WEBVTT - The Hater's Guide To The AI Bubble, Pt, 2 0:00:02.800 --> 0:00:06.880 Zone Media Hell, and welcome to Better Offline. I'm your 0:00:06.880 --> 0:00:21.799 host ed Zeitron. Subscribe to the newsletter by the merchandise. 0:00:21.840 --> 0:00:23.560 It's all in the notes. And we're on the second 0:00:23.640 --> 0:00:26.680 installment of our three part Hater's Guide to the AI 0:00:26.800 --> 0:00:30.040 bubble and the cracks within the generative AI industry and 0:00:30.040 --> 0:00:32.600 how they're becoming bigger and scarier and the potential economic 0:00:32.680 --> 0:00:35.760 meltdown course by a collapse in generative AI spending. Well, 0:00:35.960 --> 0:00:38.440 it's not really general if AI spending. It's literally just 0:00:38.479 --> 0:00:41.159 fucking GPUs, and I think it might be sooner and 0:00:41.320 --> 0:00:44.360 likelier than many think. Toward the end of the last episode, 0:00:44.360 --> 0:00:46.360 we talked about one of the inane comparisons we hear 0:00:46.400 --> 0:00:49.760 between today's nation state size spending on jen ai capital 0:00:49.760 --> 0:00:52.840 expenditures and the investments that Amazon made when scaling Amazon 0:00:52.880 --> 0:00:56.440 Web Services, which was literally the foundation of Dow Computing 0:00:56.480 --> 0:00:58.920 at scale. I would say someone's going to email and 0:00:58.920 --> 0:01:01.280 say I'm wrong, not going to read it, and I 0:01:01.320 --> 0:01:03.640 had to cut things short because we ran out of time. 0:01:03.760 --> 0:01:04.200 But I want to. 0:01:04.160 --> 0:01:06.600 Continue the conversation because I think it's important to examine 0:01:06.600 --> 0:01:10.040 this comparison thoroughly, if not just to explain why it 0:01:10.080 --> 0:01:12.080 doesn't work. It's also I want to stop. I want 0:01:12.080 --> 0:01:13.880 to stop hearing it. I want when people say it 0:01:13.880 --> 0:01:15.280 to me, I just want to send them this fucking 0:01:15.319 --> 0:01:17.520 episode and say leave me alone, buddy boy. 0:01:18.720 --> 0:01:20.360 But but bye. 0:01:20.400 --> 0:01:22.160 But the first point I want to make in this 0:01:22.200 --> 0:01:24.800 episode is that generative AI and large language models do 0:01:24.880 --> 0:01:28.840 not resemble Amazon Web Services or the greater cloud compute boom, 0:01:28.920 --> 0:01:32.520 and generative AI is not infrastructure. Now, some people compare 0:01:32.640 --> 0:01:35.720 llms and their associated services to Amazon Web Services or 0:01:35.760 --> 0:01:39.279 services like Microsoft Zero or Google Cloud, their giant, multi 0:01:39.280 --> 0:01:43.280 billion dollar operations that basically share their server capacity with 0:01:43.319 --> 0:01:46.120 companies wanting to run stuff on the Internet or within 0:01:46.160 --> 0:01:49.280 their own within their own systems. A very fudgy way 0:01:49.280 --> 0:01:52.440 of putting. They help make sure that applications work online. 0:01:52.720 --> 0:01:56.600 These are very very useful services, and by the way, 0:01:56.600 --> 0:01:58.280 people are wrong to make the comparison between them and 0:01:58.280 --> 0:02:01.880 the l lms. As I'll get into now. Amazon Web 0:02:01.920 --> 0:02:04.440 Services when it launched, comprised of things like and forgive 0:02:04.600 --> 0:02:07.400 me how much I'm going to dilute this, Amazon's Elastic 0:02:07.440 --> 0:02:10.000 Compute Cloud EC two, where you rent space in Amazon 0:02:10.080 --> 0:02:12.480 service to run applications in the cloud, or Amazon Simple 0:02:12.520 --> 0:02:15.680 Storage S three, which is enterprise level storage for applications 0:02:15.680 --> 0:02:18.679 and storing things, is not just like a simple hard drive. 0:02:18.720 --> 0:02:21.560 It's redundancy, it's making sure it's copied in places, so 0:02:21.639 --> 0:02:25.200 latency comes down tons of other things. But in simpler terms, 0:02:25.800 --> 0:02:27.840 if you were providing a cloud based service, you used 0:02:27.880 --> 0:02:30.280 Amazon to both stored a stuff that the service needed 0:02:30.360 --> 0:02:33.320 and the cloud actual cloud based processing. So of compute, 0:02:33.320 --> 0:02:35.680 so like your compute loads and runs applications, but delivered 0:02:35.680 --> 0:02:37.760 the thousands of millions of people online. 