WEBVTT - The AI Money Trap, Part One 0:00:02.440 --> 0:00:08.240 Also media, Hello and welcome to Better Offline. I'm, of 0:00:08.280 --> 0:00:21.960 course your host ed zitron. As ever, go to the 0:00:22.000 --> 0:00:24.960 episode notes and buy some merchandise. We have the Beautiful 0:00:25.040 --> 0:00:28.440 challenge coin available until August twenty four. I want to say, 0:00:28.640 --> 0:00:31.000 subscribe to my newsletter or send me a threatening email. 0:00:31.040 --> 0:00:32.879 I love hearing from you all, and I also love 0:00:32.920 --> 0:00:34.840 to see that our subreddit has gone over twelve thousand 0:00:34.920 --> 0:00:38.760 vagabonds and scallywags. I'm so proud of our community. Anyway, 0:00:38.800 --> 0:00:40.440 this week's going to be a fun one and two 0:00:40.440 --> 0:00:42.839 part of plus a special report on GPT five, and 0:00:42.880 --> 0:00:44.720 it's kind of a culmination of the last year and 0:00:44.760 --> 0:00:48.080 a half of work that I've done covering the AI bubble. Now, 0:00:48.120 --> 0:00:51.000 the reason I return to this subject so often is 0:00:51.000 --> 0:00:53.120 because each week I've become more and more convinced that 0:00:53.159 --> 0:00:55.880 the only way the AI bubble ends is well badly. 0:00:56.400 --> 0:00:58.279 Every week I see more and more evidence that this 0:00:58.320 --> 0:01:00.640 is structured in a way that is doomed when not 0:01:00.720 --> 0:01:04.320 even abruptly, but they were going to be abrupt little farts, 0:01:04.400 --> 0:01:07.440 little burps as the bubble begins letting out air, and 0:01:07.480 --> 0:01:09.920 in a way where I see people in innovation as 0:01:09.959 --> 0:01:13.600 the collateral damage in the global financial crisis. The Federal 0:01:13.640 --> 0:01:15.920 Reserve and the US government, as well as other equivalents 0:01:15.920 --> 0:01:18.040 around the world, work to ensure that when a bank 0:01:18.120 --> 0:01:20.559 or hedge FuMB went under or a big insurance company 0:01:20.560 --> 0:01:23.200 became insolvent, the collapse was managed in a way that 0:01:23.240 --> 0:01:26.959 didn't implode the entire financial system. The difference with AI 0:01:27.040 --> 0:01:29.400 is first that it's nowhere here as essential as financial 0:01:29.400 --> 0:01:32.000 services to the global economy and thus unlikely to be 0:01:32.040 --> 0:01:34.600 saved as AIG was in two thousand and eight, but 0:01:34.640 --> 0:01:38.160 also that there's no real way for this to end gracefully, 0:01:38.840 --> 0:01:40.880 just due to the way all of this is happening, 0:01:40.920 --> 0:01:44.120 how it's built. I'll get into it, and all the 0:01:44.160 --> 0:01:46.640 signs are now there, And I have been hesitant to 0:01:46.640 --> 0:01:49.560 make this call the whole time. By the way, where well, 0:01:49.600 --> 0:01:53.000 there's a contagion here, where one precarious company falls, the 0:01:53.000 --> 0:01:56.000 rest eventually tumble, sparking a chain of events that will 0:01:56.000 --> 0:02:00.760 be well, pretty bad. But how does this end? At 0:02:00.760 --> 0:02:02.800 the start of August, we saw no less than three 0:02:02.800 --> 0:02:04.760 different pieces, two in the Wall Street Journal and one 0:02:04.760 --> 0:02:09.560 in the Financial Times, Financial Nacial Times, Who Cares, published 0:02:09.560 --> 0:02:11.680 within the span of one week, all asking the same 0:02:11.760 --> 0:02:14.880 question whether the massive proliferation of data centers is a 0:02:14.880 --> 0:02:18.239 massive bubble, and honestly, since then there have been several more. 0:02:18.840 --> 0:02:21.240 Although these publications at times seem to have taken the 0:02:21.240 --> 0:02:24.640 default position of AI's inevitable value, they've begun to sour 0:02:24.680 --> 0:02:26.760 on the idea that it's going to happen anytime soon. 0:02:27.160 --> 0:02:30.040 To be clear, these are sensible, well respected business newspapers, 0:02:30.080 --> 0:02:34.840 hardly partisan voices from the AI wars. Separately, CNBC reported 0:02:34.840 --> 0:02:36.920 that quirked up three hundred cores and open ai is 0:02:36.960 --> 0:02:39.320 either raised or is about to raise another eight point 0:02:39.400 --> 0:02:42.760 three billion dollars