WEBVTT - The Case Against Generative AI (Part 4) 0:00:02.440 --> 0:00:03.199 Zoe Media. 0:00:04.400 --> 0:00:05.360 All Right, we're back. 0:00:05.440 --> 0:00:20.000 I'm ed Zetron and this is better offline. This is 0:00:20.000 --> 0:00:22.280 the fourth and final episode of this four part series 0:00:22.320 --> 0:00:26.480 where I dissect in excruciating comprehensive detail the AI bubble 0:00:26.520 --> 0:00:28.840 and tell you why I think its implosion is inevitable. 0:00:29.480 --> 0:00:31.520 We're at the end, which is fitting because in this 0:00:31.600 --> 0:00:33.559 episode we're going to talk about why and how this 0:00:33.760 --> 0:00:37.440 entire sham dies again. If you jumping in now, start 0:00:37.440 --> 0:00:39.080 from the beginning. We talk about a lot of things 0:00:39.120 --> 0:00:40.920 and it all kind of threads together. I know, I 0:00:40.960 --> 0:00:42.720 know you want to hear my explanation of how open 0:00:42.760 --> 0:00:45.080 ai is fucked with a capital fuck. But we've got 0:00:45.120 --> 0:00:48.160 to lay the groundwork to get there. Okay, got up, good, Good, 0:00:48.240 --> 0:00:51.279 We're good. We're god, You're ready, We're good, You're got yep, yep. 0:00:51.200 --> 0:00:52.040 Okay, we'll begin. 0:00:52.720 --> 0:00:55.200 One of the comfortable lies that people tell themselves is 0:00:55.240 --> 0:00:57.200 that the AI bubble is similar to the fiber boom 0:00:57.240 --> 0:00:59.279 or the dot com boom or uber, or that we're 0:00:59.280 --> 0:01:01.800 in the growth stay, that this is what software companies do. 0:01:01.880 --> 0:01:03.720 They spend a bunch of money, then pull the profit lever. 0:01:04.720 --> 0:01:07.560 The thing is, this is nothing like anything you've ever 0:01:07.600 --> 0:01:09.880 seen before, because this is the dumbest shit the tech 0:01:09.920 --> 0:01:13.959 industry has ever ever done. AI data centers are nothing 0:01:14.040 --> 0:01:16.679 like fiber because there are very few actual use cases 0:01:16.680 --> 0:01:19.200 for these GPUs outside of AI, and none of them 0:01:19.200 --> 0:01:22.840 are remotely hyper scale revenue drivers. As I discussed a 0:01:22.840 --> 0:01:25.400 month or so ago, data center development accounted for more 0:01:25.440 --> 0:01:28.520 of America's GDP growth than all consumers spending combined in 0:01:28.560 --> 0:01:29.280 the first half of. 0:01:29.240 --> 0:01:30.119 Twenty twenty five. 0:01:30.360 --> 0:01:32.679 And there really isn't any demand for AI in general, 0:01:32.760 --> 0:01:34.800 let alone at the scale that these hundreds of billions 0:01:34.840 --> 0:01:37.960 of dollars are being sunk into. The conservative estimate of 0:01:38.040 --> 0:01:41.080 capital expenditures related to data centers is around four hundred 0:01:41.080 --> 0:01:43.760 billion dollars, but given the fifty billion dollar in a 0:01:43.880 --> 0:01:46.240 quarter in private equity invested, I'm going to guess it 0:01:46.240 --> 0:01:48.560 breaks half a trillion, all to build capacity for an 0:01:48.560 --> 0:01:51.680 industry yet to prove itself. And this whole in video 0:01:51.720 --> 0:01:54.680 OpenAI one hundred billion dollar funding news should only fill 0:01:54.720 --> 0:01:57.360 you full of dread. But also it isn't fucking finalized. 0:01:57.360 --> 0:02:02.240 Stop reporting as if it's done, it's fucking anyway good. 0:02:02.680 --> 0:02:05.360 According to CNBC and I quote the initial ten billion 0:02:05.400 --> 0:02:07.760 dollar tranch is locked in at a five hundred billion 0:02:07.760 --> 0:02:10.200 dollar valuation and expected to close within a month or 0:02:10.200 --> 0:02:13.359 so once the transaction has been finalized, with successive ten 0:02:13.360 --> 0:02:16.240 billion dollar rounds planned, each to be priced at the 0:02:16.240 --> 0:02:19.040 company's then current valuation. Is new capacity comes online, what 0:02:19.080 --> 0:02:21.440 a horrible deal. They can just raise whatever they can, 0:02:21.560 --> 0:02:24.640 just go back to video. But let me just be clear. 