0:02:38.200 --> 0:02:39.600 And this is a huge industry. 0:02:40.280 --> 0:02:42.680 Amazon Web services are alone brought in WHEB revenues are 0:02:42.680 --> 0:02:44.840 over one hundred billion dollars in twenty twenty four. And 0:02:44.840 --> 0:02:47.280 while Microsoft and Google don't break out their cloud revenues, 0:02:47.280 --> 0:02:50.200 they're similarly large parts of their companies, and Microsoft is 0:02:50.240 --> 0:02:52.440 used as zero and the past to patch over shoddy growth. 0:02:53.000 --> 0:02:56.440 These services are also selling infrastructure. You aren't just paying 0:02:56.480 --> 0:02:58.799 for compute, but the ability to access storage and deliver 0:02:58.880 --> 0:03:02.080 services with low lateent so users have a snappy experiences 0:03:02.680 --> 0:03:04.360 wherever they are in the world. And I know I 0:03:04.440 --> 0:03:07.680 just said a snappy experiences. I'm not editing it. The 0:03:07.720 --> 0:03:10.440 subtle magic of the Internet is that it works at all, 0:03:10.480 --> 0:03:12.359 and a large part of that is the cloud compute 0:03:12.360 --> 0:03:15.040 infrastructure and oligopoly of the main cloud providers. Having such 0:03:15.040 --> 0:03:17.919 a vast data centers, this is much cheaper than doing 0:03:17.919 --> 0:03:20.520 it yourself, and to a certain point, Jobbox moved away 0:03:20.520 --> 0:03:22.880 from Amazon Web Services is it at scale, for example, 0:03:23.440 --> 0:03:25.200 but this also allows someone to take care of the 0:03:25.240 --> 0:03:27.960 maintenance of the hardware and make sure it actually gets 0:03:28.040 --> 0:03:31.160 your stuff to your customers. You also don't have to 0:03:31.160 --> 0:03:34.000 worry about spikes and usage because these things are usage based, 0:03:34.000 --> 0:03:36.440 hence the elastic and you can always add more compute 0:03:36.440 --> 0:03:38.560 to meet demand or just have it in a particular time. 0:03:38.960 --> 0:03:42.560 There is, of course nuance, security specific features, content specific 0:03:42.600 --> 0:03:46.680 delivery services, data based services. There's nuance behind these clouds. 0:03:47.080 --> 0:03:49.640 You're buying into the infrastructure of the infrastructure provider, and 0:03:49.680 --> 0:03:52.120 the reason these products are so profitable is that in 0:03:52.200 --> 0:03:54.480 part you are handing off the problems and responsibility to 0:03:54.480 --> 0:03:57.680 somebody else. And also, most web applications are not that 0:03:57.800 --> 0:04:01.520 demanding of cloud compute. They might be at scale expensive 0:04:01.560 --> 0:04:04.680 to provide to millions of people, but Facebook was not 0:04:04.840 --> 0:04:09.200 a super complex, I don't know website depending on thousands 0:04:09.280 --> 0:04:12.400 or millions of GPUs, and based on the idea, there 0:04:12.400 --> 0:04:14.560 are multiple product categories you can build on top of 0:04:14.560 --> 0:04:17.920 something like edblys Because ultimately cloud services are about Amazon, 0:04:18.000 --> 0:04:22.000 Microsoft and Google running your infrastructure for you. Large language 0:04:22.000 --> 0:04:25.160 models and their associated services are completely different, despite these 0:04:25.160 --> 0:04:27.280 companies attempting to prove otherwise. And it starts with a 0:04:27.360 --> 0:04:31.279 very very simple problem. Why did any of these companies 0:04:31.279 --> 0:04:33.400 build these giant data centers and why did they fill 0:04:33.400 --> 0:04:36.840 them full of GPUs? Amazon Web Services was created out 0:04:36.839 --> 0:04:40.320 of necessity. Amazon's infrastructure needs were so great that it 0:04:40.360 --> 0:04:42.799 effectively had to build out the software and hardware necessary 0:04:42.839 --> 0:04:46.240 to deliver a store that sold theoretically everything, the theoretically anywhere, 0:04:46.520 --> 0:04:49.320 handling both the traffic and customers, delivering the software that 0:04:49.400 --> 0:04:52.920 runs Amazon dot Com quickly and reliably and well, making 0:04:52.960 --> 0:04:56.039 sure things kept working, making sure they were stable. And 0:04:56.080 --> 0:04:57.920 it didn't need to come up with a reason for 0:04:58.000 --> 0:05:01.159 people to run web applications. They were already running applications 0:05:01.160 --> 0:05:04.680 client side on their computers. They realized that doing so 0:05:04.760 --> 0:05:07.159 at scale would be cool, or they were already doing 0:05:07.200 --> 0:05:10.479 so in a way that was likely not particularly cost effective. 