WEBVTT - OpenAI Is Not A Real Company 0:00:02.440 --> 0:00:06.600 All media, soul of an angel, body of a devil, 0:00:06.680 --> 0:00:09.280 chosen by God and perfected by science. This is better 0:00:09.320 --> 0:00:23.560 offline and I'm your host ed Zitron. Now we're working 0:00:23.560 --> 0:00:25.479 on my newsletter. Last week I was chatting with my 0:00:25.520 --> 0:00:28.640 friend friend of the show casey Kagawa about generative AI, 0:00:28.720 --> 0:00:32.000 and we kept coming back to one thought, where's the money? 0:00:32.120 --> 0:00:34.680 Where is it? Not really where is the money? Where 0:00:34.760 --> 0:00:38.640 is the money that this supposedly revolutionary, world changing industry 0:00:38.680 --> 0:00:41.400 is making and of course we'll make in the future. 0:00:42.240 --> 0:00:45.920 And the answer is simple. After hours of hours of 0:00:45.920 --> 0:00:49.000 grinding through earnings, of grinding through media articles, of grinding 0:00:49.000 --> 0:00:51.519 through all sorts of things, I just don't believe it 0:00:51.560 --> 0:00:55.680 really exists. It's real, but it's small. Generative AI lacks 0:00:55.720 --> 0:00:59.240 the basic unit economics, product market fit, or market penetration 0:00:59.320 --> 0:01:02.880 associated with any meaningful software boom, and outside of open AI, 0:01:02.960 --> 0:01:06.360 the industry may be pathetically hopelessly small or while providing 0:01:06.400 --> 0:01:10.200 few meaningful business returns and constantly losing money. I'm going 0:01:10.240 --> 0:01:12.520 to be pretty straightforward with everything I say in this 0:01:12.600 --> 0:01:15.040 two parter because the numbers and the facts in my 0:01:15.120 --> 0:01:18.000 hypotheses are pretty fucking damning of both the generative AI 0:01:18.120 --> 0:01:22.080 industry and its associated boosters. You're going to get this episode. 0:01:22.080 --> 0:01:23.920 Then there's going to be a monologue about something else 0:01:24.040 --> 0:01:26.000 or something related. I really haven't got to it yet, 0:01:26.240 --> 0:01:28.640 and then a second part which I'm recording immediately after 0:01:28.640 --> 0:01:32.120 this one, a little behind the curtain there for you. Anyway, 0:01:32.600 --> 0:01:35.360 in reporting this analysis, I've done everything I can to 0:01:35.400 --> 0:01:37.200 try and push back against my own work, and I've 0:01:37.200 --> 0:01:39.840 saw evidence to counter the things that I've seen, like 0:01:39.880 --> 0:01:43.080 the revenue and the business models of these companies. Yet 0:01:43.200 --> 0:01:45.920 in doing so, I've only become more convinced of the 0:01:46.000 --> 0:01:49.320 flimsiness of generative AI and the associated industry, and the 0:01:49.400 --> 0:01:51.640 likelihood of this bubble bursting in a way that kneecaps 0:01:51.640 --> 0:01:54.920 tech valuations for a prolonged period, or worse, hits the 0:01:54.960 --> 0:01:58.720 major stock market. Now, I really had originally written a 0:01:58.800 --> 0:02:02.040 far more jocular and perc script, but while I was 0:02:02.040 --> 0:02:04.640 writing it, I realized I really had to be blunt 0:02:04.640 --> 0:02:08.640 because what I'm describing is a systemic failure. Venture capital 0:02:08.680 --> 0:02:11.600 has propped up Open AI and Anthropic, two companies that 0:02:11.680 --> 0:02:14.520 burned are combined ten point five billion dollars in twenty 0:02:14.560 --> 0:02:16.960 twenty four, and that number is set to double or 0:02:17.040 --> 0:02:20.120 more in twenty twenty five. The tech media has allowed 0:02:20.160 --> 0:02:23.520 Sam Mortman to twist them to validate completely fictional ideas 0:02:23.880 --> 0:02:27.360 as a means of propping up this unprofitable, environmentally destructive 0:02:27.360 --> 0:02:30.600 software company, and big tech has become so disconnected from 0:02:30.680 --> 0:02:33.840 reality that it is incapable of seeing how little actual 0:02:33.880 --> 0:02:37.240 returns there are in generative AI, and they're failing. By 0:02:37.280 --> 0:02:40.200 the way. As I'll walk you through in these episodes, 0:02:40.240 --> 0:02:43.840 the GENERATIVEAI industry is very small, with the consumer market 0:02:43.840 --> 0:02:47.280 of the entire American GENERATIVEAI industry outside of chat GBT 