0:02:24.919 --> 0:02:26.919 Open Ai has written a lot of checks that neither 0:02:26.960 --> 0:02:28.720 it nor anyone else can cash, and for it to 0:02:28.720 --> 0:02:31.720 survive it needs to raise more than, honestly about five 0:02:31.800 --> 0:02:35.240 hundred billion dollars and maybe another four hundred billion dollars 0:02:35.280 --> 0:02:38.800 and other people paying for stuff. It's astonishing, really, and 0:02:38.880 --> 0:02:42.120 no point is anyone asking how exactly open ai builds 0:02:42.120 --> 0:02:44.160 the data centers to fill for of the GPUs to 0:02:44.160 --> 0:02:46.840 get the rest of Nvidia's funding. In fact, I'm genuinely 0:02:46.840 --> 0:02:49.160 shocked and a little disgusted by how poorly this story 0:02:49.160 --> 0:02:49.800 has been told. 0:02:50.120 --> 0:02:51.120 Let's go point by point. 0:02:52.040 --> 0:02:54.160 Ten billion dollars is not enough for open ai to 0:02:54.160 --> 0:02:56.480 build a data center. The one point one to one 0:02:56.480 --> 0:02:59.920 point two Gigawa Gabiline Texas data center being built by 0:03:00.040 --> 0:03:02.960 Oracle and Cruiser for open Ai required fifteen billion dollars 0:03:03.040 --> 0:03:05.720 in debt, and that's not including the forty billion dollars 0:03:05.760 --> 0:03:08.079 of chips needed to power it. Though a question whether 0:03:08.120 --> 0:03:10.120 all of those are going into Abilene or somewhere else, 0:03:10.120 --> 0:03:13.640 which you get to later. Open Ai can also not 0:03:13.720 --> 0:03:16.560 afford to pay for everything it's promised. Open Ai will 0:03:16.560 --> 0:03:18.799 burn they say one hundred and fifteen billion dollars in 0:03:18.800 --> 0:03:20.680 the next four years ago into the information, but I 0:03:20.720 --> 0:03:23.560 believe the company intentionally leaked those numbers before the announcement 0:03:23.560 --> 0:03:26.600 of their three hundred billion dollar deal with Oracle, as 0:03:26.639 --> 0:03:28.359 when you take into account the numbers share but the 0:03:28.440 --> 0:03:31.360 information which involved them burning forty seven billion dollars in 0:03:31.360 --> 0:03:33.280 twenty twenty eight a year when it open ai is 0:03:33.280 --> 0:03:36.120 meant to make seventy billion dollars in payments to compute 0:03:36.160 --> 0:03:38.840 to Oracle, not for twenty eight billion dollars to Microsoft 0:03:38.920 --> 0:03:41.240 or any other cost. There's no space for the three 0:03:41.320 --> 0:03:44.320 hundred billion dollars to go, and there's actually another one 0:03:44.400 --> 0:03:47.280 hundred billion dollars they're promising for backup compute as well. 0:03:47.600 --> 0:03:49.680 On top of all of this, open ai is still 0:03:49.720 --> 0:03:52.200 yet to convert to a for profit entity, and must 0:03:52.240 --> 0:03:54.360 do so by the end of twenty twenty five, or 0:03:54.360 --> 0:03:57.440 they'll lose twenty billion dollars in funding from SoftBank. A 0:03:57.480 --> 0:04:00.119 few weeks ago, Microsoft and open Ai co published an 0:04:00.120 --> 0:04:03.320 announcement that said they'd signed a non binding memorandum of 0:04:03.400 --> 0:04:06.000 understanding for the next phase of their partnership, which the 0:04:06.040 --> 0:04:08.840 media quickly took to mean the deal was done. No 0:04:09.000 --> 0:04:11.920 deal has been done. This is an announcement of nothing. 0:04:12.840 --> 0:04:15.000 On top of all of this, venture capital might run 0:04:15.040 --> 0:04:17.760 out at the current rate of investment in around six quarters, 0:04:17.760 --> 0:04:20.719 and open ai needs more investment than ever. The Information 0:04:20.839 --> 0:04:24.520 recently published some worrying data from venture capitalist John Sokoda 0:04:24.560 --> 0:04:26.600 that a Today's clip the industry would run out of 0:04:26.600 --> 0:04:29.240 money in six quarters, adding that the money wouldn't run 0:04:29.279 --> 0:04:30.960 out until the end of twenty twenty eight if it 0:04:31.000 --> 0:04:31.840 wasn't for open. 0:04:31.600 --> 0:04:32.560 Ai and Anthropic. 0:04:33.000 --> 0:04:35.760 In a very real sense, open ai threatens the future 0:04:35.760 --> 0:04:39.240 of available capital for the tech industry. Well, the good 0:04:39.240 --> 0:04:41.240 news is that they're going to make lots of money, right. 