0:05:11.240 --> 0:05:13.039 And the ways that we're doing so, they were inflexible, 0:05:13.080 --> 0:05:16.720 and they required specially skills and indeed physical infrastructure personnel. 0:05:16.720 --> 0:05:20.120 They were quite expensive. So Amazon Web Services took something 0:05:20.120 --> 0:05:22.400 that people already did and what there was actually a 0:05:22.400 --> 0:05:25.360 proven demand for, and made it better and scaled it. Eventually, 0:05:25.440 --> 0:05:28.160 Google and Microsoft copied done because that's all they can do. 0:05:28.880 --> 0:05:31.520 And that appears to be the only similarity with generative AI. 0:05:31.640 --> 0:05:33.640 That due to the ridiculous costs of both data centers 0:05:33.640 --> 0:05:36.520 and GPUs necessary to provide these services, it's largely impossible 0:05:36.560 --> 0:05:37.760 for others to enter the market. 0:05:38.520 --> 0:05:38.719 You know. 0:05:38.760 --> 0:05:41.120 After that, generative AI feels more like a feature of 0:05:41.160 --> 0:05:45.120 cloud infrastructure rather than the infrastructure itself. AWS and similar 0:05:45.120 --> 0:05:48.279 medic clouds are versatile, flexible, and multi faceted. Generative AI 0:05:48.440 --> 0:05:52.880 does what generative AI does well, that's about it. You 0:05:52.920 --> 0:05:55.480 can run lots of different things in AWS. What are 0:05:55.480 --> 0:05:58.480 the different things you can run using large language models? 0:05:58.480 --> 0:06:01.599 What are the different use cases and indeed user requirements 0:06:01.680 --> 0:06:04.880 that make this the supposed next big thing. Perhaps the 0:06:05.000 --> 0:06:07.520 argument is that generator of AI is the next AWS 0:06:07.600 --> 0:06:09.440 or similar cloud service because you can build the next 0:06:09.440 --> 0:06:11.880 great companies on the infrastructure of others. The models of 0:06:11.960 --> 0:06:16.920 say open AI and anthropic and the service of Microsoft. Okay, okay, 0:06:17.839 --> 0:06:20.200 let's humor this point too. You can build the next 0:06:20.200 --> 0:06:22.560 great AI startup, and you have to build it on 0:06:22.560 --> 0:06:24.960 one of the megaclouds because they're the only ones that 0:06:25.000 --> 0:06:29.839 can afford to build the infrastructure. One inc wincteny small problem. 0:06:30.400 --> 0:06:32.600 Companies built on top of large language models don't make 0:06:32.720 --> 0:06:35.200 much money, and in fact they're almost all deeply unprofitable. 0:06:35.640 --> 0:06:40.320 But let's establish a few flats to get going. I said, flats, flats, 0:06:40.520 --> 0:06:41.240 Jesus Christ. 0:06:41.480 --> 0:06:41.880 Facts. 0:06:42.520 --> 0:06:45.880 Here are the flats I'm establishing. Outside of one exception, 0:06:46.080 --> 0:06:48.600 mid Journey, which claimed it was profitable in twenty twenty two, 0:06:48.640 --> 0:06:50.839 which may not still be the case. I've actually reached 0:06:50.839 --> 0:06:52.159 out to ask them and they didn't get. 0:06:52.000 --> 0:06:52.359 Back to me. 0:06:52.680 --> 0:06:58.560 Every single LLM model is company is unprofitable, often wildly so. 0:06:58.600 --> 0:07:00.960 Outside of open ai and oropic, in any sphere which 0:07:00.960 --> 0:07:03.520 makes the AI coding app cursor, there are no large 0:07:03.560 --> 0:07:06.680 language model companies either building models or services on top 0:07:06.720 --> 0:07:08.599 of others models that make more than five hundred million 0:07:08.640 --> 0:07:12.360 dollars in annualized revenue meaning month times twelve. Outside mid 0:07:12.400 --> 0:07:14.960 Journeys two hundred million arr and Iron clouds one hundred 0:07:14.960 --> 0:07:21.120 and fifty million arr Also fucking perplexity, there are only 0:07:21.200 --> 0:07:24.160 twelve generative AI powered companies making one hundred million dollars 0:07:24.200 --> 0:07:27.320 annualized or eighteen point three million dollars a month in revenue. 