0:02:47.680 --> 0:02:51.520 barely cracking one hundred million monthly active users, which puts 0:02:51.520 --> 0:02:53.880 them below a lot of free to play games that 0:02:53.919 --> 0:02:57.760 you get on your iPhone. Hyperscalers have already spent hundreds 0:02:57.800 --> 0:03:00.359 of billions of dollars in capital expenditures for an AI 0:03:00.400 --> 0:03:03.280 industry that has the combined monthly active users of a 0:03:03.360 --> 0:03:06.399 free to play mobile game. I really must repeat myself, 0:03:06.600 --> 0:03:11.800 it's insane, But unlike most mobile games, GENERATIVEAI doesn't really 0:03:11.840 --> 0:03:14.960 make any money. And for those of you wondering if 0:03:15.000 --> 0:03:18.040 selling access to AI models is the solution, it's important 0:03:18.040 --> 0:03:20.760 to know that open AI the market leader in generative AI, 0:03:21.040 --> 0:03:23.280 made less than a billion dollars an API calls in 0:03:23.320 --> 0:03:25.959 twenty twenty four, and that's when people plug their models 0:03:26.000 --> 0:03:27.760 them For those of you who don't understand, so it's 0:03:27.800 --> 0:03:30.320 the difference between you to load up the chat GPT 0:03:30.440 --> 0:03:34.160 app or someone has an AI generative AI like chat 0:03:34.160 --> 0:03:37.520 gpt plugged into it. Now, Microsoft pays open Ai a 0:03:37.560 --> 0:03:40.920 revenue share of twenty percent on them selling open AI's models, 0:03:40.960 --> 0:03:44.680 so two hundred billion dollars. So this means that Microsoft 0:03:44.760 --> 0:03:46.560 likely when he makes a billion dollars in revenue from 0:03:46.600 --> 0:03:50.160 API calls themselves, this is a pathetic amount of money 0:03:50.160 --> 0:03:52.800 and suggests there really isn't significant demand at all or 0:03:53.520 --> 0:03:57.080 they're not charging enough. Neither of these are great, And honestly, 0:03:57.080 --> 0:03:59.120 I'm sick and tired of hearing people prop up this 0:03:59.200 --> 0:04:02.600 fucking industry. In these episodes, I will explain as calmly 0:04:02.640 --> 0:04:05.760 as possible how the generative AI industry barely exists outside 0:04:05.800 --> 0:04:09.000 of open Ai, And honestly, in writing this, I've become 0:04:09.000 --> 0:04:12.480 completely disgusted at Silicon Valley at the waste. Why is 0:04:12.560 --> 0:04:17.200 nobody talking about the revenues, Why is nobody sharing real 0:04:17.279 --> 0:04:20.400 user numbers other than open Ai. Well, I believe it's 0:04:20.440 --> 0:04:22.960 because there isn't that much money, and there certainly aren't 0:04:22.960 --> 0:04:26.839 that many users. Nobody is making a profit from this 0:04:26.920 --> 0:04:29.920 other than consultants, and that's because this is a hype 0:04:30.080 --> 0:04:34.200 driven movement. What you see on TV and in the 0:04:34.240 --> 0:04:37.480 newspaper is not the advent of a revolutionary piece of technology. 0:04:37.760 --> 0:04:41.160 It's a cynical marketing campaign for one company, open Ai. 0:04:42.560 --> 0:04:45.440 I need you to understand how precarious this all is. 0:04:46.160 --> 0:04:48.640 So much money has been wasted propping up an industry 0:04:48.640 --> 0:04:51.799 that only burns money, that does not have mass market appeal. 0:04:52.360 --> 0:04:55.839 Chat GPT is not significant enough, or useful enough, or 0:04:55.880 --> 0:04:59.880 meaningful enough to justify spending nine billion dollars to lose 0:05:00.000 --> 0:05:02.960 five billion dollars. And yes, those are the raw economics 0:05:02.960 --> 0:05:05.960 of open Ai. Now you may say, well, Ubber lost 0:05:05.960 --> 0:05:08.600 a lot of money, didn't it, Edward? Guess what? Uber 0:05:08.640 --> 0:05:10.720 lost about six point two billion dollars in one year, 0:05:10.760 --> 0:05:12.839 and that was in twenty twenty when they couldn't run 0:05:12.839 --> 0:05:16.200 their bloody service. Uber is a very different company. I 0:05:16.279 --> 0:05:18.279 will gladly if you email me, I will explain this 0:05:18.320 --> 0:05:20.359 to you. But Uber is not a comparison. There is 0:05:20.400 --> 0:05:23.240 no comparison to what open ai is doing, what Anthropic 0:05:23.320 --> 0:05:26.080 is doing. I sound crazy as ever, but you're going 0:05:26.160 --> 0:05:29.760 to understand why when I'm done. I am deeply worried 0:05:29.760 --> 0:05:32.000 about this industry, and I need you to know why. 0:05:33.080 --> 0:05:36.039 But in this first episode, I'm going to focus on 0:05:36.040 --> 0:05:38.720 one specific thing, and that's the capitalist delusion known as 0:05:38.760 --> 0:05:42.720 open Ai, a company that encompasses almost all of the traffic, funding, 0:05:42.760 --> 0:05:46.240 and attention in generative AI, and I believe they die 0:05:46.279 --> 0:05:50.080 without a constant flow of venture capital and hyperscale or welfare. 