0:04:41.520 --> 0:04:44.119 I mean, open Ai leaked that they'll make two hundred 0:04:44.200 --> 0:04:48.400 billion dollars in four years time. I mean, if you're concussed, 0:04:48.720 --> 0:04:50.839 I can see how you believe that. Let's go, I'm 0:04:50.839 --> 0:04:52.680 going to read you a bunch of numbers. I know 0:04:52.720 --> 0:04:55.040 it's kind of going to be annoying to hear this. 0:04:56.200 --> 0:04:57.760 For the next thirty second, I'm going to just throw 0:04:57.800 --> 0:04:59.800 figures at you, but pay attention because it's important to 0:04:59.800 --> 0:05:02.240 really us how insane this is. So. In twenty twenty five, 0:05:02.320 --> 0:05:05.120 open ai projects to make thirteen billion dollars and have 0:05:05.200 --> 0:05:08.159 free cash flow of negative nine billion dollars. In twenty 0:05:08.160 --> 0:05:10.920 twenty six, open ai will make twenty nine billion dollars 0:05:10.960 --> 0:05:13.440 or thirty billion dollars, but have free cash flow of 0:05:13.480 --> 0:05:17.360 negative seventeen billion. In twenty twenty seven, open ai will 0:05:17.360 --> 0:05:20.200 make sixty billion dollars but have negative cash flow of 0:05:20.560 --> 0:05:23.600 negative thirty five billion dollars. And in twenty twenty eight, 0:05:23.640 --> 0:05:25.640 open ai will make one hundred billion dollars but have 0:05:25.720 --> 0:05:28.520 free cash flow of negative forty seven billion dollars. In 0:05:28.560 --> 0:05:30.680 twenty twenty nine, open ai will make one hundred and 0:05:30.720 --> 0:05:32.880 forty five billion dollars but a free cash flow of 0:05:32.960 --> 0:05:36.440 negative eight billion dollars. Somehow, I just want to reiterate 0:05:36.480 --> 0:05:39.640 there open AI's projection. See it reduce its negative cash 0:05:39.640 --> 0:05:42.120 flow by thirty nine billion dollars or one and a 0:05:42.120 --> 0:05:44.800 half times three m's revenue from the last financial year 0:05:44.920 --> 0:05:47.200 in a single year, while also increasing their revenue by 0:05:47.200 --> 0:05:49.960 a similar amount, give or take a few billion. Who's counting, 0:05:50.279 --> 0:05:52.920 certainly not Sarah Fryer, the CFO of open Ai. 0:05:53.440 --> 0:05:54.719 It's also fucking stupid. 0:05:54.720 --> 0:05:57.960 But don't worry. I know, I know some of you 0:05:58.120 --> 0:06:00.760 have been worried. You see big strong men with tears. 0:06:00.839 --> 0:06:03.520 Now is going to be saying cirsa open Ai. They're 0:06:03.520 --> 0:06:05.120 not gonna They're going to have enough money. 0:06:05.160 --> 0:06:06.320 They're gonna die. But don't worry. 0:06:06.360 --> 0:06:09.040 In twenty thirty, open ai will make two hundred billion 0:06:09.040 --> 0:06:12.000 dollars and somehow have positive free cash flow of thirty 0:06:12.040 --> 0:06:14.920 eight billion dollars. It's just that easy. I teach you 0:06:14.960 --> 0:06:17.479 this in business school. It's just that easy. 0:06:17.480 --> 0:06:19.160 How they're gonna make it. They're gonna make it. They'll 0:06:19.200 --> 0:06:19.719 be fine. 0:06:20.000 --> 0:06:22.320 Open ai is current reported burn is one hundred and 0:06:22.360 --> 0:06:24.760 fifteen billion dollars through twenty thirty, which means there's no 0:06:24.800 --> 0:06:27.279 way that these projections that were leaked included three hundred 0:06:27.320 --> 0:06:30.240 billion dollars in compute costs. Even when you factor in 0:06:30.279 --> 0:06:33.120 the revenue, there's no space in the projections to absorb 0:06:33.160 --> 0:06:34.840 the Oracle money, and from what I can tell, by 0:06:34.880 --> 0:06:37.600 twenty twenty nine, open AI will burned up have burned 0:06:37.680 --> 0:06:40.280 upwards of two hundred ninety billion dollars, assuming it suffves 0:06:40.360 --> 0:06:43.039 that long, which I do not believe it will. Don't 0:06:43.040 --> 0:06:46.039 worry though, open ai is about to make some crazy money. 