0:07:27.560 --> 0:07:31.200 The database then, this is the Information's AI. Generative AI 0:07:31.280 --> 0:07:34.480 database doesn't have replt, which also announced it hit one 0:07:34.520 --> 0:07:36.960 hundred million in analyzed revenue. I've included it in my 0:07:37.040 --> 0:07:41.000 statement of facts. Of these companies, two of them have 0:07:41.040 --> 0:07:43.360 been acquired, move Works acquired by service Now in March 0:07:43.360 --> 0:07:45.800 twenty twenty five after the company shit the Big Big Time, 0:07:46.040 --> 0:07:48.640 and Windsurf, which was acquired by Google and Cognition in 0:07:48.720 --> 0:07:50.600 July twenty twenty five and one of the most annoying 0:07:50.640 --> 0:07:53.360 deals of all time. But for the sake of simplicity, 0:07:53.400 --> 0:07:55.920 I've left out companies like Surge, Scale, Cheering, and Together, 0:07:56.000 --> 0:07:59.120 all of whom run consultancies selling services and training stuff 0:07:59.120 --> 0:08:02.600 for training models. Otherwise, there are seven companies total that 0:08:02.640 --> 0:08:05.200 make fifty million dollars or more annual recurring revenue, which 0:08:05.240 --> 0:08:07.320 is four point one six million dollars a month. 0:08:08.440 --> 0:08:09.680 Now, none of this is to say. 0:08:09.520 --> 0:08:11.440 That one hundred million dollars isn't a lot of money 0:08:11.480 --> 0:08:12.080 to you and me. 0:08:12.520 --> 0:08:13.320 I just want to be clear. 0:08:13.320 --> 0:08:15.000 If you want to give me one hundred million dollars, 0:08:15.080 --> 0:08:18.280 I'll do anything. I'll wink like a pig for you anyway. 0:08:18.360 --> 0:08:20.040 But in the world of software as a service or 0:08:20.120 --> 0:08:23.640 enterprise software, this is jump change HubSpot At revenues are 0:08:23.640 --> 0:08:26.200 two point six three billion dollars in its twenty twenty 0:08:26.240 --> 0:08:29.520 four financial year. Three years into this crap, and Generative 0:08:29.520 --> 0:08:32.800 AI's highest grossing companies outside of open Ai ten billion 0:08:33.040 --> 0:08:36.400 annualized as of June and Anthropic four billion annualized as July. 0:08:36.559 --> 0:08:39.000 Don't like saying that word. Both of them loose billions 0:08:39.000 --> 0:08:42.640 a year after revenue. There are really three problems here. 0:08:42.960 --> 0:08:45.920 Businesses powered by Generative AI do not seem to be popular, 0:08:46.280 --> 0:08:49.680 Those businesses that are remotely popular are deeply unprofitable, and 0:08:49.720 --> 0:08:52.920 even the less popular generative AI powered businesses are also 0:08:53.000 --> 0:08:56.280 deeply unprofitable. But I want to start somewhere because I 0:08:56.440 --> 0:09:01.120 keep hearing about fucking Cursor. Fucking's start with any sphere 0:09:01.120 --> 0:09:05.520 and Cursor and their app Cursor. It's an AI powered 0:09:05.559 --> 0:09:08.040 coding app and they have five hundred million dollars of 0:09:08.120 --> 0:09:09.080 annualized revenue. 0:09:09.200 --> 0:09:10.720 Pretty great, right, ha. 0:09:11.240 --> 0:09:13.640 It hit two hundred million dollars in annualized revenue in 0:09:13.679 --> 0:09:16.560 March and then hit five hundred million in June after 0:09:16.640 --> 0:09:20.800 raising nine hundred million dollars. That's amazing, ed, ed it's 0:09:20.880 --> 0:09:24.000 time walk to the garage. ED, it's over for you. Wrong, 0:09:24.240 --> 0:09:26.760 it's a mirage. Cursor's growth was the result of an 0:09:26.800 --> 0:09:29.000 unsustainable business model that it's now had to replace with 0:09:29.040 --> 0:09:32.600 opaque terms of service, dramatically restricting access to models, and 0:09:32.679 --> 0:09:35.040 rate limits that effectively stop its users using the product 0:09:35.080 --> 0:09:37.040 at the price point they were used to go to 0:09:37.160 --> 0:09:40.640 Arsnash Cursor on red app Take a look. Take a 0:09:40.640 --> 0:09:42.120 look at how happy everyone is. 0:09:42.440 --> 0:09:43.240 I