 Speaker 1: to it. Here are the science to look out for, 167 00:08:42,480 --> 00:08:44,920 Speaker 1: freshly updated, and if you have any ideas, please do 168 00:08:45,040 --> 00:08:47,319 Speaker 1: reach out. I'd love to hear from you. We'll start 169 00:08:47,320 --> 00:08:50,240 Speaker 1: with an obvious one. Any price increases or decreases are 170 00:08:50,240 --> 00:08:53,920 Speaker 1: open AI and Anthropic. I called these increases in August, 171 00:08:53,960 --> 00:08:56,120 Speaker 1: and they had decreases because, as I've mentioned in the 172 00:08:56,120 --> 00:08:58,600 Speaker 1: subprime AI crisis, there is now a race to the 173 00:08:58,600 --> 00:09:01,520 Speaker 1: boond for model developers. Open Ai cut the price of 174 00:09:01,559 --> 00:09:03,880 Speaker 1: their own through Reasoning model by eight percent in June 175 00:09:03,880 --> 00:09:06,520 Speaker 1: in an attempt to undercut Anthropics claud for Opus, which 176 00:09:06,600 --> 00:09:11,000 Speaker 1: launched May twenty second, I believe, and we've already begun 177 00:09:11,040 --> 00:09:14,240 Speaker 1: to see price increases too. Anthropic added service theres and 178 00:09:14,280 --> 00:09:17,240 Speaker 1: open ai priority processing for enterprise customers to give them 179 00:09:17,320 --> 00:09:20,160 Speaker 1: uninterrupted service in June. And I must be clear when 180 00:09:20,200 --> 00:09:22,760 Speaker 1: I say enterprise, I mean any AI company that has 181 00:09:22,840 --> 00:09:27,360 Speaker 1: customers at scale. So any Sphere, Cursor, Glean and the like, 182 00:09:27,440 --> 00:09:30,440 Speaker 1: they are enterprise customers and they are very likely having 183 00:09:30,440 --> 00:09:33,960 Speaker 1: to pay these prices. Now here's another one. Anthropic and 184 00:09:34,000 --> 00:09:37,720 Speaker 1: open Ai may move away from monthly subscriptions, and indeed 185 00:09:37,800 --> 00:09:41,280 Speaker 1: ANYAI company may do this. This is a huge payer 186 00:09:41,400 --> 00:09:45,640 Speaker 1: horse when companies start moving away from simple business models, 187 00:09:45,679 --> 00:09:48,679 Speaker 1: so twenty bucks one hundred bucks a month, or they 188 00:09:48,679 --> 00:09:51,720 Speaker 1: start changing those so the returns on a sort of 189 00:09:51,720 --> 00:09:55,160 Speaker 1: returns the actual usership isn't obvious. Isn't obvious how much 190 00:09:55,200 --> 00:09:57,440 Speaker 1: you can or can't use. That's a sign that the 191 00:09:57,440 --> 00:10:00,160 Speaker 1: company is trying to contain costs and indeed fuck with 192 00:10:00,200 --> 00:10:01,960 Speaker 1: their customers a bit. And we've already seen a little 193 00:10:01,960 --> 00:10:04,520 Speaker 1: bit of this with how Cursor changed their pricing or 194 00:10:04,640 --> 00:10:06,960 Speaker 1: Claude Max the one hundred dollar or two hundred dollars 195 00:10:06,960 --> 00:10:10,160 Speaker 1: a month subscription from Anthropic, which has now changed things 196 00:10:10,160 --> 00:10:12,839 Speaker 1: to say you get more usage at the higher tiers 197 00:10:12,880 --> 00:10:15,720 Speaker 1: than the twenty dollars a month pro subscription. How much 198 00:10:15,760 --> 00:10:19,040 Speaker 1: more it isn't clear. Go to Claude subreadit. You'll see 199 00:10:19,040 --> 00:10:21,160 Speaker 1: a lot of people who are constantly saying, I don't 200 00:10:21,200 --> 00:10:24,000 Speaker 1: know what the rate limits are, and that's working as 201 00:10:24,040 --> 00:10:27,560 Speaker 1: intended by the way Anthropic is tweaking things very clearly 202 00:10:27,600 --> 00:10:29,760 Speaker 1: on the back end, we do not know what they are. 203 00:10:29,800 --> 00:10:32,760 Speaker 1: There are rumors that they are quantizing models, so making 204 00:10:32,840 --> 00:10:36,960 Speaker 1: them smaller, dumber, cheaper to run, but still doing comparable things, 205 00:10:37,360 --> 00:10:40,679 Speaker 1: or just not having as much access within peak hours. 