0:06:46.080 --> 0:06:49.440 I say in no way being totally sarcastic. I'm gonna 0:06:49.440 --> 0:06:53.839 now read you some of the projections that CFO Sarah 0:06:53.880 --> 0:06:58.800 Fryer signed off on. This is a professional chief financial officer. Again, 0:06:58.839 --> 0:07:02.359 more numbers, but pay attention, Pay attention. This is I 0:07:02.400 --> 0:07:04.520 want to give you, because these are numbers you just heard, 0:07:04.560 --> 0:07:07.560 but I want to give you some comparison points. So 0:07:07.600 --> 0:07:09.920 in twenty twenty six, open AI will make, according to 0:07:09.960 --> 0:07:12.760 the projections, or these are the projections, thirty billion dollars 0:07:12.800 --> 0:07:15.320 in revenue, or just under thirty the thirty four billion 0:07:15.360 --> 0:07:18.120 dollars that Salesforce made in twenty twenty four. In twenty 0:07:18.120 --> 0:07:20.960 twenty seven, open AI will allegedly make sixty billion dollars 0:07:21.000 --> 0:07:23.240 or two point six billion, more than the fifty seven 0:07:23.280 --> 0:07:25.800 point four billion dollars the Oracle made in its fiscal 0:07:25.840 --> 0:07:28.440 year twenty twenty five, or eight point four billion dollars 0:07:28.480 --> 0:07:31.280 more than Broadcom made in twenty twenty four. In twenty 0:07:31.280 --> 0:07:33.520 twenty eight, this is crazy. Open ai will make one 0:07:33.560 --> 0:07:36.920 hundred billion dollars two point three billion dollars more than 0:07:36.960 --> 0:07:40.000 Tesla made in twenty twenty four, or eleven point eleven 0:07:40.040 --> 0:07:42.800 point sixty six billion dollars more than TSMC, the single 0:07:42.880 --> 0:07:44.880 largest chip manufacturer in the world. 0:07:44.720 --> 0:07:45.920 Made in twenty twenty four. 0:07:46.480 --> 0:07:49.640 In twenty twenty nine, things get crazier because open ai 0:07:49.720 --> 0:07:51.720 will then make one hundred and fifty four billion dollars 0:07:51.800 --> 0:07:53.760 or fourteen point five billion dollars more than the one 0:07:53.840 --> 0:07:55.960 hundred and thirty point five billion dollars that in Video 0:07:56.000 --> 0:07:59.000 made and its fiscal year twenty twenty five. But in 0:07:59.040 --> 0:08:01.440 twenty thirty, this is when they really blow the doors off. 0:08:01.480 --> 0:08:04.120 They will make two hundred billion dollars or thirty five 0:08:04.160 --> 0:08:06.120 point five billion dollars in the one hundred and sixty 0:08:06.160 --> 0:08:09.280 four point five billion that Meta made in twenty twenty four. 0:08:09.720 --> 0:08:12.280 Just so we are clear, open ai intends to ten 0:08:12.320 --> 0:08:14.840 exits revenue in the space of four years, selling software 0:08:14.880 --> 0:08:17.920 and access to its models in an industry with about 0:08:17.960 --> 0:08:21.120 sixty billion dollars of revenue. In twenty twenty five, How 0:08:21.200 --> 0:08:24.240 will it do this? Open ai does not say. I 0:08:24.400 --> 0:08:27.160 don't know open ai CFO Sarah Fryer, but I do 0:08:27.240 --> 0:08:29.200 know that signing off on these numbers is at the 0:08:29.320 --> 0:08:33.599 very least ethically questionable. But putting aside the ridiculousness of 0:08:33.640 --> 0:08:37.000 open AI's deals or its funding requirements, Fryar has wilfully 0:08:37.080 --> 0:08:39.839 allowed Samulman and open ai to state goals that defy 0:08:39.920 --> 0:08:43.239 reality or good sense, all to take advantage of investors 0:08:43.360 --> 0:08:46.800 and public markets that have completely lost the plot. I'm 0:08:46.840 --> 0:08:49.760 actually you know what, I'm actually going to be blunter. 0:08:50.200 --> 0:08:53.040 Open ai has signed multiple different deals and contracts for 0:08:53.080 --> 0:08:55.800 amounts it cannot afford to pay that it cannot hope 0:08:55.800 --> 0:08:58.240 to raise the money to pay for that defy the 0:08:58.280 --> 0:09:01.840 amounts of venture capital and private capital available, or to 0:09:01.880 --> 0:09:05.640 sustain a company that will burn half a trillion dollars 0:09:05.679 --> 0:09:08.800 in the next four years based on their own expenditures, 0:09:08.880 --> 0:09:11.360 and they have no path of the profitability of any kind. 0:09:12.240 --> 0:09:14.400 But what about that Invidia deal? And one hundred billion 0:09:14.440 --> 0:09:16.280 dollars is a lot of money, right, it is? But 0:09:16.320 --> 0:09:20.040 the announcement is bullshit. I know you may have read otherwise, 0:09:20.040 --> 0:09:22.680 but in Vidia didn't give open Ai a hundred billion dollars. 