want to know one. 0:09:43.240 --> 0:09:44.960 My peers in the media don't seem to have the 0:09:44.960 --> 0:09:48.200 ability to talk to actual fucking customers. It's ridiculous. This 0:09:48.240 --> 0:09:51.120 company is circling the drain, and nobody seems to want 0:09:51.160 --> 0:09:53.880 to talk about it, despite how big a deal that is. 0:09:54.240 --> 0:09:54.440 Oh. 0:09:54.480 --> 0:09:57.160 Also, Curse is horribly unprofitable, and I believe there are 0:09:57.160 --> 0:09:59.520 a sign of things to come in generative AI. A 0:09:59.520 --> 0:10:01.319 couple of weeks weeks ago, I wrote up the dramatic 0:10:01.400 --> 0:10:03.600 changes that Cursor made to its service in the middle 0:10:03.600 --> 0:10:06.199 of June or my premium newsletter and discovered that they 0:10:06.480 --> 0:10:09.520 timed these changes precisely with Anthropic and open Ai to 0:10:09.559 --> 0:10:13.240 a lesser extent, adding service tiers and priority processing, which 0:10:13.280 --> 0:10:15.360 is tech language for pay us extra if you have 0:10:15.400 --> 0:10:17.920 a lot of customers or face rate limits or service delays. 0:10:17.960 --> 0:10:18.480 Asshole. 0:10:19.160 --> 0:10:21.480 These price ships have also led to companies like replt 0:10:21.600 --> 0:10:24.000 having to make significant changes to their pricing model that 0:10:24.040 --> 0:10:29.360 disfavors users. People are finding in really simple terms that 0:10:29.520 --> 0:10:32.079 what they used to get for twenty bucks is much much, much, 0:10:32.200 --> 0:10:35.640 much much smaller curse the users hit rate limits. Replit 0:10:35.720 --> 0:10:38.199 users are hitting rate limits, and even then when they 0:10:38.200 --> 0:10:39.840 try and do the same things, they're spending way more 0:10:39.880 --> 0:10:41.560 money if they go pay as you go. It's a 0:10:41.559 --> 0:10:44.400 complete fast But I'm going to repeat some of this 0:10:44.440 --> 0:10:46.520 stuff from the premium newsletter because there is a time 0:10:46.520 --> 0:10:49.000 of events that I believe are going to be in 0:10:49.040 --> 0:10:52.000 the big short to AI Boogloo all right. In or 0:10:52.080 --> 0:10:54.440 around May fifth, twenty twenty five, Cursor closed the nine 0:10:54.480 --> 0:10:57.320 hundred million dollar funding round in a Around May twenty second, 0:10:57.360 --> 0:11:00.640 twenty twenty five, Anthropic launched Clawed four Opus and new 0:11:00.679 --> 0:11:03.480 models with Sona and Opus, both of them kind of 0:11:03.480 --> 0:11:05.800 well known for coding, and on May thirtieth, twenty twenty five, 0:11:05.840 --> 0:11:09.440 they added service tiers, including priority pricing specifically focused on 0:11:09.520 --> 0:11:12.160 cash heavy products like Cursor and the cash is when 0:11:12.160 --> 0:11:14.240 you put stuff that you're going to be looking at regularly, 0:11:14.440 --> 0:11:15.880 take a look at it, and you can use it 0:11:15.880 --> 0:11:19.200 more readily. Cash is the CAC eight G, by the way, 0:11:19.840 --> 0:11:21.800 is generally something that's for efficiency. 0:11:22.360 --> 0:11:23.160 The idea that you. 0:11:23.160 --> 0:11:26.920 Would add a toll onto the cash is fucking disgusting 0:11:26.920 --> 0:11:30.560 and only targeted coding startups. But on May thirtieth, twenty 0:11:30.600 --> 0:11:33.960 twenty five, Reutter's reported the Anthropics annualized revenue hit three 0:11:34.000 --> 0:11:37.760 billion dollars, with a key driver being code generation. This 0:11:37.800 --> 0:11:40.200 translates to around two hundred and fifty million dollars in 0:11:40.280 --> 0:11:43.880 monthly revenue. June ninth, twenty twenty five, CNBC reported open 0:11:43.880 --> 0:11:47.680 Ai'd hit ten billion dollars in annualized revenue. And yeah, 0:11:47.720 --> 0:11:50.360 when they said ann your recurring revenue, they meant annualized. 