in cash, less than two months since 0:02:42.760 --> 0:02:45.760 it raised ten billion dollars from SoftBank and a selection 0:02:45.840 --> 0:02:50.000 of other venture catital firms. While this sounds ambiguous, this 0:02:50.080 --> 0:02:52.600 reporting only claims that it has secured that funding, not 0:02:52.639 --> 0:02:55.440 that it's received it, which is an important distinction. I've 0:02:55.480 --> 0:02:57.760 pointed out in the past that secured funding often comes 0:02:57.760 --> 0:03:00.560 with its own conditions attached. On toime top of this, 0:03:00.960 --> 0:03:03.160 I should add that there is currently a rumor that 0:03:03.280 --> 0:03:06.240 open ai insiders, so the people working at the company, 0:03:06.400 --> 0:03:08.480 are selling six billion dollars with stock at a five 0:03:08.560 --> 0:03:11.960 hundred billion dollar valuation that's around the price of Netflix. 0:03:12.000 --> 0:03:14.600 This is fucking stupid. It's so stupid, and every time 0:03:14.639 --> 0:03:16.959 I read about it, I feel a little insane. But 0:03:17.960 --> 0:03:20.320 it's also weird that you've got people like soft Bank 0:03:20.360 --> 0:03:23.239 and Dragonea who invested in the eight point through billion 0:03:23.280 --> 0:03:26.400 dollar round, who are also buying the stock. All this 0:03:26.480 --> 0:03:29.120 sounds stinky and I hate to be crude. I hate 0:03:29.120 --> 0:03:30.679 to be too crude here, But where the fuck is 0:03:30.720 --> 0:03:33.920 all this money going. Is open ai just incinerating capital? 0:03:34.160 --> 0:03:37.000 Is it compute? Is it salaries? Is it more compute? 0:03:37.040 --> 0:03:40.480 Is it data centers? Because soft Bank isn't actually building 0:03:40.040 --> 0:03:44.520 anything for stargate It's all okay, I calm down a 0:03:44.560 --> 0:03:47.440 little bit, but it's all getting a little bit silly now. 0:03:47.480 --> 0:03:49.520 The information, which has some of the best sourcing in 0:03:49.560 --> 0:03:52.600 any publication covering the AI bubble, i'd argue, though I've 0:03:52.600 --> 0:03:55.000 taken issues with their reporting in the past, suggested that 0:03:55.080 --> 0:03:57.640 open ai intends to use this money to build data centers, 0:03:57.800 --> 0:04:01.160 possibly the only worst investment it can make than general ivai, 0:04:01.560 --> 0:04:03.960 and it's the one that open ai can't really avoid 0:04:04.000 --> 0:04:07.280 because they're also somehow running out of compute two and 0:04:07.320 --> 0:04:10.440 amongst this already ridiculous situation. Since the issue of open 0:04:10.440 --> 0:04:13.160 Ai and Anthropic's actual revenues, which I wrote about my 0:04:13.200 --> 0:04:16.360 premium newsletter on the first of August Please give Me Money, 0:04:16.640 --> 0:04:19.000 and have roughly estimated to be five point twenty six 0:04:19.040 --> 0:04:21.000 billion dollars in the case of open Ai and one 0:04:21.000 --> 0:04:24.680 point five billion dollars in Anthropic as of July. Now, 0:04:24.800 --> 0:04:26.799 to be clear, there are very different numbers going around. 0:04:27.080 --> 0:04:29.560 Get to in a second. In any case, these estimates 0:04:29.560 --> 0:04:32.080 were made based on both companies predilection for leaking their 0:04:32.120 --> 0:04:37.240 annualized revenue or month times twelve. Now, this extremely annoying 0:04:37.400 --> 0:04:39.800 term annualized or recurring revenue is one that I keep 0:04:39.800 --> 0:04:42.039 bringing up because it's become the de facto wait for 0:04:42.080 --> 0:04:45.360 GENERATIVEAI companies to express their revenue, and both open ai 0:04:45.360 --> 0:04:47.760 and Anthropic leaking them intentionally and doing so in a 0:04:47.760 --> 0:04:50.640 way that suggests that they're not even using the traditional 0:04:50.680 --> 0:04:53.880 ways of calculating them. Because arr isn't in and of 0:04:53.920 --> 0:04:56.920 itself an evil thing, it's fairly stand up within software 0:04:56.920 --> 0:05:00.960 as a service companies, except they run completely different to 0:05:01.040 --> 0:05:05.760 open ai and Anthropic anyway. Open ai leaked on July thirtieth, 0:05:05.760 --> 0:05:07.920 twenty twenty five, that it was at twelve billion dollars 0:05:07.960 --> 0:05:10.560 of annualized revenue, so it earned around eight hundred and 0:05:10.600 --> 0:05:13.880 thirty three million dollars in some sort of thirty day period, 0:05:14.080 --> 0:05:17.040 and then two days later, on August first, twenty twenty five, 0:05:17.400 --> 0:05:19.719 the New York Times reported it was at thirteen billion 0:05:19.760 --> 0:05:23.719 dollars annualized, at one point eight billion dollars of monthly revenue. 