1 00:00:02,480 --> 00:00:07,040 Speaker 1: Zone Media, Hello im ed Zetron, and this, of course 2 00:00:07,160 --> 00:00:21,000 Speaker 1: is better offline what Welcome to the second part of 3 00:00:21,000 --> 00:00:24,120 Speaker 1: our four part series, where I give you my most comprehensive, 4 00:00:24,320 --> 00:00:26,599 Speaker 1: most up to day explanation of why we're in a 5 00:00:26,600 --> 00:00:29,640 Speaker 1: bubble and what that even means. The reason why I'm 6 00:00:29,680 --> 00:00:32,280 Speaker 1: taking my time to be descriptive and comprehensive is because 7 00:00:32,320 --> 00:00:34,280 Speaker 1: I want this to make sense to those who listen 8 00:00:34,320 --> 00:00:37,159 Speaker 1: to it. Having written hundreds of thousands of words this 9 00:00:37,280 --> 00:00:39,879 Speaker 1: year about the AI bubble, so many of the arguments 10 00:00:39,880 --> 00:00:42,320 Speaker 1: I've made and the secrets I've exposed are contained in 11 00:00:42,360 --> 00:00:45,320 Speaker 1: their own discreet little episodes or newsletters. This is my 12 00:00:45,440 --> 00:00:48,319 Speaker 1: series to consolidate all of the information I've put out 13 00:00:48,320 --> 00:00:51,040 Speaker 1: there in one place, and I want to make it 14 00:00:51,080 --> 00:00:53,760 Speaker 1: makes sense to anyone who listens to him. I want anyone, 15 00:00:53,840 --> 00:00:56,440 Speaker 1: even who someone who doesn't even know that much about AI, 16 00:00:56,520 --> 00:00:58,360 Speaker 1: to listen to the audience I've been making for the 17 00:00:58,400 --> 00:01:01,160 Speaker 1: past three years, to understand why things are dire and 18 00:01:01,200 --> 00:01:03,360 Speaker 1: to feel the same alarm I'm feeling, or at least 19 00:01:03,440 --> 00:01:05,800 Speaker 1: understand why I'm alarmed. Because I don't like to tell 20 00:01:05,800 --> 00:01:08,120 Speaker 1: you how you feel. Old school bit of FEEDBACKERD got 21 00:01:08,160 --> 00:01:10,560 Speaker 1: from a listener once, and I appreciate that to this day. 22 00:01:11,840 --> 00:01:14,560 Speaker 1: Now today I'll make the case the generative AI's fundamental 23 00:01:14,600 --> 00:01:17,000 Speaker 1: growth story is flawed and explain why we're in the 24 00:01:17,040 --> 00:01:20,320 Speaker 1: midst of an egregious bubble. This industry is sold by 25 00:01:20,400 --> 00:01:22,679 Speaker 1: keeping things vague and knowing that most people don't dig 26 00:01:22,800 --> 00:01:25,679 Speaker 1: much deeper than a headliner problem I simply do not have. 27 00:01:26,319 --> 00:01:29,600 Speaker 1: This industry is effectively in service of two companies, open 28 00:01:29,640 --> 00:01:32,800 Speaker 1: Ai and Nvidia, who pump headlines out through endless contracts 29 00:01:32,840 --> 00:01:35,760 Speaker 1: between them or subsidiaries or investments to give the illusion 30 00:01:35,760 --> 00:01:39,040 Speaker 1: of activity. Open Ai has now promised over four hundred 31 00:01:39,080 --> 00:01:42,000 Speaker 1: billion dollars in the next four years, though honestly they 32 00:01:42,080 --> 00:01:44,119 Speaker 1: might owe about a trillion dollars with all the data 33 00:01:44,120 --> 00:01:46,959 Speaker 1: centers they're signed up for. All of these are egregious 34 00:01:47,000 --> 00:01:50,040 Speaker 1: sums for a company that have already forecasted billions in 35 00:01:50,120 --> 00:01:52,720 Speaker 1: losses with no clear explanation as to how it will 36 00:01:52,720 --> 00:01:54,920 Speaker 1: afford any of this beyond we need more money and 37 