0:11:50.559 --> 0:11:52.319 But the very same day they cut the price of 0:11:52.360 --> 0:11:54.960 their three model by eighty percent, which competes directly with 0:11:55.000 --> 0:11:56.760 Clawed four Opus by the way, and This was a 0:11:56.800 --> 0:11:59.760 direct and aggressive attempt to force Anthropic to kind of 0:11:59.800 --> 0:12:03.120 like make too ether lower prices or compete. It's just 0:12:03.400 --> 0:12:07.160 shtheads fuckinging around with assholes. But on or around June sixteen, 0:12:07.200 --> 0:12:09.600 twenty twenty five, Cursor changed its pricing, added a new 0:12:09.640 --> 0:12:11.880 two hundred dollar a month Ultra tier that, in their 0:12:11.920 --> 0:12:14.800 own words, was made possible by multi year partnerships with 0:12:14.880 --> 0:12:18.680 open Ai, Anthropic, Google, an Xai, which translates to multi 0:12:18.720 --> 0:12:21.960 year commitments to spend which can be amortized as monthly amounts. 0:12:22.400 --> 0:12:25.440 A day later, on June seventeenth, Cursor dramatically changed its 0:12:25.440 --> 0:12:28.679 offering to it for its twenty dollars a month subscriptions 0:12:28.720 --> 0:12:31.520 to usage base, where one got at least the value 0:12:31.520 --> 0:12:33.400 of their subscription, so a twenty bus a month person 0:12:33.440 --> 0:12:37.520 would get more than twenty dollars of API course in compute, 0:12:37.720 --> 0:12:41.319 along with arbitrary rate limits and unlimited access to Cursor's 0:12:41.320 --> 0:12:44.280 own slow model that its users hey. Then on June eighteenth, 0:12:44.360 --> 0:12:46.679 Repler and other vibe coding company that I had previously 0:12:46.720 --> 0:12:50.079 mentioned announced their effort based pricing increases that were massive. 0:12:50.559 --> 0:12:53.280 July first, the Information reported. 0:12:52.920 --> 0:12:56.320 The Anthropic hit four billion dollars of annualized revenue making 0:12:56.360 --> 0:12:59.080 three hundred and thirty million dollars a month, an increase 0:12:59.120 --> 0:13:01.520 of eighty three million dollars a month. We'll just under 0:13:01.520 --> 0:13:03.240 twenty five percent in the space of a month. 0:13:04.920 --> 0:13:05.160 Hmm. 0:13:05.960 --> 0:13:18.800 Where could that money have come from? In simpler terms, 0:13:18.840 --> 0:13:21.080 Cursor raised nine hundred million dollars and very likely had 0:13:21.120 --> 0:13:23.000 to hand large amounts of that money over to Open 0:13:23.040 --> 0:13:25.480 Air and Anthropic to keep doing business with them, then 0:13:25.480 --> 0:13:28.000 immediately change the terms of service to make them worse 0:13:28.000 --> 0:13:30.320 for their customers. And as I said at the time, 0:13:30.360 --> 0:13:32.320 and this is a direct quote from my news there, 0:13:33.440 --> 0:13:35.640 while some met, no, I can't do the Kevin Ruth's 0:13:35.679 --> 0:13:38.679 voice and doing my own stuff, pardon me. While some 0:13:38.760 --> 0:13:41.520 may believe that Open AI and Anthropic hitting annualized revenue 0:13:41.559 --> 0:13:43.920 milestones is good news, you have to consider how these 0:13:43.960 --> 0:13:46.559 milestones were hit. Based on my reporting, I believe that 0:13:46.600 --> 0:13:50.200 both companies are effectively doing steroids, forcing massive infrastructural costs 0:13:50.200 --> 0:13:52.600 onto big customers as a means of covering the increasing 0:13:52.640 --> 0:13:55.160 costs of their own models. There is simply no other 0:13:55.200 --> 0:13:58.160 aid to read this situation. By making these changes, Anthropic 0:13:58.200 --> 0:14:00.800 is intentionally making it harder for its larger costs largest 0:14:00.800 --> 0:14:03.320 customer to do business. By the way, Cursor is their 0:14:03.400 --> 0:14:06.920 largest customer, creating the extra revenue by making Cursors product 0:14:07.000 --> 0:14:10.240 worse by proxy. What's sickening about this particular situation. It 0:14:10.240 --> 0:14:12.720 doesn't really matter if Curs's customers are happy or sad. 0:14:13.080 --> 0:14:17.920 They like open AI's Enterprise Priority Access API Anthropic in 0:14:17.920 --> 0:14:20.280 this case, require a long term commitment which involves a 0:14:20.280 --> 0:14:22.680 minimum through put of tokens per second as part of 0:14:22.680 --> 0:14:26.000 their tiered access program. If Curs's customers drop off, both 0:14:26.040 --> 0:14:28.360 Anthropic and open Ai still get their cut, and if 0:14:28.400 --> 0:14:31.120 curses customers somehow out spend those commitments, they'll either still 0:14:31.120 --> 0:14:34.280 get rate limited or any sphere willkin cur more costs. 