0:09:22.720 --> 0:09:25.080 In fact, ninety billion dollars of that is contingent on 0:09:25.160 --> 0:09:28.000 open Ai building roughly one hundred and twenty five billion 0:09:28.040 --> 0:09:30.520 dollars worth of data centers with two hundred billion dollars 0:09:30.520 --> 0:09:33.439 worth of GPUs inside them. And those data centers will 0:09:33.440 --> 0:09:35.520 have to be built faster than anyone could possibly build 0:09:35.520 --> 0:09:37.680 something of that scale. And there's no evidence that open 0:09:37.720 --> 0:09:40.280 ai knows where or when these facilities will. 0:09:40.080 --> 0:09:41.760 Be built or who will build them. 0:09:42.160 --> 0:09:45.400 Important detail as well, in Vidia's data centers have nothing 0:09:45.440 --> 0:09:47.680 to do with stargate. So when you read that open 0:09:47.720 --> 0:09:50.920 ai has seven gigawatts of stargate whatever, it doesn't exist. 0:09:50.960 --> 0:09:54.679 Who cares? That's nothing to do with Nvidia. Oracle is 0:09:54.720 --> 0:09:57.120 not building these data centers. The ten gigawatts that they 0:09:57.160 --> 0:10:10.800 need to build with video are something completely separate. But 0:10:10.840 --> 0:10:13.079 you know what, let's take a step back with something 0:10:13.080 --> 0:10:17.640 really simple. Data centers take forever to build. As I've 0:10:17.640 --> 0:10:20.560 said previously, based on current reports, it's taking Oracle and 0:10:20.600 --> 0:10:22.000 Cruiser around two point five. 0:10:21.920 --> 0:10:22.559 Years per giga. 0:10:22.559 --> 0:10:25.680 What of data circumpans city and nowhere in these reports 0:10:25.679 --> 0:10:28.040 there's one reporter take a fucking second to say, hey, 0:10:28.080 --> 0:10:29.520 what data centers are you talking about? 0:10:29.559 --> 0:10:30.640 Hey, didn't Sam alm. 0:10:30.440 --> 0:10:32.280 And say back in July he was building ten gigawats 0:10:32.280 --> 0:10:35.320 of data center capacity with Oracle. But wait now Oracle 0:10:35.360 --> 0:10:37.600 and open Ai have done another announcement that says they're 0:10:37.679 --> 0:10:40.240 only doing seven gigawats, but they already had a schedule 0:10:40.240 --> 0:10:44.640 on ten gigawatts. Like I said, totally different ten gigawats 0:10:44.640 --> 0:10:47.400 to the one within video. But you know what, got 0:10:47.440 --> 0:10:49.800 a few hundred billion dollars of data centers. Now you're 0:10:49.800 --> 0:10:52.880 breaking real money anyway. I cannot be clear enough how 0:10:52.960 --> 0:10:54.880 unlikely it is that the first giga what of in 0:10:54.960 --> 0:10:57.120 video systems will be deployed in the second half of 0:10:57.200 --> 0:10:59.760 twenty twenty six, as in Video has stated with their 0:10:59.760 --> 0:11:02.360 deal with open Ai. And that's as if the land 0:11:02.440 --> 0:11:05.120 has been bought and got permits and the construction has started. 0:11:05.360 --> 0:11:06.400 None of this has happened. 0:11:06.760 --> 0:11:09.840 But you know, let's get really specific on costs. 0:11:10.440 --> 0:11:13.080 Kruso's one point two gigawatts of compute for open Ai 0:11:13.240 --> 0:11:15.599 is a fifteen billion dollar joint venture, which means a 0:11:15.640 --> 0:11:18.800 gigawat of compute runs about twelve point five billion dollars 0:11:18.800 --> 0:11:21.960 worth of construction, Abilene's eight buildings are meant to hold 0:11:21.960 --> 0:11:24.959 fifty thousand in video GB two hundred GPUs and their 0:11:25.000 --> 0:11:28.240 associated networking infrastructure. So let's say a Gigawa is around 0:11:28.280 --> 0:11:31.040 three hundred and thirty three thousand Blackwell GPUs, even though 0:11:31.040 --> 0:11:33.360 the GB two hundreds are technically two of them put together. 