00:01:54,960 --> 00:01:57,400 Speaker 1: they're vague hope that there's another soft bank or Microsoft 38 00:01:57,400 --> 00:02:00,160 Speaker 1: waiting in the wings to swoop in and save the day. Now, 39 00:02:00,160 --> 00:02:01,880 Speaker 1: I'm going to walk you through where I see this 40 00:02:01,960 --> 00:02:04,240 Speaker 1: industry today and why I see no future for it 41 00:02:04,360 --> 00:02:09,880 Speaker 1: beyond a horrible, fiery car rack. While everybody reasonably harps 42 00:02:09,880 --> 00:02:12,840 Speaker 1: on about hallucinations, which to remind you, is when a 43 00:02:12,880 --> 00:02:16,560 Speaker 1: model authoritativety states something that isn't true, the truth of 44 00:02:16,960 --> 00:02:19,160 Speaker 1: why that's bad is far more complex and actually far 45 00:02:19,200 --> 00:02:22,119 Speaker 1: worse than it seems. You cannot rely on a large 46 00:02:22,200 --> 00:02:24,320 Speaker 1: language model to do what you want, even though its 47 00:02:24,440 --> 00:02:27,320 Speaker 1: highly tuned models on the most expensive and intricate platforms 48 00:02:27,320 --> 00:02:30,560 Speaker 1: can't actually be relied upon to do exactly what you want. 49 00:02:30,880 --> 00:02:33,080 Speaker 1: And I know some people might say, well, yes, they 50 00:02:33,080 --> 00:02:36,720 Speaker 1: do every time, one hundred percent of the time. A 51 00:02:36,760 --> 00:02:40,040 Speaker 1: hallucination isn't just when these models say something that isn't true. 52 00:02:40,280 --> 00:02:42,520 Speaker 1: It's when they decide to do something wrong because it 53 00:02:42,560 --> 00:02:44,520 Speaker 1: seems the most likely thing to do, or when a 54 00:02:44,560 --> 00:02:46,800 Speaker 1: coding model decides to go on a wild goose chase, 55 00:02:47,200 --> 00:02:49,160 Speaker 1: failing the user and burning a ton of money in 56 00:02:49,200 --> 00:02:53,000 Speaker 1: the process. The advent of reasoning models those engineered to 57 00:02:53,240 --> 00:02:55,480 Speaker 1: think through problems in the way reminiscent of a human. 58 00:02:55,520 --> 00:02:58,440 Speaker 1: But it's not thinking. They don't think. They have no consciousness. 59 00:02:58,680 --> 00:03:01,480 Speaker 1: They literally you ask them something and they break down 60 00:03:01,520 --> 00:03:04,560 Speaker 1: what the prompt might mean and then choose it's not thinking, 61 00:03:04,919 --> 00:03:07,120 Speaker 1: and the expansion of what people are trying to use 62 00:03:07,240 --> 00:03:10,320 Speaker 1: llms for demands that the definition of an AI hallucination 63 00:03:10,480 --> 00:03:13,720 Speaker 1: be widened, not merely referring to factual errors, but fundamental 64 00:03:13,800 --> 00:03:16,600 Speaker 1: errors in understanding the user's request or intent or what 65 00:03:16,680 --> 00:03:20,080 Speaker 1: constitutes a task, in part because these models, as I said, 66 00:03:20,120 --> 00:03:23,360 Speaker 1: cannot think and do not know anything. However successful a 67 00:03:23,400 --> 00:03:25,760 Speaker 1: model might be in generating something good once, it will 68 00:03:25,800 --> 00:03:29,120 Speaker 1: also often generate something bad, or it will generate the 69 00:03:29,200 --> 00:03:32,120 Speaker 1: right thing, but in an inefficient and over verbos fashion, 70 00:03:32,520 --> 00:03:34,640 Speaker 1: you do not know what you're going to get each time, 71 00:03:34,720 --> 00:03:37,680 Speaker 1: and hallucinations multiply with the complexity of the thing you're 72 00:03:37,720 --> 