206 00:10:40,800 --> 00:10:43,679 Speaker 1: These are unsubstantiated rumors, but I've seen enough claims of 207 00:10:43,720 --> 00:10:46,840 Speaker 1: them that they're worth mentioning. These are signs that things 208 00:10:46,920 --> 00:10:49,800 Speaker 1: are getting bad, or that they need to make things 209 00:10:49,840 --> 00:10:53,840 Speaker 1: more profitable. And really, any and all cost increases on 210 00:10:53,880 --> 00:10:56,640 Speaker 1: consumers by any AI service are a pay or horse, 211 00:10:56,640 --> 00:10:58,640 Speaker 1: as they suggest the company in question is trying to 212 00:10:58,679 --> 00:11:02,240 Speaker 1: deal with these ruinous costs. Remember, there are no profitable 213 00:11:02,240 --> 00:11:05,480 Speaker 1: AI companies. It's fucking crazy. We are three years into 214 00:11:05,520 --> 00:11:10,079 Speaker 1: this dog shit and there's none of them. Okay, surge AI. 215 00:11:10,559 --> 00:11:14,240 Speaker 1: They are an AI data training company. That does not matter. 216 00:11:14,679 --> 00:11:16,480 Speaker 1: It does not matter. If you've got a car that 217 00:11:16,480 --> 00:11:19,079 Speaker 1: has a horrible fuel economy. You don't point out how 218 00:11:19,120 --> 00:11:22,400 Speaker 1: well oil is selling to say that the car is good. 219 00:11:22,800 --> 00:11:25,920 Speaker 1: The different things anyway, Any and all rate limits that 220 00:11:26,040 --> 00:11:29,800 Speaker 1: any AI startup are a huge pay or horse because 221 00:11:29,840 --> 00:11:32,240 Speaker 1: these companies are spending a shit ton of money and 222 00:11:32,280 --> 00:11:35,839 Speaker 1: they're having to run these massive infrastructural operations to keep 223 00:11:35,880 --> 00:11:40,120 Speaker 1: things going. Those things are expensive. Power in particular is expensive, 224 00:11:40,160 --> 00:11:43,640 Speaker 1: so any mediating those costs is necessary. If you see 225 00:11:43,679 --> 00:11:46,320 Speaker 1: it with other companies, it's even worse because it means 226 00:11:46,360 --> 00:11:49,600 Speaker 1: that open ai and Anthropic are turning the screws on them. Now. 227 00:11:49,600 --> 00:11:52,280 Speaker 1: Another obvious one, of course, is any funding of revenue 228 00:11:52,280 --> 00:11:55,920 Speaker 1: problems from the remaining AI software companies Glean, Harvey, Alpha Sense, 229 00:11:55,960 --> 00:11:58,559 Speaker 1: any sphere who makes Cursor as I mentioned, or replet 230 00:11:58,640 --> 00:12:01,240 Speaker 1: for example. It's just the funding required to keep these 231 00:12:01,240 --> 00:12:03,760 Speaker 1: companies going is running out, and their business models aren't working. 232 00:12:03,920 --> 00:12:06,040 Speaker 1: And I want to make one real clear point here. 233 00:12:06,040 --> 00:12:09,800 Speaker 1: I'm kind of repeating myself, but it's necessary. These companies 234 00:12:09,880 --> 00:12:14,040 Speaker 1: are symbolic of whether generative AI will work. I realized 235 00:12:14,120 --> 00:12:16,480 Speaker 1: chet GPT is the big sexy one. But if you 236 00:12:16,600 --> 00:12:19,720 Speaker 1: can't build software on GENERATIVEAI and sell it for a profit, 237 00:12:20,000 --> 00:12:24,120 Speaker 1: there is no industry here. Every single major cloud computing 238 00:12:24,200 --> 00:12:28,680 Speaker 1: revolution has started because there is profit, and fucking hell, 239 00:12:28,760 --> 00:12:31,680 Speaker 1: it's nothing like Amazon Web Services when it grew. I'm 240 00:12:31,760 --> 00:12:33,960 Speaker 1: tired of hearing this point. I think I have to 241 00:12:34,000 --> 00:12:36,720 Speaker 1: do a thing about it because I'm fucking sick of it. 242 00:12:36,720 --> 00:12:40,360 Speaker 1: It's nothing to do with it, nothing, It's not remotely similar. 