0:11:33.840 --> 0:11:35.599 Though this math is a little funky due to in 0:11:35.720 --> 0:11:38.160 video promising to install its new Reuben GPUs in these 0:11:38.240 --> 0:11:41.120 theoretical data centers. That means these data centers will require 0:11:41.120 --> 0:11:42.640 a little under two hundred billion. 0:11:42.440 --> 0:11:43.559 Dollars worth of GPUs. 0:11:43.880 --> 0:11:45.920 By my maths, that's about three hundred and twenty five 0:11:45.960 --> 0:11:47.880 billion dollars, but in videos saying it could be half 0:11:47.920 --> 0:11:50.040 a trillion. But you know, who gives a shit, right, 0:11:50.600 --> 0:11:55.160 numbers don't matter. You've got Jensen Wong telling CNBC that 0:11:55.320 --> 0:11:57.840 one giga what costs fifty sixty billion dollars. You've got 0:11:57.840 --> 0:12:00.760 Altman claiming the same thing to Bloomberg. But my mass 0:12:00.880 --> 0:12:04.640 is something completely fucking different, But no one seems. 0:12:04.360 --> 0:12:05.319 To want to bother. 0:12:06.360 --> 0:12:09.679 You know. It's only like the places I'm quoting are 0:12:09.840 --> 0:12:12.520 leading business outlets with teams of thousands of people, with 0:12:12.559 --> 0:12:15.680 incredible infrastructure where they could sit down and say, well, 0:12:15.920 --> 0:12:19.040 you know it costs this much to build stargate ABERI, 0:12:19.080 --> 0:12:21.920 and why don't we work out actually what this costs? No, no, no, 0:12:22.080 --> 0:12:25.000 just eat the info slop yum yum yum. You know 0:12:25.320 --> 0:12:28.120 it doesn't exhaust me. It doesn't make me cynical and 0:12:28.200 --> 0:12:31.640 frustrated with people who could do a good job. Indeed, 0:12:31.640 --> 0:12:34.000 I've seen them do a good job, but they just 0:12:34.040 --> 0:12:35.920 seem to not want to do one with this, and 0:12:36.320 --> 0:12:39.920 it genuinely bothers me on a day to day basis. Okay, 0:12:40.120 --> 0:12:43.280 I realize I'm just ranting at this point, but I'm 0:12:43.320 --> 0:12:45.280 gonna be honest. I'm just I'm so tired of this. 0:12:47.800 --> 0:12:50.400 As a reminder as well, Open Eyes agreed to spend 0:12:50.400 --> 0:12:52.600 three hundred billion dollars on compute with Oracle. But at 0:12:52.600 --> 0:12:56.120 this point, as you can tell, numbers basically don't have meaning. 0:12:56.760 --> 0:12:59.800 Numbers have stopped meaning anything. If you believe these numbers, 0:13:00.280 --> 0:13:03.440 you're a fantasist or a liar, or you just don't 0:13:03.440 --> 0:13:05.120 want to think about them too much, because when you 0:13:05.160 --> 0:13:08.200 think about them, you start to realize how unrealistic it is. 0:13:08.559 --> 0:13:10.960 And even if you remain steadfast in your belief of 0:13:11.040 --> 0:13:14.800 the transformative potential of large language models. Eventually, there comes 0:13:14.800 --> 0:13:17.040 a point where reality must prevail and you accept that 0:13:17.080 --> 0:13:18.680 we're in a bubble, whether you like it or not. 0:13:20.080 --> 0:13:22.480 What we're witnessing is one of the most egregious wastes 0:13:22.520 --> 0:13:25.520 of capital in history, sold by career charlatans, with their 0:13:25.520 --> 0:13:28.320 reputations laundered by a tech and business media afraid to 0:13:28.320 --> 0:13:30.959 criticize the powerful an analyst that don't seem to want 0:13:31.000 --> 0:13:34.079 to tell their investors the truth. There are no historic 0:13:34.160 --> 0:13:38.240 comparisons here. Even Britain's abominable eighteen hundreds railway bubble, which 0:13:38.280 --> 0:13:42.040 absorbed half of the country's national income, created valuable infrastructure 0:13:42.040 --> 0:13:44.080 for trains, a vehicle that can move people to and 0:13:44.120 --> 0:13:47.920 from places. So the train doesn't randomly just go backwards 0:13:47.920 --> 0:13:49.800 when you make it go forwards. It doesn't become a 0:13:49.840 --> 0:13:53.440 bus or a dog at random. It doesn't hallucinate doors 0:13:53.440 --> 0:13:57.000 that open and close at random. Trains work and GPUs 0:13:57.040 --> 0:13:59.560 are not trains, nor are they cars or even CPUs, 0:14:00.080 --> 0:14:02.240 not adaptable to many other kinds of work, nor are 0:14:02.240 --> 0:14:04.760 they the infrastructure of the future of tech, because they're 0:14:04.800 --> 0:14:07.160 already quite old and with everybody focused on buying them, 0:14:07.200 --> 0:14:09.600 you'd absolutely see one other use case by now that 0:14:09.679 --> 0:14:14.800 actually mattered. GPUs are expensive, power hungry, environmentally destructive, and 0:14:14.840 --> 0:14:17.800 require their own kinds of cooling and server infrastructure, making 0:14:17.800 --> 0:14:22.120 every GPU data center an environmental and fiscal bubble onto themselves. 