00:03:40,680 Speaker 1: asking for or whether a task contains multiple steps, which 73 00:03:40,720 --> 00:03:43,520 Speaker 1: is a fatal blow to the idea of agents. You 74 00:03:43,560 --> 00:03:46,080 Speaker 1: can add as many levels of intrigue and reasoning as 75 00:03:46,080 --> 00:03:48,600 Speaker 1: you want, but large language models cannot be trusted to 76 00:03:48,600 --> 00:03:51,840 Speaker 1: do something correctly or even consistently, let alone every time. 77 00:03:53,040 --> 00:03:56,400 Speaker 1: Model companies have successfully convinced everybody that the issue is 78 00:03:56,400 --> 00:03:58,880 Speaker 1: that users are prompting the models wrong and that the 79 00:03:58,920 --> 00:04:01,000 Speaker 1: people need to be trained to use AI. But what 80 00:04:01,040 --> 00:04:04,040 Speaker 1: they're doing is training people to explain away the inconsistencies 81 00:04:04,080 --> 00:04:07,600 Speaker 1: of large language models and to assume individual responsibility for 82 00:04:07,640 --> 00:04:10,720 Speaker 1: what is an innate floor in how these fucking things work. 83 00:04:11,720 --> 00:04:16,040 Speaker 1: Large language models are also uniquely expensive. Many mistakenly try 84 00:04:16,080 --> 00:04:18,080 Speaker 1: and claim that this is like the dot com boom 85 00:04:18,160 --> 00:04:21,160 Speaker 1: or uber, but the basic unique economics of generative AI 86 00:04:21,360 --> 00:04:25,600 Speaker 1: are insane. Providers must purchase tens or hundreds of thousands 87 00:04:25,600 --> 00:04:29,360 Speaker 1: of GPUs, each costing fifty thousand to seventy thousand apiece, 88 00:04:29,720 --> 00:04:32,279 Speaker 1: and the hundreds of millions or billions of dollars of 89 00:04:32,320 --> 00:04:34,960 Speaker 1: infrastructure that goes around them are so expensive and hard 90 00:04:34,960 --> 00:04:38,599 Speaker 1: to install, and that's without mentioning things like staffing or construction, 91 00:04:38,720 --> 00:04:42,360 Speaker 1: or power or water or even permitting. Then you turn 92 00:04:42,400 --> 00:04:45,359 Speaker 1: them on and immediately they start losing new money. Despite 93 00:04:45,440 --> 00:04:48,320 Speaker 1: hundreds of billions of GPUs sold, nobody seems to actually 94 00:04:48,360 --> 00:04:50,560 Speaker 1: make any of it, other than Video of course, the 95 00:04:50,600 --> 00:04:53,120 Speaker 1: company that makes them, and resellers like Dell and Supermicro 96 00:04:53,240 --> 00:04:55,360 Speaker 1: who buy the GPUs put them in servers and sell 97 00:04:55,360 --> 00:04:58,200 Speaker 1: them to other people. Now, if you're an eager listener, 98 00:04:58,800 --> 00:05:01,320 Speaker 1: I would love to hear from you on one question, 99 00:05:01,400 --> 00:05:03,600 Speaker 1: then this is just something that's been bouncing around my head. 100 00:05:03,760 --> 00:05:07,760 Speaker 1: Super Micro is in Video a customer of super Micro. 101 00:05:07,800 --> 00:05:10,640 Speaker 1: They're a custom in super Micro is a huge customer 102 00:05:10,640 --> 00:05:12,440 Speaker 1: of Video. I've read something like seventy percent of the 103 00:05:12,560 --> 00:05:16,920 Speaker 1: cost of goods sold is buying GPUs. But I read 104 00:05:16,960 --> 00:05:18,279 Speaker 1: that in Video as a customer of them. But I 105 00:05:18,279 --> 00:05:21,360 Speaker 1: can't find anything else. Reach out easy at better Offline 106 00:05:21,400 --> 00:05:24,039 Speaker 1: dot com if you've got any thoughts there anyway, But 107 00:05:24,160 --> 00:05:26,560 Speaker 