0:14:35.480 --> 0:14:36.600 Why do you care about this? 0:14:37.040 --> 0:14:39.600 Well, Cursor is the largest and most successful genetive AI 0:14:39.760 --> 0:14:42.080 company by far. In these aggressive and desperate changes to 0:14:42.120 --> 0:14:45.080 its products suggest that a that its products are deeply unprofitable, 0:14:45.080 --> 0:14:46.880 and b that its current growth was the result of 0:14:46.880 --> 0:14:48.560 offering a product that it was not the one it 0:14:48.560 --> 0:14:51.920 would sell in the long term. Cursor misled its customers, 0:14:52.120 --> 0:14:55.920 and its current revenue is as a result, highly unlikely 0:14:55.960 --> 0:14:59.240 to stay at this level. Worse still, two anthropic engineers 0:14:59.360 --> 0:15:01.560 left from the the Clawed Code team to go and 0:15:01.600 --> 0:15:04.160 work at Cursor two weeks ago, and they have already 0:15:04.200 --> 0:15:06.600 come back. This heavily suggests that whatever they saw over 0:15:06.600 --> 0:15:09.240 there wasn't compelling enough to make them stay. As I 0:15:09.280 --> 0:15:12.480 also said, while Cursor may have raised nine hundred million dollars, 0:15:12.640 --> 0:15:15.760 it was really open Aianthropic XAI and Google that got 0:15:15.800 --> 0:15:18.640 that money. At this point, there are no profitable price 0:15:18.680 --> 0:15:21.040 AI startups, and it's highly unlikely that the new pricing 0:15:21.120 --> 0:15:23.120 models by both Cursor and Replet are going to help. 0:15:23.720 --> 0:15:24.600 These are now the. 0:15:24.600 --> 0:15:27.160 New terms of doing business with the big model companies, 0:15:27.320 --> 0:15:29.840 a shakedown where you pay for priority access or tears, 0:15:29.920 --> 0:15:33.640 or face indeterminate delays or rate limits. Any start up 0:15:33.680 --> 0:15:36.640 scaling into an enterprise integration of General AVII, which means 0:15:36.680 --> 0:15:39.280 in this case anything that requires a level of service 0:15:39.360 --> 0:15:41.880 uptime has to commit to both a minimum amount of 0:15:41.920 --> 0:15:43.720 months and the throughput of tokens, which means that the 0:15:43.720 --> 0:15:46.280 price of starting an AI company that gets any kind 0:15:46.320 --> 0:15:50.120 of real market traction just dramatically increased. Well, one could say, oh, 0:15:50.160 --> 0:15:52.280 perhaps you don't need priority access. The need here is 0:15:52.320 --> 0:15:54.920 something that can be entirely judged by anthropic and open 0:15:54.960 --> 0:15:58.000 ai in a totally opaque manner. They can and they 0:15:58.040 --> 0:16:00.520 will throttle companies that are two demanding on their systems. 0:16:00.600 --> 0:16:02.120 It's proven by the fact that they've done this to 0:16:02.120 --> 0:16:06.120 curse them multiple times. But okay, why does curse them 0:16:06.120 --> 0:16:09.160 out so much? And it's simple. Generative AI will not 0:16:09.240 --> 0:16:13.600 get big on selling consumer software without an enterprise SaaS story, 0:16:13.840 --> 0:16:19.080 they're dead And I realize, I know, okay, folks, it's 0:16:19.160 --> 0:16:21.120 kind of a little boring hearing about software as a 0:16:21.160 --> 0:16:23.840 service despite the fact that it's a huge, several hundred 0:16:23.880 --> 0:16:26.800 billion dollar industry. But this is the only place where 0:16:26.840 --> 0:16:29.760 generative AI can really make money. Companies buying hundreds of 0:16:29.800 --> 0:16:33.120 thousands of seats or how industries that rely on compute 0:16:33.160 --> 0:16:35.920 grow and without that growth, they're going nowhere. To give 0:16:35.920 --> 0:16:38.840 you some context, Netflix makes about thirty nine billion dollars 0:16:38.880 --> 0:16:42.280 a year in subscription new from consumers, and Spotify about 0:16:42.280 --> 0:16:42.920 eighteen billion. 0:16:43.360 --> 0:16:43.840 These are the. 0:16:43.800 --> 0:16:46.800 Single most popular consumer software subscriptions in the world, and 0:16:46.840 --> 0:16:49.800 open ai is fifteen point five million subscribers. Suggest that 0:16:50.000 --> 0:16:51.760 open ai can't rely on them for the kind of 0:16:51.760 --> 0:16:54.360 growth that would actually make the company worth three hundred billion. 