0:14:22.520 --> 0:14:25.360 And whereas the Victorian train and infrastructure still exists in 0:14:25.400 --> 0:14:27.520 the UK, though it has been upgraded over the years, 0:14:27.680 --> 0:14:30.760 a GPU has a limited useful lifespan. These are cards 0:14:30.800 --> 0:14:34.120 that can and will break after a period of extended usage, 0:14:34.200 --> 0:14:36.280 whether that period is five years or later, though I 0:14:36.320 --> 0:14:39.800 hear it's one to three, and they'll inevitably be superseded 0:14:39.840 --> 0:14:42.200 by something better and more powerful, meaning that the resale 0:14:42.240 --> 0:14:44.280 value of that GPU will only go down with a 0:14:44.280 --> 0:14:47.400 price depreciation. That's the kin to a new car, and 0:14:47.440 --> 0:14:49.280 then a used car, and then a car that you 0:14:49.360 --> 0:14:53.160 kick the door over every so often. I am telling you, 0:14:53.440 --> 0:14:55.600 as I have been telling you for years, again and 0:14:55.640 --> 0:14:58.000 again and again, that the demand is not there for 0:14:58.040 --> 0:15:01.280 generative AI, and the demand is never ever arriving. The 0:15:01.320 --> 0:15:03.920 only reason anyone humors any of this crap is the 0:15:04.000 --> 0:15:07.400 endless hoarding of GPUs to build capacity for a revolution. 0:15:07.040 --> 0:15:08.240 That will never arrive. 0:15:08.880 --> 0:15:11.200 Well, that and open Ai, a company that's built and 0:15:11.240 --> 0:15:15.520 sold on lies about chat GPT's capabilities. Chat GPT's popularity 0:15:15.520 --> 0:15:18.680 in open hunger for endless amounts of compute have created 0:15:18.720 --> 0:15:20.600 the illusion of demand due to the sheer amount of 0:15:20.640 --> 0:15:23.600 capacity required to keep their services operational. Also that they 0:15:23.640 --> 0:15:25.760 can burn around eight billion dollars or more in twenty 0:15:25.800 --> 0:15:28.600 twenty five and hundreds of billions of dollars more by 0:15:28.960 --> 0:15:32.400 twenty thirty if they make him, which they won't. The 0:15:32.560 --> 0:15:35.120 Nvidia dealers are fast an obvious attempt by the largest 0:15:35.120 --> 0:15:37.280 company on the American stock market to prop up the 0:15:37.320 --> 0:15:40.760 one significant revenue generator in the entire industry, knowing that 0:15:40.840 --> 0:15:42.800 time is running out for it to create new avenues 0:15:42.840 --> 0:15:46.320 for eternal growth. I'd argue that Nvidia's deal also shows 0:15:46.320 --> 0:15:48.880 the complete content that these companies have for the media. 0:15:49.040 --> 0:15:51.520 There are no details about how this deal works beyond 0:15:51.520 --> 0:15:54.200 the initial ten billion dollars. There's no land purchase, no 0:15:54.280 --> 0:15:57.440 data center construction started, yet the media slurps it down 0:15:57.480 --> 0:16:01.200 without a second thought. I am but one man, and 0:16:01.240 --> 0:16:05.160 I'm fucking peculiar. I did not learn financial analysis in school, 0:16:05.160 --> 0:16:06.760 but I appear to be one of the few people 0:16:06.800 --> 0:16:09.480 doing even the most basic analysis of these deals. And 0:16:09.480 --> 0:16:11.440 while I'm having a great time doing so, I'm are 0:16:11.520 --> 0:16:14.320 so extremely frustrated about how little effort is being put 0:16:14.320 --> 0:16:16.200 into praying apart these deals by other people. 