0:05:24.600 --> 0:05:27.600 It's taking the pierce a little bit. It's very clear 0:05:27.680 --> 0:05:30.560 open AI is not talking in actual calendar months, at 0:05:30.560 --> 0:05:33.400 which point we can assume they're using like a trailing 0:05:33.480 --> 0:05:35.960 thirty day window, as in the month is just thirty days, 0:05:36.040 --> 0:05:39.280 rather than a calendar month like May or June or July. 0:05:39.880 --> 0:05:43.600 We can, however, declaratively say that it's not saying the 0:05:43.640 --> 0:05:45.719 month of June or the month of July was twelve 0:05:45.880 --> 0:05:48.479 or thirteen billion dollars annualized, because if it was, they 0:05:48.520 --> 0:05:51.080 wouldn't have given two vastly different goddamn numbers in the 0:05:51.080 --> 0:05:54.000 same two day period. It doesn't make any sense. And 0:05:54.040 --> 0:05:55.840 to be clear, while I can't say for certain, I 0:05:55.839 --> 0:05:58.840 believe these leaks are deliberate. Open AI is timing matches 0:05:58.880 --> 0:06:02.720 exactly with the fundraise. Similarly, on Anthropic side, these revenues 0:06:02.720 --> 0:06:05.920 are beginning to get really, really weird. Anthropic went from 0:06:05.920 --> 0:06:08.760 making around seventy two million dollars eight hundred and seventy 0:06:08.760 --> 0:06:11.920 five million dollars annualized in January to four hundred and 0:06:11.960 --> 0:06:14.880 thirty three million dollars in monthly revenue in July, or 0:06:14.880 --> 0:06:17.120 at least it leaked on July first, twenty twenty five 0:06:17.160 --> 0:06:19.520 that it was at four billion annuallyzed to the Information, 0:06:19.800 --> 0:06:21.880 which is three hundred and thirty three million dollars, and 0:06:21.920 --> 0:06:25.160 then claimed it had reached five billion annualized four hundred 0:06:25.160 --> 0:06:28.760 and sixteen million to Bloomberg on July twenty ninth, twenty 0:06:28.760 --> 0:06:33.840 twenty five. I'm so tired of this. Something stinky is happening. 0:06:34.200 --> 0:06:36.919 The people who will not admit something stinky is happening, 0:06:37.080 --> 0:06:40.320 I'm beginning to question their sanity or whether they're just 0:06:40.400 --> 0:06:43.040 inherently corrupt in some way. And you can be corrupt 0:06:43.080 --> 0:06:45.640 without money, you can just do it for your mates. Anyway. 0:06:46.720 --> 0:06:49.560 How did Anthropic get there, I'm guessing it was from 0:06:49.600 --> 0:06:52.599 cranking up prices on Cursor, a product made by a 0:06:52.600 --> 0:06:55.280 company called any Sphere that lets developers generate code using 0:06:55.320 --> 0:06:58.440 Anthropics models and other models, and we've had the confirmation 0:06:58.560 --> 0:07:00.600 that's the case thanks to the Information reporting the one 0:07:00.600 --> 0:07:03.280 point four billion dollars of the annualized revenue from Anthropic. 0:07:03.279 --> 0:07:05.720 It's from its two top customers, so around one hundred 0:07:05.720 --> 0:07:07.880 and sixteen million dollars in a month, and the biggest 0:07:07.880 --> 0:07:11.320 of which is Cursor. Confusingly, the Information also says that 0:07:11.320 --> 0:07:14.240 Anthropics Clad Code is generating nearly four hundred million dollars 0:07:14.240 --> 0:07:17.360 in annualized revenue, roughly doubling from just a few weeks ago. 0:07:17.440 --> 0:07:19.920 Per their report, it's around thirty three million dollars a 0:07:20.000 --> 0:07:24.200 monthly revenue. They really jacked that product up. I need 0:07:24.240 --> 0:07:27.080 to do a whole thing on glored code if I'm honest, 0:07:27.280 --> 0:07:30.040 because that whole thing has become a complete nightmare. They've 0:07:30.080 --> 0:07:33.520 had to completely the weekly rate limits are coming. God, 0:07:33.520 --> 0:07:35.360 what a fucking mess. Every time