0:05:50.120 --> 0:05:52.440 And I actually don't know why I said I believe 0:05:52.480 --> 0:05:57.440 that they will. They cannot survive without that money. But okay, 0:05:57.560 --> 0:06:01.200 let's take a second. Let's talk about open as unit economics. 0:06:02.000 --> 0:06:04.520 Putting aside the high the blaster, open Ai, as with 0:06:04.560 --> 0:06:08.640 all GENERATIVAI model developers, loses money on every single prompt 0:06:08.680 --> 0:06:12.159 and output. Its products do not scale like traditional software, 0:06:12.240 --> 0:06:14.680 in that the more users it gets, the more expensive 0:06:14.760 --> 0:06:17.039 its services are to run because its models are so 0:06:17.120 --> 0:06:21.760 compute intensive. For example, Chat GPT having four hundred million 0:06:21.800 --> 0:06:23.880 weekly active users is not the same thing as a 0:06:23.880 --> 0:06:26.839 traditional app like Instagram or Facebook having that many users. 0:06:26.920 --> 0:06:29.960 Or indeed Uber. The cost of serving a regular user 0:06:29.960 --> 0:06:32.800 of an app like Instagram is significantly smaller because these 0:06:32.880 --> 0:06:37.720 are effectively websites with connecting APIs, images, videos, and user interactions. 0:06:38.000 --> 0:06:41.120 These platforms aren't innately compute heavy, and so you don't 0:06:41.160 --> 0:06:44.520 need to have the same level of infrastructure to support 0:06:44.560 --> 0:06:48.440 the same amount of people. Conversely, GENERATIVAI requires expensive to 0:06:48.440 --> 0:06:51.560 buy and expensive to run and expensive to maintain graphics 0:06:51.600 --> 0:06:55.960 processing units GPUs, both for inference and training the models themselves. 0:06:56.480 --> 0:06:58.720 These GPUs must be run at full tilt for both 0:06:58.720 --> 0:07:02.039 inference and training, which organs their lifespan while also consuming 0:07:02.120 --> 0:07:05.000 ungodly amounts of energy. And by the way, inference is 0:07:05.080 --> 0:07:07.440 just the thing that happens when you tell chat GPTs. 0:07:07.480 --> 0:07:09.200 I think it infers the meaning of the prompt. And 0:07:09.640 --> 0:07:11.640 the training is what they do when they throw all 0:07:11.640 --> 0:07:14.560 the training data to make the model huh smart. Not 0:07:14.600 --> 0:07:17.520 really though, And by the way, surrounding the GPU in 0:07:17.560 --> 0:07:19.320 there isn't that the GPUs just kind of hang out. 0:07:19.320 --> 0:07:21.760 There's the rest of a computer, which is usually highly 0:07:21.760 --> 0:07:25.600 specked and incredibly difficult to cool and thus very very expensive. 0:07:26.480 --> 0:07:30.000 These generative AI models also require endless amounts of training 0:07:30.080 --> 0:07:32.600 data and supplies of that training data have been running 0:07:32.600 --> 0:07:35.640 out for a long time. While synthetic data might bridge 0:07:35.640 --> 0:07:38.040 some of the gap, there are likely diminishing returns due 0:07:38.040 --> 0:07:40.080 to the sheer amount of data necessary to make a large, 0:07:40.160 --> 0:07:43.200 loud and quich model even larger, as much as more 0:07:43.240 --> 0:07:47.920 than four times the size of the Internet. This is insane. 