0:16:17.040 --> 0:16:19.720 I realize how ridiculous all that this sounds. I get it. 0:16:19.760 --> 0:16:22.120 There's so much money being promised, that so many people 0:16:22.160 --> 0:16:24.600 market rallies built off the back of these massive deals, 0:16:24.640 --> 0:16:26.680 and I get that the assumption that this is that 0:16:26.760 --> 0:16:29.160 this much money can't be wrong, that this many people 0:16:29.160 --> 0:16:31.600 wouldn't just say stuff without intending to follow through, or 0:16:31.600 --> 0:16:34.240 without considering whether their company could afford it. I know 0:16:34.360 --> 0:16:36.560 it's hard to conceive that these hundreds of billions of 0:16:36.600 --> 0:16:39.160 dollars could be invested in something for no apparent reason, 0:16:39.520 --> 0:16:42.480 but it's happening, right goddamn now, in front of your eyes. 0:16:42.640 --> 0:16:44.880 And I am going to be merciless on anyone who 0:16:44.880 --> 0:16:47.800 attempts to write, how could we possibly have seen this coming? 0:16:48.880 --> 0:16:52.360 Generator AI has never been reliable, has always been unprofitable, 0:16:52.400 --> 0:16:55.200 and has always been unsustainable. And I've been saying so 0:16:55.240 --> 0:16:58.520 since at least February twenty twenty four. These economics have 0:16:58.600 --> 0:17:01.600 never made sense. Something I've said repeatedly since April twenty 0:17:01.600 --> 0:17:03.880 twenty four, and when I wrote how does open Ai 0:17:03.960 --> 0:17:06.320 Survive in July twenty twenty four, I had multiple people 0:17:06.359 --> 0:17:10.280 suggest I was being an alarmist. Here's some alarmism for you. 0:17:10.760 --> 0:17:12.719 The longer it takes for open ai to die, the 0:17:12.720 --> 0:17:15.600 more damage it will cause once it does. As I 0:17:15.680 --> 0:17:18.120 talked about earlier in this series, venture capital could run 0:17:18.119 --> 0:17:20.480 out in six quarters, with investor and research of John 0:17:20.520 --> 0:17:23.240 Socoda estimating that there will be only around one hundred 0:17:23.240 --> 0:17:25.960 and sixty four billion dollars of dry powder, which is 0:17:25.960 --> 0:17:28.320 available capital, by the way, in USVC firms, by the 0:17:28.400 --> 0:17:31.880 end of twenty twenty five. In July, the French Tech 0:17:31.960 --> 0:17:35.040 journal reported using Pitchbook DAYA, the global venture capital deal 0:17:35.040 --> 0:17:38.320 activity reached its lowest first half total since twenty eighteen, 0:17:38.520 --> 0:17:40.879 with one hundred and thirty nine point four billion dollars 0:17:40.880 --> 0:17:42.960 in deal value in the first half of twenty twenty five, 0:17:43.200 --> 0:17:45.240 down from one hundred and eighty three point four billion 0:17:45.240 --> 0:17:47.439 in the first half of twenty twenty four, meaning that 0:17:47.480 --> 0:17:50.040 any further expansion or demands for venge capital from open 0:17:50.119 --> 0:17:52.120 ai will likely sap dwindling. 0:17:51.720 --> 0:17:53.600 Funds available for other startups. 0:17:54.240 --> 0:17:56.000 Things get worse when you narrow things to the US 0:17:56.119 --> 0:17:59.080 venture capital In a piece from April Erstin Young report, 0:17:59.160 --> 0:18:01.760 the VC backed in in US companies say eighty billion 0:18:01.760 --> 0:18:04.399 dollars in Q one twenty twenty five, but one forty 0:18:04.400 --> 0:18:06.600 billion dollar deal account for half of the investment. 0:18:06.680 --> 0:18:08.280 Open Ai is forty billion. 0:18:08.000 --> 0:18:11.440 Dollar deal, of which only ten billion dollar has actually closed, 0:18:11.520 --> 0:18:15.199 and that didn't happen until fucking June. Without the imaginary 0:18:15.240 --> 0:18:17.719 money from open Ai, US venture capital would have declined 0:18:17.720 --> 0:18:20.560 by thirty six percent. I should also add another eight 0:18:20.640 --> 0:18:22.600 point three billion dollars was added on the top of that, 0:18:22.680 --> 0:18:26.280 but still that's less than forty dollars, folks. The longer 0:18:26.280 --> 0:18:28.320 the open Ai survives, the longer it will sap the 0:18:28.359 --> 0:18:30.880 remaining billions from the tech ecosystem, and I expect its 0:18:30.920 --> 0:18:34.200 tendrils to extend to private credit soon too. The three 0:18:34.320 --> 0:18:36.359