I think about this 0:07:35.400 --> 0:07:38.040 company get upset. In any case, I think Cursor is 0:07:38.040 --> 0:07:40.360 a huge indicator of the current fragility of the bubble 0:07:40.440 --> 0:07:42.760 and the fact that for most AI startups there's simply 0:07:42.760 --> 0:07:45.880 no way out because being acquired or going public does 0:07:45.920 --> 0:07:49.000 not appear to be a viable route. But before I explain, 0:07:49.080 --> 0:07:52.160 please listen to one of the many stable, normal advertisers 0:07:52.280 --> 0:07:54.880 that gives money to the show, not to me, though 0:07:55.120 --> 0:08:08.920 I don't get the money from the ads. And we're back. 0:08:09.600 --> 0:08:12.440 So how exactly is one AI powered coding environment such 0:08:12.440 --> 0:08:15.080 a load bearing part of the AI trade. We're going 0:08:15.120 --> 0:08:17.400 to be focusing now on both the fact that Cursor 0:08:17.480 --> 0:08:19.960 is a systemic risk to the AI industry, one where 0:08:20.040 --> 0:08:22.320 if it fails, could make every other player that bit 0:08:22.400 --> 0:08:24.920 less secure, and the fact that Cursor is emblematic of 0:08:24.920 --> 0:08:27.840 what makes general if AI companies so paradoxical, where you 0:08:27.840 --> 0:08:30.240 can have bonkers valuations on one hand, but also no 0:08:30.320 --> 0:08:32.920 way for these companies to go public or sell to 0:08:32.960 --> 0:08:37.120 someone else and justified the actual valuation itself. I know 0:08:37.200 --> 0:08:39.480 it sounds a little insane, but I believe that Cursor 0:08:39.559 --> 0:08:41.679 is the weak point of the entire bubble, and I'm 0:08:41.679 --> 0:08:43.959 going to explain why and how this could go. This 0:08:44.000 --> 0:08:46.560 is by no means inevitable, but I cannot work out 0:08:46.679 --> 0:08:50.640 what Cursor does other than this Cursor makes before at 0:08:50.720 --> 0:08:52.959 least the massive changes to its service. Though Tom the 0:08:53.040 --> 0:08:55.680 town over at Newcomer suggests that the revenue is still 0:08:55.720 --> 0:08:58.320 going up. For now, they make about five hundred million 0:08:58.360 --> 0:09:01.720 dollars in annualized revenue around forty two million dollars a month. 0:09:01.840 --> 0:09:04.439 This makes it the single highest earning general IFAI company 0:09:04.520 --> 0:09:07.959 that isn't called AI or Anthropic, and the highest learning 0:09:08.000 --> 0:09:12.080 company built on top of primarily Anthropics models. Its success 0:09:12.120 --> 0:09:14.800 is symbolic to the greater AI movement and just as 0:09:14.840 --> 0:09:17.240 it hit its peak, Anthropic and open Ai to a 0:09:17.280 --> 0:09:21.400 lesser extent, though they seem more interconnected. During GPT five's launch, 0:09:21.600 --> 0:09:25.440 decided to add priority processing in priority service tiers, demanding 0:09:25.440 --> 0:09:28.680 more money upfront and causing Cursor to have to massively 0:09:28.720 --> 0:09:31.680 degrade its service to keep up. I also have a 0:09:31.679 --> 0:09:33.760 premium piece that explains this from a few weeks go 0:09:33.880 --> 0:09:37.120 within a monologue two and it's over at wezered dot app. 0:09:37.160 --> 0:09:40.520 Please give me money to explain In short, curses AI 0:09:40.559 --> 0:09:43.280 powered coding Editor used to have fairly unrestrained access to 0:09:43.320 --> 0:09:46.839 the various models provided by companies like open Ai and Anthropic. 0:09:48.120 --> 0:09:51.000 In mid June, a few weeks after Anthropic introduced priority 0:09:51.000 --> 0:09:54.040 tiers that required enterprise companies to pay upfront and guarantee 0:09:54.040 --> 0:09:57.360 a certain throughput of tokens and increase costs on prompt cashing, 0:09:57.400 --> 0:10:00.000 which is a big part of AI coding. Curse mass 0:10:00.160 --> 0:10:02.400 the change the amount its users could use its product 0:10:02.520 --> 0:10:05.640 and introduce the two hundred dollars a month subscription. As 0:10:05.679 --> 0:10:08.680 an aside to this, Anthropic also competes with Curser's AI 0:10:08.760 --> 0:10:11.720 coding product with Clawed Code, its own service. I've talked 0:10:11.720 --> 0:10:13.120 about it in the past, and I'll talk about it 0:10:13.160 --> 0:10:16.960 