1: back to those resourers. This arrangement works out great for 108 00:05:26,640 --> 00:05:29,000 Speaker 1: Jensen Wang, the CEO of in Video, and terribly for 109 00:05:29,040 --> 00:05:32,600 Speaker 1: everybody else. Today, I'm gonna explain the insanity of the 110 00:05:32,640 --> 00:05:35,080 Speaker 1: situation we'll find ourselves in and why I continue to 111 00:05:35,120 --> 00:05:38,640 Speaker 1: do this work underterred. The bubble has entered its most pornographic, 112 00:05:38,680 --> 00:05:41,440 Speaker 1: aggressive and destructive stage, where the more obvious it becomes 113 00:05:41,440 --> 00:05:43,960 Speaker 1: that we're all cooked here in AI land, the more 114 00:05:44,080 --> 00:05:47,839 Speaker 1: ridiculous the GENERATIVAI industry will act a dark juxtaposition against 115 00:05:47,880 --> 00:05:50,480 Speaker 1: every new study that says generative AI does not work, 116 00:05:50,720 --> 00:05:54,039 Speaker 1: or new story about chat gpd's uncannyability to activate mental 117 00:05:54,080 --> 00:05:57,560 Speaker 1: illness in people. And we're going to start looking at 118 00:05:57,600 --> 00:06:00,520 Speaker 1: one company in Video which now dominates the stock market 119 00:06:00,600 --> 00:06:03,760 Speaker 1: and has taken extraordinary and dangerous measures to sustain growth. 120 00:06:03,760 --> 00:06:07,680 Speaker 1: That is, to any sane person, completely unsustainable and unrealistic 121 00:06:07,720 --> 00:06:11,279 Speaker 1: on every level. But let's start simple. Nvidia is a 122 00:06:11,320 --> 00:06:14,800 Speaker 1: hardware company that sells GPUs, including consumer GPUs that you'd 123 00:06:14,839 --> 00:06:17,599 Speaker 1: see in a modern gaming PC. But when you read 124 00:06:17,680 --> 00:06:20,520 Speaker 1: someone say GPU within the context of AI, they mean 125 00:06:20,760 --> 00:06:24,640 Speaker 1: enterprise focused GPUs like the A one hundred, h one hundred, 126 00:06:24,920 --> 00:06:27,719 Speaker 1: H two hundred, and more modern GPUs like the Blackwell 127 00:06:27,839 --> 00:06:31,560 Speaker 1: Series B two hundred and GB two hundred, which combines 128 00:06:31,560 --> 00:06:35,680 Speaker 1: two GPUs with an Nvidia CPU. This is all complex sounding, 129 00:06:35,720 --> 00:06:38,320 Speaker 1: but I want you to have the groundwork. These GPUs 130 00:06:38,360 --> 00:06:41,120 Speaker 1: cost anywhere from fifty to seventy thousand dollars and require 131 00:06:41,160 --> 00:06:44,640 Speaker 1: tens of thousands of dollars more of infrastructure networking to 132 00:06:45,080 --> 00:06:48,000 Speaker 1: cluster these server acts of GPUs together to provide compute 133 00:06:48,040 --> 00:06:50,360 Speaker 1: and massive cooling systems to deal with the massive amounts 134 00:06:50,360 --> 00:06:52,960 Speaker 1: of heat they produce, as well as servers themselves that 135 00:06:53,000 --> 00:06:54,920 Speaker 1: they run on, which typically use top of the line 136 00:06:55,000 --> 00:06:58,160 Speaker 1: data center CPUs and contain vast quantities of high speed 137 00:06:58,240 --> 00:07:02,640 Speaker 1: memory and storage. GPUs itself is likely the most expensive 138 00:07:02,640 --> 00:07:05,240 Speaker 1: single it and within an AI server the other costs. 