0:07:48.120 --> 0:07:50.960 There is not enough data, and it already kind of sucks, 0:07:51.120 --> 0:07:55.120 and it's not getting better. These companies also must spend 0:07:55.240 --> 0:07:57.760 hundreds of millions of dollars on salaries to attract and 0:07:57.800 --> 0:08:00.880 retain AI talent, as much as one point five billion 0:08:00.920 --> 0:08:03.119 dollars a year in open AI's case, and that's before 0:08:03.160 --> 0:08:07.120 stock based compensation. In twenty sixteen, Microsoft claimed that top 0:08:07.160 --> 0:08:09.720 AI talent could cost as much as an NFL quarterback 0:08:09.720 --> 0:08:12.360 to hire, and that some has likely only increased since 0:08:12.400 --> 0:08:15.280 then given the GENERATIVAI frenzy and the fact they're overpaying 0:08:15.360 --> 0:08:18.360 quarterbacks as in the side. One analyst told The Wall 0:08:18.360 --> 0:08:21.000 Street Journal that companies running generative AI models could and 0:08:21.080 --> 0:08:24.560 I quote be utilizing half their capital expenditures because all 0:08:24.600 --> 0:08:29.280 of these things could break down. That's that's really bad. 0:08:29.400 --> 0:08:32.079 These costs are not a burden on open ai or Anthropic, 0:08:32.120 --> 0:08:35.520 but they absolutely are on Microsoft, Google and Amazon. This 0:08:35.600 --> 0:08:39.000 shit's crazy anyway. As a result of the costs of 0:08:39.080 --> 0:08:41.520 running these services, a free user of Chat GPT is 0:08:41.559 --> 0:08:44.680 a cost burden on open Ai, as is every single 0:08:44.720 --> 0:08:48.520 free customer of Google's Gemini, Anthropics, Clawed, Perplexity, or any 0:08:48.520 --> 0:08:52.200 other generative AI company. Said costs are also so severe 0:08:52.200 --> 0:08:56.440 that even paying customers lose these companies' money. Even the 0:08:56.440 --> 0:08:58.680 most successful company in the business appears to have no 0:08:58.760 --> 0:09:01.520 way to stop burning cap And as I'll explain, there's 0:09:01.559 --> 0:09:05.560 only really one real company in the industry, Open Ai, 0:09:06.160 --> 0:09:09.640 and open ai is not a real business. But let's 0:09:09.720 --> 0:09:12.559 start with a really, really important fact. If you forget 0:09:12.600 --> 0:09:15.600 everything I say, I want you to remember this. Open 0:09:15.640 --> 0:09:19.160 Ai spent nine billion dollars to make just under four 0:09:19.200 --> 0:09:22.079 billion dollars in twenty twenty four, and the entirety of 0:09:22.120 --> 0:09:25.720 their revenue that's about four billion dollars, is spent on compute, 0:09:25.840 --> 0:09:28.360 two billion dollars to run models and three billion dollars 0:09:28.440 --> 0:09:31.680 to train them. That is completely and utterly fucking insane. 0:09:31.880 --> 0:09:35.240 That is bonkers. That is crazy. That is completely nuts. 0:09:35.400 --> 0:09:37.680 This is not a real company. It is insane. We're 0:09:37.720 --> 0:09:40.880 allowing this. Everyone should be screaming this at everyone. We 0:09:40.960 --> 0:09:43.719 live in an alternate reality where this is acceptable. There 0:09:43.760 --> 0:09:46.320 has been no precedent for this. Not Amazon Web Services, 0:09:46.360 --> 0:09:49.240 not Uber, not anyone. No one has done this, and 0:09:49.280 --> 0:09:52.000 it's sickening and wasteful that we continue to. And in 0:09:52.040 --> 0:09:54.600 the past I've repeatedly said that open ai lost five 0:09:54.600 --> 0:09:57.440 billion dollars after revenue. Now that is true. By the way, 0:09:57.720 --> 0:10:00.320 that is completely true. They made money and they lost money, 0:10:00.320 --> 0:10:03.920 but ended up losing five billion either either way. However, 0:10:04.000 --> 0:10:06.640 I really just I can't in good conscience suggest that 0:10:06.679 --> 0:10:10.080 open ai only spent five billion dollars. It's time to 0:10:10.080 --> 0:10:12.319 be honest about these numbers. While it's fair to say 0:10:12.360 --> 0:10:15.880 that their net losses are five billion, they spent nine 0:10:15.920 --> 0:10:30.880 billion dollars to lose five billion dollars. Let's really get 0:10:30.880 --> 0:10:33.320 down into the nitty gritty of these numbers. So, as 0:10:33.360 --> 0:10:36.200 discussed previously, according to the reporting by the Information, open 0:10:36.240 --> 0:10:38.320 AI's revenue was likely somewhere in the region of four 0:10:38.320 --> 0:10:41.840 billion dollars in twenty twenty four. Their burn ray, according 0:10:41.880 --> 0:10:45.000 to the Information, was five billion dollars after revenue in 0:10:45.000 --> 0:10:48.560 twenty twenty four, excluding stock based compensation, which open Ai, 0:10:48.760 --> 0:10:51.200 like other startups, uses as a means of compensation on 0:10:51.240 --> 0:10:54.080 top of cash. Nevertheless, the more it gives away, the 0:10:54.160 --> 0:10:57.439 