some more in this series too. Cursor, as Anthropic's largest client, 0:10:17.040 --> 0:10:20.480 the second being GitHub Copilot, represents a material part of 0:10:20.520 --> 0:10:23.040 its revenue, and its surging popularity meant that it was 0:10:23.080 --> 0:10:26.480 sending more and more revenue to Anthropic. Anthropic use this 0:10:26.480 --> 0:10:29.760 opportunity to raise prices on accessing its models to continue 0:10:29.920 --> 0:10:33.040 providing service at an acceptable level to Cursor's customers by 0:10:33.080 --> 0:10:36.600 introducing priority toier access on May thirtieth, twenty twenty five. 0:10:37.280 --> 0:10:39.600 This has allowed Anthropic to duce its revenues, and due 0:10:39.640 --> 0:10:42.400 to the upfront nature of these contracts, Cursor is locked 0:10:42.400 --> 0:10:45.280 in regardless of how well it does. The net result 0:10:45.280 --> 0:10:47.720 of these cost increases means that Curser's product is less 0:10:47.720 --> 0:10:50.679 attractive to its customers and will thus eventually make it 0:10:50.800 --> 0:10:53.560 less money. At this point, one has to ask, how 0:10:53.600 --> 0:10:57.160 does Cursor survive. Its product isn't profitable and the means 0:10:57.160 --> 0:10:59.920 it used to make the company so successful have become untenable. 0:11:00.200 --> 0:11:02.640 It is guaranteed a certain throughput of tokens per second 0:11:02.640 --> 0:11:05.280 to the major model developers. Chief of the meanthropic. But 0:11:05.320 --> 0:11:08.720 Cursor itself said in the whole Cursor Ultra two hundred 0:11:08.760 --> 0:11:11.320 dollars a month thing that it signed multi year deals 0:11:11.360 --> 0:11:14.680 with multiple cloud providers like open Ai, Xai and Google. 0:11:15.760 --> 0:11:19.000 Cursor's product is now worse by the way. They also 0:11:19.040 --> 0:11:21.200 have made it so the automodel that used to be 0:11:21.240 --> 0:11:24.200 unlimited will now as of September fifteenth, charge a fee. 0:11:24.280 --> 0:11:26.600 It's so fucking funny. I love this shit. It's so 0:11:26.720 --> 0:11:29.640 funny watching this happen. But really, people are going to 0:11:29.640 --> 0:11:33.640 cancel their subscriptions, their annualized revenue will drop, and their 0:11:33.679 --> 0:11:36.160 ability to raise capital will suffer. As a direct result, 0:11:36.600 --> 0:11:39.000 Cursor will, regardless of this drop in revenue, have to 0:11:39.000 --> 0:11:40.840 pay the cloud companies what it owes them as if 0:11:40.880 --> 0:11:43.079 it had the business it used to because it was 0:11:43.120 --> 0:11:46.160 a guaranteed throughput. I've spoken to a few different people, 0:11:46.160 --> 0:11:48.840 including a company with an enterprise contract, either planning to 0:11:48.840 --> 0:11:50.640 cancel or trying to find a way out of their 0:11:50.679 --> 0:11:52.840 agreements with Cursor, and they were planning to do so 0:11:52.920 --> 0:11:56.600 as recently as April. I'm sorry recently as April. I 0:11:56.640 --> 0:11:58.520 mean they were planning to do back in April, before 0:11:58.559 --> 0:12:02.640 this shit happened. I'm not a here, you do understand anyway. 0:12:02.720 --> 0:12:04.760 If Cursor is allowed to die, it will be unable 0:12:04.800 --> 0:12:07.160 to pay a chunk of Anthropic and open AI's revenue, 0:12:07.160 --> 0:12:10.400 and yes, the revenue of people like Xai in Google 0:12:10.440 --> 0:12:13.520 as well. It also brings into question whether it's possible 0:12:13.559 --> 0:12:16.480 to build, putting aside any questions of profitability, a business 0:12:16.480 --> 0:12:18.760 of any kind offering services built on top of generative 0:12:18.800 --> 0:12:21.400 AI models, and in turn bring into doubt the veracity 0:12:21.440 --> 0:12:24.520 of investing in this sector writ large. It will also 0:12:24.559 --> 0:12:27.200 call into question whether any other generativai company is a 0:12:27.240 --> 0:12:29.960 real business. This will naturally lead to the question of 0:12:29.960 --> 0:12:33.679 why we're building all these goddamn data centers. Cursor at 0:12:33.679 --> 0:12:37.000 this point faces two options, die or get acquiet. This 0:12:37.040 --> 0:12:39.079 is not an attack on anyone who works at the company, 0:12:39.160 --> 0:12:42.080 nor anything personal. The unit economics of this business do 0:12:42.200 --> 0:12:44.840 not make sense, and yet on some level. Its existence 0:12:44.880 --> 0:12:48.000 is deeply important to the valley's future. And a large 0:12:48.040 --> 0:12:51.640 chunk of anthropics revenue I should add, and Tom totan 0:12:51.760 --> 0:12:54.280 over it, and Newcomer added something like ten percent of 0:12:54.320 --> 0:12:58.080 anthropics revenue comes from Cursor. This is so bad, but 0:12:58.160 --> 0:13:00.800 I'm gonna humor this. I'm gonna hue in this big time. 0:13:01.400 --> 0:13:06.200 Who would actually acquire Cursor? Maybe open Ai, but they 0:13:06.480 --> 0:13:10.040 couldn't acquire Windsurf for another AI company because they were 0:13:10.080 --> 0:13:12.680 too worried that Microsoft was going to get the somehow 0:13:12.840 --> 0:13:15.360 essential ip of what appears to be one of one 0:13:15.400 --> 0:13:19.200 hundred different AI powered coding environments. They also already trade 0:13:19.200 --> 0:13:22.040 and failed to buy Cursor. And if I'm honest, I 0:13:22.080 --> 0:13:23.880 would not be surprised if Curser would sell. 0:13:23.920 --> 0:13:24.079 Now. 0:13:24.160 --> 0:13:26.120 I don't know if open ai has the money, but 0:13:26.240 --> 0:13:30.760 you know they probably would sell. Honestly, Cursor fucked up 0:13:30.800 --> 0:13:33.400 bad not selling to open Ai. They could have got 0:13:33.400 --> 0:13:35.240 ten billion dollars in Sam Mortman would have had to 0:13:35.280 --> 0:13:38.840 accelerate the funding clause. It would have been so goddamn sick. 0:13:38.880 --> 0:13:41.760 But now the only sick here is Cursor's fragile, plagued 0:13:41.800 --> 0:13:45.480 business model. But how about Anthropic. I don't know about 0:13:45.480 --> 0:13:49.080 that one either. They've already got their own extremely expensive 0:13:49.080 --> 0:13:52.280 coding environment, which I estimated in my premium newsletter loses 0:13:52.320 --> 0:13:54.720 the company one hundred percent to ten thousand percent of 0:13:54.720 --> 0:13:59.000 a subscript subscription per customer. That's like a monthly subscription 0:13:59.679 --> 0:14:01.560 just from us. And they did that a few weeks ago. 0:14:01.640 --> 0:14:04.120 And now Anthropic, as of August twenty eighth, is adding 0:14:04.120 --> 0:14:07.080 weekly limits on accounts, which I believe creates some of 0:14:07.080 --> 0:14:09.880 the most gnarly churn in SaaS history. It's going to 0:14:09.920 --> 0:14:13.400 be quite nice. Also, does Anthropic really want to acquire 0:14:13.400 --> 0:14:18.880 its largest customer also with more money. They're not raising 0:14:18.960 --> 0:14:22.760 five billion dollars to bail out Cursor. Anthropic needs that 0:14:22.800 --> 0:14:25.960 to feed directly into Andy Jesse's ass pocket to keep 0:14:26.000 --> 0:14:29.000 offering increasingly more complex models that never quite seemed good 0:14:29.080 --> 0:14:32.360 enough to make any money. But how about Google? They 0:14:33.200 --> 0:14:36.920 know they just sort of bought Windsurf, but I'll get 0:14:37.000 --> 0:14:39.520 it's a complete mess, and they can't do that again. 0:14:39.920 --> 0:14:43.640 They've already given out the participation trophy multiple billions of 0:14:43.640 --> 0:14:46.000 dollars to investors and founders, so nobody has to get 0:14:46.040 --> 0:14:48.880 embarrassed about this. And then allowed Cognition to pick up 0:14:48.920 --> 0:14:50.880 the scraps of a business that made six point eighty 0:14:50.880 --> 0:14:53.440 three million dollars a month after burning one hundred and 0:14:53.480 --> 0:14:56.640 forty three million of investor capital, and tech runch reports 0:14:56.680 --> 0:14:58.840 that Windsurf was left for about one hundred million dollars 0:14:58.840 --> 0:15:02.400 in cash post acquisition. Tech Crunch also reports that Cognition 0:15:02.440 --> 0:15:04.480 paid about two hundred and fifty million dollars for what 0:15:04.560 --> 0:15:08.600 remained of the company. Google paid like two point something 0:15:08.680 --> 0:15:12.360 billion dollars to just take the founders, and Cognition picked 0:15:12.400 --> 0:15:14.600 up the rest. And this deal, by the way, didn't 0:15:14.600 --> 0:15:18.520 actually pay out the majority of Windsurf's employees. He I'll 0:15:18.520 --> 0:15:20.640 have the links to the links to all of this 0:15:20.720 --> 0:15:23.280 in the notes. This whole deal is real sickly though, 0:15:23.680 --> 0:15:26.440 that the investors got paid out, the founders got paid out, 0:15:26.520 --> 0:15:29.320 the employees got fugged. It really sucks. I'll get into 0:15:29.320 --> 0:15:31.880 it in a bit, but let's go with meta. Here's 0:15:31.920 --> 0:15:34.440 the thing. If I'm Michael Truell, the CEO of Cursor 0:15:34.480 --> 0:15:37.240 of any Sphere, even I am calling Mark Zuckerberg, and 0:15:37.240 --> 0:15:38.800 I'm pretending that I think the only person in the 0:15:38.840 --> 0:15:41.280 world who can usher in superintelligence is the guy who 0:15:41.320 --> 0:15:43.600 burned more than forty five billion dollars on the metaverse 0:15:43.720 --> 0:15:45.920 and believes that not wearing AI glasses in the future 0:15:45.920 --> 0:15:48.480 will be a disadvantage. I would say all manner of 0:15:48.480 --> 0:15:50.440 shit about the future, and that the only way to 0:15:50.480 --> 0:15:52.680 do this was to buy my AI powered coding startup 0:15:52.680 --> 0:15:57.080 that literally cannot afford to exist. I would tell Zuckerberg anything. 0:15:57.080 --> 0:15:59.560 I would be like, mate, I just watched the actual 0:15:59.600 --> 0:16:03.080 star I think you're the people from Stargate. He's never 0:16:03.120 --> 0:16:05.200 gonna watch Stargate. He's just gonna be like, whoa, yeah, 0:16:05.760 --> 0:16:09.400 I've heard that's a TV show. Wow, A great. And 0:16:09.440 --> 0:16:11.640 then you get however much money you need, just twenty 0:16:11.680 --> 0:16:14.040 billion dollars, He'll just pull that out of his pocket. 0:16:14.080 --> 0:16:16.520 He has it. He has twenty billion. He'll just handed 0:16:16.520 --> 0:16:18.600 to you. He's probably got it in is like in 0:16:18.640 --> 0:16:20.600 his safe. Just go and ask mine, how much money 0:16:20.640 --> 0:16:22.800 do you think Mark Zuckerberg's got his safe? Email me, 0:16:23.040 --> 0:16:27.120 email me with the guests. Anyway, This whole can't afford 0:16:27.120 --> 0:16:30.080 to exist thing that really is the problem. These companies 0:16:30.080 --> 0:16:32.200 are all going through the same motions that every company 0:16:32.240 --> 0:16:34.800 before them did. Raise as much money as possible, get 0:16:34.840 --> 0:16:37.000 as big as possible and eventually scale to the point 0:16:37.080 --> 0:16:40.600 of the e fat with enterprise cash. Except the real 0:16:40.640 --> 0:16:43.160 problem is that, just like the Big text New gluttony 0:16:43.200 --> 0:16:46.080 of physical real estate, it's taken on general ifai company 0:16:46.120 --> 0:16:48.200 is a burden with a constant aggressive form of cloud 0:16:48.280 --> 0:16:51.520 damn the endless punishment of the costs for accessing the 0:16:51.560 --> 0:16:54.360 API for general if ai models that always seem to 0:16:54.360 --> 0:16:56.240 get a little bit better, but never in such a 0:16:56.280 --> 0:16:59.360 way that anything really changes other than how much Anthropic 0:16:59.440 --> 0:17:01.280 and open Ai are going to need at the end 0:17:01.280 --> 0:17:03.800 of the month, or they break your startup's legs. I'm 0:17:03.800 --> 0:17:06.600 not even trying to be funny. Anthropic raised its prices 0:17:06.600 --> 0:17:09.800 on curses so severely it broke its already unprofitable business model. 0:17:10.040 --> 0:17:12.400 These products, while also for the most part not producing 0:17:12.400 --> 0:17:14.760 that much revenue, need to be sold with users being 0:17:14.800 --> 0:17:18.080 aware of and or sensitive to the costs of providing them. 0:17:18.400 --> 0:17:20.960 The Cursor's original product was a twenty dollars a month 0:17:21.240 --> 0:17:24.040 service for five hundred fast requests of different models, in 0:17:24.040 --> 0:17:26.480 the same way that accessing clawed code on any subscription 0:17:26.560 --> 0:17:28.840 is either twenty one hundred or two hundred dollars a month. 