243 00:12:40,440 --> 00:12:43,320 Speaker 1: The only similar thing is that goddamn data centers. God 244 00:12:44,760 --> 00:12:49,160 Speaker 1: calming down. But on the subject to data centers, any 245 00:12:49,240 --> 00:12:53,000 Speaker 1: delays to Stargate or Amazon and Anthropics new Carlisle, Indiana 246 00:12:53,080 --> 00:12:56,880 Speaker 1: build out, both Anthropic and OpenAI are running into capacity issues. 247 00:12:56,920 --> 00:12:59,640 Speaker 1: This is documented well both by the companies and reporting 248 00:12:59,679 --> 00:13:02,760 Speaker 1: from outs like the Information. And they need these projects 249 00:13:02,840 --> 00:13:06,600 Speaker 1: to happen because right now it's very obvious Microsoft is 250 00:13:06,600 --> 00:13:09,800 Speaker 1: not going to expand their infrastructure for Open AI and 251 00:13:09,840 --> 00:13:12,720 Speaker 1: Anthropic well, they need Amazon to do it. But data 252 00:13:12,760 --> 00:13:15,680 Speaker 1: centers don't grow like weeds. They take forever to build. 253 00:13:16,000 --> 00:13:18,560 Speaker 1: I don't even know when the Carlysle Indiana data center 254 00:13:18,640 --> 00:13:21,720 Speaker 1: is going to get built. Fucking hell, what a mess. 255 00:13:21,920 --> 00:13:24,000 Speaker 1: Every reason I must think about this, it's such a mess, 256 00:13:24,280 --> 00:13:26,400 Speaker 1: all right. One final pale horse, and this is my 257 00:13:26,480 --> 00:13:29,839 Speaker 1: favorite one. Anything that happens with soft Bank and their 258 00:13:29,880 --> 00:13:32,960 Speaker 1: ability to raise money is a big pale horse the 259 00:13:33,000 --> 00:13:35,720 Speaker 1: paylist of them all. Actually, because soft Bank is critical 260 00:13:35,760 --> 00:13:38,839 Speaker 1: to everything. Soft Bank is the money behind open Ai. 261 00:13:39,440 --> 00:13:42,520 Speaker 1: Soft Bank is the money behind Stargate. Soft Bank is 262 00:13:42,800 --> 00:13:46,920 Speaker 1: financially responsible for Stargate, by the way, And if open 263 00:13:46,960 --> 00:13:51,480 Speaker 1: Ai doesn't move in to Stargate, and indeed if Stargate 264 00:13:51,520 --> 00:13:54,640 Speaker 1: never happens, Oracle gets left with forty billion dollars of 265 00:13:54,679 --> 00:13:56,960 Speaker 1: chips unused and Cruso, who they have to pay a 266 00:13:57,000 --> 00:14:00,000 Speaker 1: billion dollars I think, have a fifteen year lease with them. 267 00:14:00,400 --> 00:14:03,200 Speaker 1: What a mess. But as I've said, the actual bubble 268 00:14:03,200 --> 00:14:05,920 Speaker 1: bursting will be slow, annoying and unsatisfying at times. But 269 00:14:05,960 --> 00:14:08,400 Speaker 1: these pale horses are the signs that things are falling apart. 270 00:14:08,600 --> 00:14:10,000 Speaker 1: I will, as ever keep you up to date on 271 00:14:10,040 --> 00:14:12,920 Speaker 1: when these signs are happening, what they mean, why you 272 00:14:12,960 --> 00:14:15,760 Speaker 1: should care. I'll probably say fucking shit a few times 273 00:14:15,760 --> 00:14:18,600 Speaker 1: when I do, so, really appreciate you, love you all, 274 00:14:18,679 --> 00:14:19,480 Speaker 1: Thanks for listening.