less it has for capital raises. And these are technically costs, 0:10:57.520 --> 0:11:00.680 though they're not real money unless there's a liquidity event, 0:11:00.720 --> 0:11:02.280 but that's our you don't need to worry about that. 0:11:02.960 --> 0:11:05.280 To put this in blunt terms, based on reporting by 0:11:05.280 --> 0:11:08.319 the Information, and I'm repeating myself here, but I really 0:11:08.360 --> 0:11:11.720 need you to remember this, running open Ai cost nine 0:11:12.080 --> 0:11:15.800 billion dollars in twenty twenty four. The cost of the computer, 0:11:16.200 --> 0:11:19.360 to the compute to train models alone three billion dollars 0:11:19.440 --> 0:11:23.480 obliterates the entirety of their subscription revenue, which was about 0:11:23.480 --> 0:11:25.559 three billion dollars by the way, and the compute from 0:11:25.600 --> 0:11:28.160 running models two billion dollars takes the rest and then some. 0:11:28.600 --> 0:11:31.920 They actually end up losing an extra billion on top 0:11:31.920 --> 0:11:34.040 of that. Sam Altman's net worth is a billion dollars. 0:11:34.080 --> 0:11:36.600 By the way, Casey Gogawa has now used this as 0:11:36.640 --> 0:11:39.560 the the Altman Index, so it's like you've lost one 0:11:39.600 --> 0:11:43.200 Sam Altman. That's a billion dollars. But just to be clear, 0:11:43.240 --> 0:11:45.720 it doesn't just cost more to run open Ai than 0:11:45.720 --> 0:11:47.679 they make. It costs them a billion dollars more than 0:11:47.679 --> 0:11:50.000 the entirety of their revenue to run the software they 0:11:50.040 --> 0:11:54.320 sell before any other costs. Why are we not more 0:11:54.360 --> 0:11:58.240 concerned about this company now? Something else to note is 0:11:58.240 --> 0:12:00.800 that open ai also spends another line amount of money 0:12:00.800 --> 0:12:03.560 on salaries, over seven hundred million dollars in twenty twenty 0:12:03.559 --> 0:12:07.000 four before you consider that compensation from stock, a number 0:12:07.000 --> 0:12:09.839 that will also have to increase because open ai is growing, 0:12:09.880 --> 0:12:12.320 which means hiring as many people as possible, and they're 0:12:12.320 --> 0:12:15.439 paying through the nose for them. But let's talk about 0:12:15.440 --> 0:12:18.760 how open ai makes money. Open Ai sells access to 0:12:18.800 --> 0:12:21.719 its models via its API and selling premium subscriptions to 0:12:21.800 --> 0:12:25.000 chat gpt. The majority of its revenue over seventy percent, 0:12:25.080 --> 0:12:28.560 comes from subscriptions to premium versions of chat GPT. The 0:12:28.640 --> 0:12:31.600 Information also reported that open ai now has fifteen point 0:12:31.679 --> 0:12:34.760 five million paying customers, though it's unclear what level of 0:12:34.800 --> 0:12:37.439 the service they're paying for or how sticky these customers 0:12:37.480 --> 0:12:39.319 are is in how luckily they are to stick around, 0:12:39.440 --> 0:12:41.760 or the cost of acquiring these customers are really any 0:12:41.760 --> 0:12:44.400 other metric to tell them tell us how valuable these 0:12:44.400 --> 0:12:47.960 customers are to the bottom line. Nevertheless, open ai loses 0:12:48.000 --> 0:12:52.200 money on every single paying customer, just like its free users. 0:12:52.760 --> 0:12:56.880 Increasing paid subscribers to open ai services somehow increases open 0:12:56.920 --> 0:13:00.640 AI's burn rate. This is not a real company. Now, 0:13:00.640 --> 0:13:02.800 The New York Times reports the open ai projects it 0:13:02.840 --> 0:13:05.800 will make eleven point six billion dollars in twenty twenty five, 0:13:06.000 --> 0:13:08.120 and assuming that open ai burns at the same rate 0:13:08.200 --> 0:13:10.240 it did in twenty twenty four, spending two point two 0:13:10.360 --> 0:13:12.640 five dollars to make one dollar, open ai is on 0:13:12.760 --> 0:13:14.920 course to bone over twenty six billion dollars in twenty 0:13:14.960 --> 0:13:17.960 twenty five for a loss of fourteen point four billion dollars. 