0:16:54.160 --> 0:16:55.160 Dollars or more. 0:16:55.960 --> 0:16:58.400 Cuzer, as it stands, is the one example of a 0:16:58.440 --> 0:17:02.640 company thriving using GENERATIVEA a software company selling software, and 0:17:02.680 --> 0:17:04.920 it appears its rapid growth was the result of selling 0:17:04.960 --> 0:17:07.560 a product at a massive loss. As it stands today, 0:17:07.640 --> 0:17:10.280 Curs's product is significantly worse and it's ready it's full 0:17:10.320 --> 0:17:12.639 of people furious at the company for the changes. In 0:17:12.720 --> 0:17:15.439 simpler terms, Curser was the company that people mentioned to 0:17:15.440 --> 0:17:17.720 prove that startups could make money by building on top 0:17:17.760 --> 0:17:20.320 products on top of open AI and Anthropics models. Yet 0:17:20.359 --> 0:17:22.159 the truth is the only way to do so is 0:17:22.200 --> 0:17:24.359 to grow, and grow is to burn tons of money. 0:17:25.119 --> 0:17:27.719 While the tempting argument is to say that Curs's customers 0:17:27.760 --> 0:17:30.280 are addicted and will keep paying, this is clearly not 0:17:30.320 --> 0:17:33.600 the case, nor is it an actual business model. Like 0:17:33.680 --> 0:17:36.199 people that say this, I've never had a drug addiction, 0:17:36.240 --> 0:17:39.320 but I know people that do it. It's nothing like software. 0:17:39.400 --> 0:17:42.520 Stop making that comparison. It's insulting to the victims of addiction. 0:17:43.119 --> 0:17:46.040 But anyway, this story showed that open A and Anthropics 0:17:46.080 --> 0:17:48.320 are actually their bigger the biggest threats to their customers 0:17:48.359 --> 0:17:50.360 and will actively rent seek can punish any of their 0:17:50.359 --> 0:17:53.000 success stories, looking to loose as much as they can from. 0:17:52.840 --> 0:17:54.200 Them before they copy their products. 0:17:54.680 --> 0:17:57.880 To put it bluntly, curses growth story was a fucking lie. 0:17:58.000 --> 0:18:00.680 It reached five hundred million dollars in annulif revenue selling 0:18:00.680 --> 0:18:02.520 a product it can no longer afford to sell and 0:18:02.560 --> 0:18:06.119 could not afford to sell long term, suggesting material weakness 0:18:06.160 --> 0:18:09.400 in its business and any and all coding startups. It's 0:18:09.440 --> 0:18:11.919 also remarkable, in the shocking failure of journalism that this 0:18:12.040 --> 0:18:16.000 isn't in every single article about any sphere. I'm doing 0:18:16.040 --> 0:18:18.520 this part time? Why am I in the asshole here? 0:18:18.560 --> 0:18:22.760 Like I'm I don't know, really, though, I do have 0:18:22.800 --> 0:18:25.640 a question. Where are all the consumer AI starts? I'm 0:18:25.800 --> 0:18:26.720 genuinely serious. 0:18:27.080 --> 0:18:28.000 What have you got for me? 0:18:28.119 --> 0:18:32.280 Perplexity Perplexity. Perplexity only has one hundred and fifty million 0:18:32.320 --> 0:18:34.639 dollars in the annualized revenue, and they spent one hundred 0:18:34.640 --> 0:18:37.360 and sixty seven percent of their revenue in twenty twenty four, 0:18:37.560 --> 0:18:40.119 or fifty seven million dollars of spending on revenues of 0:18:40.200 --> 0:18:43.480 thirty four million dollars on computer services from Anthropic, Open 0:18:43.520 --> 0:18:47.879 AI and Amazon. They lost sixty eight million dollars and 0:18:47.920 --> 0:18:50.520 worse still, they still have no path to profitability and 0:18:50.560 --> 0:18:53.439 it's not even making anything new. They're a search engine, 0:18:53.480 --> 0:18:57.160 they have an AI browser. But don't worry. Professional gas 0:18:57.160 --> 0:19:00.399 bag Alex Heath just did this insane and flumm mixing 0:19:00.400 --> 0:19:04.560 interview with CEO Aravins Ravinas, who, when asked how it 0:19:04.600 --> 0:19:08.200 perplexed you would become profitable, appeared to experience what seems 0:19:08.200 --> 0:19:12.240 to be a stroke like I'm about to read something 0:19:12.240 --> 0:19:15.480 to you and it's gonna sound strange, but this is 0:19:15.520 --> 0:19:18.600