139 00:07:05,279 --> 00:07:07,520 Speaker 1: And I'm not even factoring in the actual physical building 140 00:07:07,600 --> 00:07:10,200 Speaker 1: that the server lives in, or the water or electricity 141 00:07:10,200 --> 00:07:14,440 Speaker 1: that uses. Well, all this crap adds up. I've mentioned 142 00:07:14,440 --> 00:07:17,160 Speaker 1: in Video because it has a virtual monopoly in this space, 143 00:07:17,720 --> 00:07:21,200 Speaker 1: Genera IVAI effectively requires in video GPUs, in part because 144 00:07:21,240 --> 00:07:23,200 Speaker 1: it's the only company really making the kinds of high 145 00:07:23,200 --> 00:07:26,520 Speaker 1: powered cards that GENERAIVAI demands, and because in Video created 146 00:07:26,560 --> 00:07:30,160 Speaker 1: something called Kuda. Cuda a collection of software tools that 147 00:07:30,240 --> 00:07:33,040 Speaker 1: lets programmers write software that runs on GPUs, which were 148 00:07:33,040 --> 00:07:36,640 Speaker 1: traditionally used primarily for rendering graphics in games. While there 149 00:07:36,680 --> 00:07:39,600 Speaker 1: are some open source alternatives as well as alternatives from 150 00:07:39,600 --> 00:07:42,320 Speaker 1: Intel with its Arc GPUs and AMD, in Vidia's main 151 00:07:42,400 --> 00:07:45,000 Speaker 1: rival in the consumer space, these aren't nearly as mature 152 00:07:45,120 --> 00:07:48,600 Speaker 1: or feature Richkuda's been around for ten fifteen years now, 153 00:07:49,360 --> 00:07:50,800 Speaker 1: they really knew what they were doing. They also have 154 00:07:50,840 --> 00:07:52,880 Speaker 1: bought a company called Melanox, which did the high speed 155 00:07:52,960 --> 00:07:57,840 Speaker 1: networking back in twenty nineteen, I think, for six billion dollars. Anyway, 156 00:07:58,360 --> 00:08:00,680 Speaker 1: due to the complexities of AI models, one cannot just 157 00:08:00,720 --> 00:08:02,880 Speaker 1: stand up a few of these GPUs either you need 158 00:08:02,960 --> 00:08:05,840 Speaker 1: clusters of thousands, tens of thousands, or hundreds of thousands 159 00:08:05,880 --> 00:08:08,520 Speaker 1: of them for it to be worthwhile making any investment 160 00:08:08,560 --> 00:08:11,120 Speaker 1: in GPUs and the hundreds of millions or billions of dollars, 161 00:08:11,320 --> 00:08:14,680 Speaker 1: especially considering they require completely different data center architecture to 162 00:08:14,680 --> 00:08:17,840 Speaker 1: make them run. You've probably read a bunch of stuff 163 00:08:18,080 --> 00:08:22,440 Speaker 1: about cryptominers turning into AI data center providers. These crypto 164 00:08:22,600 --> 00:08:24,840 Speaker 1: data centers have to be knocked down and replaced. They 165 00:08:25,160 --> 00:08:27,440 Speaker 1: you can't just put the same GPUs in isn't going 166 00:08:27,520 --> 00:08:29,920 Speaker 1: to work, and with the new Blackwell ones, the brand 167 00:08:29,960 --> 00:08:32,360 Speaker 1: new ones, and then the Rubens following them, same deal. 168 00:08:33,280 --> 00:08:35,880 Speaker 1: A common request like asking a generative AI model to 169 00:08:35,880 --> 00:08:37,800 Speaker 1: pass through thousands of lines of code and make a 170 00:08:37,880 --> 00:08:40,800 Speaker 1: change or in addition, may use multiples of these fifty 171 00:08:40,800 --> 00:08:43,880 Speaker 1: thousand dollars GPUs at the same time. And so if 172 00:08:43,920 --> 00:08:46,680 Speaker 1: you aspire to serve thousands or millions of concurrent users, 173 00:08:46,720 --> 00:08:50,880 Speaker 1: you need to spend big, really really really big. It's 174 00:08:50,920 --> 00:08:53,439 Speaker 1: these factors, the vendor lock in the ecosystem and the 175 00:08:53,440 --> 00:08:55,680 Speaker 1: fact that generative AI really only works when you're buying 176 00:08:55,760 --> 