0:13:18.360 --> 0:13:21.439 Who knows what their actual cost will be. Now you've 0:13:21.480 --> 0:13:24.000 probably heard about soft Bank coming in. Soft Bank's going 0:13:24.080 --> 0:13:25.680 to feed the money, and soft banks said they're going 0:13:25.760 --> 0:13:28.160 to spend money on this, that, and the other. That 0:13:28.280 --> 0:13:31.960 round has not closed yet. Masayoshi Son a complete fucking 0:13:32.000 --> 0:13:35.280 idiot who's lost thirty odd billion dollars for soft Bank, 0:13:35.320 --> 0:13:40.080 the Japanese make a conglomerate. He's dedicating billions of dollars 0:13:40.080 --> 0:13:43.360 of revenue to buying open ai services, unless this is 0:13:43.400 --> 0:13:45.640 a straight up trade where he's just sending money before 0:13:45.679 --> 0:13:47.920 the services come in. I don't know if it happens, 0:13:48.320 --> 0:13:50.240 and I'm going to get into things like agents later, 0:13:50.280 --> 0:13:53.360 but the information reported that open ai expects to make 0:13:53.400 --> 0:13:57.200 three billion dollars in revenue from agents. By the end 0:13:57.200 --> 0:13:59.240 of this episode. You're going to realize how fucking stupid 0:13:59.240 --> 0:14:02.719 that sounds. We'll get their layer. It's also important to 0:14:02.800 --> 0:14:05.160 note that open AI's costs are partially subsidized by its 0:14:05.200 --> 0:14:08.200 relationship with Microsoft, which provides cloud compute credits for its 0:14:08.320 --> 0:14:11.439 zero cloud service. Not super technical, it's just when they 0:14:11.440 --> 0:14:15.080 host people's software and files and such and the compute 0:14:15.160 --> 0:14:18.720 to run these models, and they also offer this a steep, 0:14:18.760 --> 0:14:21.640 steep discount to open Ai. Or put another way, it's 0:14:21.680 --> 0:14:23.640 like open ai got paid with air miles, but the 0:14:23.680 --> 0:14:26.000 airline lowered the redemption cost of booking a flight with 0:14:26.000 --> 0:14:28.240 those air miles, allowing it to take more flights than 0:14:28.280 --> 0:14:31.800 any other person with the equivalent amount of points. Until recently, 0:14:31.880 --> 0:14:35.400 open ai exclusively used Microsoft as Zuo services to train 0:14:35.520 --> 0:14:38.120 host and run its models, but recent changes to its 0:14:38.120 --> 0:14:40.240 deal means that open ai is now working with Oracle 0:14:40.240 --> 0:14:42.240 to build up further data centers to train host and 0:14:42.320 --> 0:14:45.680 run its models. It's unclear whether this partnership will work 0:14:45.680 --> 0:14:47.720 in the same way as the Microsoft deal with open 0:14:47.720 --> 0:14:51.280 Ai provided credits and discounts like before. If not, open 0:14:51.320 --> 0:14:54.920 AI's operating costs will only go up. For previous reporting 0:14:54.960 --> 0:14:57.680 from the information, open Ai pays just over twenty five 0:14:57.680 --> 0:15:00.240 percent of the cost of a zero's GPU compute part 0:15:00.280 --> 0:15:02.760 of their deal with Microsoft, and that's about a dollar 0:15:02.840 --> 0:15:05.560 thirty per GPU per hour versus the regular is your 0:15:05.600 --> 0:15:07.840 cost of three dollars and forty cents to four dollars 0:15:07.920 --> 0:15:10.480 an hour. I know that this sounds really technical, but 0:15:11.400 --> 0:15:14.160 in very short they're getting a sweet deal from Microsoft, 0:15:14.200 --> 0:15:16.640 and if anything happens in that, they're completely fucked. They're 0:15:16.640 --> 0:15:19.760 fucked anyway. They don't have They're burning billions of dollars. 0:15:19.800 --> 0:15:23.680 It's insane. But let's talk about user numbers, because open 0:15:23.680 --> 0:15:26.360 ai has quite a few. They recently announced that they 0:15:26.400 --> 0:15:30.320 have four hundred million weekly active users. Now, weekly active 0:15:30.400 --> 0:15:32.720 users is a wanky number and a very strange one 0:15:32.720 --> 0:15:35.520 for a company like this. Open Ai may pretend to 0:15:35.520 --> 0:15:38.120 be a consumer company, but the majority of their revenue 0:15:38.120 --> 0:15:41.000 comes from monthly subscriptions, making them kind of a cloud 0:15:41.040 --> 0:15:45.160 software company. Classically cloud software companies report monthly active users. 0:15:45.440 --> 0:15:48.880 That way, you can, I don't know, compare one number 0:15:48.920 --> 0:15:50.920 which is the amount of active users you have with 0:15:51.000 --> 0:15:52.880 the paid users you have, and then say, oh, that's 0:15:52.880 --> 0:15:55.200 a good business. That's a good business, right there. Man, 0:15:55.960 --> 0:15:58.480 guess what open ai isn't given their monthly active users. 