0:17:29.119 --> 0:17:32.000 Rather than paying per API cores, because these companies also 0:17:32.160 --> 0:17:34.840 products that shield the customer from the actual costs of 0:17:34.880 --> 0:17:38.879 running the services, and those costs could be vast. One article, 0:17:38.880 --> 0:17:40.920 which I'll link to in the spreadsheet, argues that given 0:17:40.960 --> 0:17:43.320 the trajectory of the development of AI models, it's not 0:17:43.400 --> 0:17:46.320 unreasonable to imagine a full time engineer spending one hundred 0:17:46.400 --> 0:17:52.040 thousand dollars in tokens each year per engineer. Just really 0:17:52.080 --> 0:17:54.480 think about that for a second. I'll have the link 0:17:54.560 --> 0:17:57.919 in there. It's fucking stupid. It's so fucking stupid, I 0:17:58.000 --> 0:18:01.200 want to scream. And the irony is that, despite being 0:18:01.200 --> 0:18:03.920 willing to kill these companies by fundamentally changing the terms 0:18:03.960 --> 0:18:06.520 and put in which they access their models, Anthropic is 0:18:06.560 --> 0:18:10.040 also in some way dependent on Cursor, Repler and other 0:18:10.119 --> 0:18:12.800 similar firms continuing to buy tokens at the same rate 0:18:12.840 --> 0:18:15.720 as before, as that consumption is baked into their own 0:18:16.119 --> 0:18:18.880 arr figures, as well as the forward looking revenue projections 0:18:18.920 --> 0:18:22.120 they've given to investors to convince them to give them 0:18:22.160 --> 0:18:25.800 five billion dollars. It is, in some sense of Kobyashi 0:18:25.920 --> 0:18:29.080 maru Anthropic is an existential need to screw over its 0:18:29.080 --> 0:18:32.199 customers by hiking rates and imposing long term commitments, but 0:18:32.240 --> 0:18:35.320 its existence is also in some way predicated on these 0:18:35.320 --> 0:18:38.919 companies continuing to exist. If Cursor and Replit both die, 0:18:39.119 --> 0:18:42.760 that's a significant chunk of anthropics API business gone, and 0:18:42.800 --> 0:18:46.240 may I remind you that significantly overshadows its subscription business, 0:18:46.240 --> 0:18:48.800 making it almost like an inverse of open AI, where 0:18:48.840 --> 0:18:52.240 subscriptions drive the bulk of revenue, also a shitty business model. 0:18:52.800 --> 0:18:56.760 Anthropic's future is wedded to Cursor, and I just don't 0:18:56.800 --> 0:18:59.560 see how Cursor survives, let alone exists or gets subsumed 0:18:59.560 --> 0:19:01.840 by another company in a way that mirrors our acquisitions 0:19:01.840 --> 0:19:05.439 have worked since Ever, if Curser does not sell for 0:19:05.440 --> 0:19:08.359 a healthy amount, I'm talking ten to twenty billion dollars, 0:19:08.400 --> 0:19:10.760 and I mean actually sell, not the founders are hired 0:19:10.760 --> 0:19:13.119 and a strange contractual agreement that pays out investors and 0:19:13.160 --> 0:19:16.320 its assets are sold to Rick from porn Stars. It's pawn. 0:19:16.359 --> 0:19:18.960 By the way, it will prove that no generative AI 0:19:19.040 --> 0:19:23.400 company to this date has actually been successful in reality. 0:19:23.440 --> 0:19:26.119 I expect a Chumley esque deal that helps CEO Michael 0:19:26.160 --> 0:19:29.280 Truell by a Porsche while his staff makes nothing. It's 0:19:29.280 --> 0:19:31.960 Cursor worth ten billion dollars or more. No, no matter 0:19:31.960 --> 0:19:33.840 how good its product may or may not be, it 0:19:33.880 --> 0:19:35.320 is not good enough to be sold at a price 0:19:35.359 --> 0:19:38.119 that doesn't require Cursor to incinerate hundreds of millions of 0:19:38.160 --> 0:19:42.000 dollars with no end in sight. And this ultimately gives 0:19:42.080 --> 0:19:46.480 us a real conundrum. Why aren't generative AI startups selling 0:19:47.280 --> 0:19:49.600 I'll tell you what, though, other people are selling stuff 0:19:49.600 --> 0:19:51.159 and it's time for you to buy it. Listen to 0:19:51.240 --> 0:19:52.719 these ads and let me know what you think at 0:19:52.720 --> 0:20:02.840 my email address, which is Robert Evans app and we're back. So, really, though, 0:20:02.960 --> 0:20:06.399 why are there so few generative AI acquisitions? I know 0:20:06.440 --> 0:20:07.960 you're probably going to assume that there have been a 0:20:07.960 --> 0:20:11.399 ton of them, right, if you've not been obsessively reading 0:20:11.480 --> 0:20:14.080