0:15:58.760 --> 0:16:02.560 Don't worry, I might estimated it. When I asked open 0:16:02.600 --> 0:16:04.800 ai to define what a weekly active user was, it 0:16:04.840 --> 0:16:07.120 responded by pointing me to a tweet by chief operating 0:16:07.120 --> 0:16:09.920 officer Brad light Cap that said chat gpt recently crossed 0:16:09.920 --> 0:16:12.920 four one hundred million weekly active users. We feel very 0:16:12.960 --> 0:16:15.200 fortunate serve five percent of the world every week. What 0:16:15.320 --> 0:16:19.840 a fucking liar. It's extremely questionable that open ai refuses 0:16:19.880 --> 0:16:21.880 to define this core metric. By the way, and without 0:16:21.880 --> 0:16:23.800 a definition, in my opinion, there is no way to 0:16:23.840 --> 0:16:27.880 assume anything other than open ai is actively gaming its numbers. Now. 0:16:27.920 --> 0:16:30.800 There's likely two reasons they focus on weekly active users. 0:16:31.000 --> 0:16:33.320 One has described these numbers are easy to game, you 0:16:33.320 --> 0:16:36.360 can choose any seven day period. And also the majority 0:16:36.360 --> 0:16:39.120 of open AI's revenue comes from paid subscriptions to chat gpt. 0:16:39.480 --> 0:16:41.800 And that latter point is crucial because it suggests open 0:16:41.840 --> 0:16:44.000 ai is not doing anywhere near as well as it 0:16:44.040 --> 0:16:47.000 seems based on the very basic metrics used to measure 0:16:47.000 --> 0:16:50.200 the success of a software product. The information reported on 0:16:50.280 --> 0:16:53.080 January thirty first, open Ai, like I mentioned, had fifteen 0:16:53.080 --> 0:16:56.840 point five million monthly paying subscribers, and they added in 0:16:56.880 --> 0:16:59.480 this piece that this was less than a five percent 0:16:59.520 --> 0:17:03.480 conversion rate of open aiy's weekly active users, a statement 0:17:03.520 --> 0:17:05.880 that's kind of like dividing the number fifty two budd 0:17:05.880 --> 0:17:08.720 a letter A. This is not an honest or reasonable 0:17:08.760 --> 0:17:11.959 way to evaluate the success of chat GPT's still unprofitable 0:17:11.960 --> 0:17:15.520 software business, because the actual metric, like I mentioned, would 0:17:15.520 --> 0:17:19.200 have been to definde paying subscribers by monthly active users 0:17:19.280 --> 0:17:21.000 or the other way around. I guess a number that 0:17:21.040 --> 0:17:24.199 would be considerably higher than four hundred million. And the 0:17:24.240 --> 0:17:25.720 reason they don't need to do that, by the way, 0:17:25.760 --> 0:17:27.840 is because you would divide them and see that they 0:17:27.840 --> 0:17:30.119 have a piss poor conversion rate. Good conversion rate is 0:17:30.359 --> 0:17:32.920 way higher than five percent, by the way, and theirs 0:17:33.000 --> 0:17:35.600 is definitely lower. But don't worry. I'm a sneaky little shit. 0:17:35.680 --> 0:17:37.119 So I went and looked some stuff up, and I 0:17:37.200 --> 0:17:39.840 talked to some people based on data from the market 0:17:39.840 --> 0:17:43.000 intelligence firm Center Tower. Open AI's chat gpt app on 0:17:43.040 --> 0:17:45.440 Android and iOS is estimated to have more than three 0:17:45.560 --> 0:17:48.600 hundred and thirty nine million monthly active users, and based 0:17:48.600 --> 0:17:52.160 on traffic data for market intelligence company similar web chatgpt 0:17:52.480 --> 0:17:55.000 dot com had two hundred and forty six million unique 0:17:55.000 --> 0:17:57.720 monthly visitors, and these were in January twenty twenty five. 0:17:58.240 --> 0:18:00.679 There's likely some crossover with people using both the mobile 0:18:00.680 --> 0:18:03.359 and web interfaces, though how big nut group is is 0:18:03.440 --> 0:18:06.760 kind of hard to tell and remains uncertain. Though every 0:18:06.840 --> 0:18:09.840 single person that visits chatgpt dot com might not become 0:18:09.880 --> 0:18:12.399 a user, it's safe to assume that chat GPT's monthly 0:18:12.480 --> 0:18:14.879 active users are somewhere in the region of five hundred 0:18:15.119 --> 0:18:18.879 to six hundred million. That's good, right, Its actual users 0:18:18.880 --> 0:18:22.359 are higher than officially claimed, right, that's good. No, it's bad. 0:18:22.720 --> 0:18:25.760 First of all, each user that uses chat gpt for 0:18:25.760 --> 0:18:28.360 free is a drain on the company, whether they're free 0:18:28.440 --> 0:18:30.719 or not, honestly, but either way, their free ones definitely are. 0:18:31.040 --> 0:18:33.320 It would also suggest that the real conversion rate is 0:18:33.359 --> 0:18:36.120 somewhere in the neighborhood of two point five eight three 0:18:36.160 --> 0:18:39.159 percent from free paid users on chat GPT, which is 0:18:39.200 --> 0:18:42.520 astonishingly bad, and it's a fact that's made worse by 0:18:42.560 --> 0:18:44.600 the fact that every single user, regardless of whether they 0:18:44.640 --> 0:18:48.320 pay or not, loses the money either way. And while 0:18:48.359 --> 0:18:50.880 it's quite common for Silicon Valley companies to play fast 0:18:50.880 --> 0:18:54.040 than loose with metrics, this particularly one is well was 0:18:54.080 --> 0:18:56.920 deeply concerning, and I hypothesize that open ay is choosing 0:18:56.960 --> 0:18:59.280 to go with the weekly versus monthly active users in 0:18:59.320 --> 0:19:02.320 an intentional attempt to avoid people calculating the conversion rate 0:19:02.320 --> 0:19:05.720 of its subscription products. As I will continue to repeat, 0:19:05.840 --> 0:19:09.520 these subscription products lose the company money every single time. 0:19:10.280 --> 0:19:12.840 Now let's talk product strategy, shall we, because I don't 0:19:12.840 --> 0:19:16.400 think open ai really has one. Open Ai makes most 0:19:16.400 --> 0:19:19.359 of its money from subscriptions approximately three billion dollars in 0:19:19.359 --> 0:19:21.800 twenty twenty four, and the rest on API access to 0:19:21.840 --> 0:19:25.080 its models approximately a billion. As a result, open Ai 0:19:25.160 --> 0:19:28.000 is chosen to monetize chat GBT in its associated products 0:19:28.040 --> 0:19:30.200 in it all you can eat software subscription model, or 0:19:30.240 --> 0:19:33.359 otherwise make money by other people productizing in and just 0:19:33.400 --> 0:19:35.560 to be clear, in both of these scenarios, open Ai 0:19:35.720 --> 0:19:39.240 loses money on every transaction. Open AI's products are not 0:19:39.320 --> 0:19:42.840 fundamentally differentiated or interesting enough to be sold separately. It 0:19:42.920 --> 0:19:45.880 has failed, as with the rest of the generative AI industry, 0:19:45.960 --> 0:19:48.760 to meaningfully productize its models due to the massive training 0:19:48.760 --> 0:19:51.600 and operational costs and a lack of any meaningful killer 0:19:51.600 --> 0:19:55.080 app use cases for large language models. The only product 0:19:55.080 --> 0:19:57.920 that OpenAI has succeeded in scaling to the mass market 0:19:57.960 --> 0:20:00.439 is the free version of chat GBT, which loses the 0:20:00.440 --> 0:20:04.040 company money with every single prompt and output. This scale 0:20:04.119 --> 0:20:06.320 isn't a result of any kind of product market fit, 0:20:06.359 --> 0:20:09.360 by the way, It's entirely media driven, with reporters making 0:20:09.480 --> 0:20:13.040 chat GPT synonymous with artificial intelligence, a thing they regularly 0:20:13.040 --> 0:20:16.000 write about without thinking. As a result, I do not 0:20:16.119 --> 0:20:19.399 believe that the generative AI industry is real. It's not 0:20:19.480 --> 0:20:21.879 a real industry, which I will define as one with 0:20:21.960 --> 0:20:26.560 multiple competitive companies with sustainable or otherwise growing revenue streams 0:20:26.560 --> 0:20:29.880 and meaningful products with actual market penetration. And I feel 0:20:29.880 --> 0:20:32.760 this way because this market is entirely subsidized by a 0:20:32.760 --> 0:20:36.160 combination of venture capital and hyperscaler cloud credits, and well 0:20:36.200 --> 0:20:39.760 real money. I guess chat GPT is popular because it's 0:20:39.800 --> 0:20:42.680 the only well known product one that's mentioned in basically 0:20:42.720 --> 0:20:45.800 every article on ai. If this were a real industry, 0:20:45.840 --> 0:20:48.679 other competitors would also be mentioned all the time. They 0:20:48.680 --> 0:20:52.600