00:08:59,120 Speaker 1: GPUs at scale that underpin the rise of Nvidia. But 177 00:08:59,200 --> 00:09:02,120 Speaker 1: Beyond the economic and technical factors, there are human ones too. 178 00:09:03,000 --> 00:09:05,680 Speaker 1: To understand the AI bubble is to understand why CEOs 179 00:09:05,720 --> 00:09:08,360 Speaker 1: do the things they do. Because the executive job is 180 00:09:08,440 --> 00:09:10,720 Speaker 1: so vague, they can telegraph the value of their labour 181 00:09:10,760 --> 00:09:15,480 Speaker 1: by spending money on initiatives and partnerships and stratagem. AI 182 00:09:15,520 --> 00:09:18,439 Speaker 1: gave hyperscalers the excuse to spend hundreds of billions of 183 00:09:18,480 --> 00:09:20,920 Speaker 1: dollars on data centers and buy a bunch of GPUs 184 00:09:20,920 --> 00:09:23,000 Speaker 1: to go in them, because that, to the markets, looks 185 00:09:23,040 --> 00:09:26,480 Speaker 1: like they're doing something. By virtue of spending a lot 186 00:09:26,520 --> 00:09:29,840 Speaker 1: of money in a frighteningly short amount of time, Satchnadella 187 00:09:29,920 --> 00:09:33,040 Speaker 1: received multiple glossy profiles, all without having to prove that 188 00:09:33,080 --> 00:09:35,440 Speaker 1: AI can really do anything, be it a job or 189 00:09:35,480 --> 00:09:39,280 Speaker 1: make Microsoft money. Nevertheless, AI allowed CEOs to look busy, 190 00:09:39,280 --> 00:09:41,160 Speaker 1: and once the markets and journalists had agreed on the 191 00:09:41,160 --> 00:09:43,719 Speaker 1: consensus opinion that AI would be big, all that these 192 00:09:43,760 --> 00:09:46,360 Speaker 1: executives had to do was buy GPUs and do AI 193 00:09:46,960 --> 00:09:49,959 Speaker 1: or plug AI within their own software profitts. But really 194 00:09:49,960 --> 00:09:52,120 Speaker 1: it was just jump on the big stupid asshole train. 195 00:10:02,800 --> 00:10:04,760 Speaker 1: We are in the midst of one of the darkest 196 00:10:04,840 --> 00:10:08,840 Speaker 1: forms of software in history, described by many as unwanted guest, 197 00:10:08,960 --> 00:10:12,480 Speaker 1: invading their products, their social media feeds, their bosses, empty minds, 198 00:10:12,520 --> 00:10:15,680 Speaker 1: and resting in the hands of monsters. Every story of 199 00:10:15,720 --> 00:10:18,840 Speaker 1: AI's success feels bereft of any real triumph, with every 200 00:10:18,920 --> 00:10:22,120 Speaker 1: literal description of its abilities involving multiple caveats about the 201 00:10:22,160 --> 00:10:24,840 Speaker 1: mistakes it makes or the incredible costs of running in 202 00:10:25,520 --> 00:10:28,320 Speaker 1: generator of AI really exists for two reasons, to cost 203 00:10:28,400 --> 00:10:31,560 Speaker 1: money and to make executives look busy. It was meant 204 00:10:31,600 --> 00:10:33,640 Speaker 1: to be the new enterprise software and the new I 205 00:10:33,760 --> 00:10:36,160 Speaker 1: phone and the new Netflix all at once, a panacea 206 00:10:36,200 --> 00:10:38,760 Speaker 1: where the software guys pay one hardware guy for GPUs 207 00:10:38,800 --> 00:10:42,320 Speaker 1: to unlock the incredible value creation of the future. In 208 00:10:42,360 --> 00:10:44,360 Speaker 1: many ways, Jenera iv AI was always set up to 209 00:10:44,360 --> 00:10:46,440 Speaker 1: fail because it was meant to be Everything. Was talked 210 00:10:46,440 --> 00:10:49,720 Speaker 1: about like it was everything. It's still sold like it's everything. 211 00:10:49,800 --> 00:10:51,920 Speaker 1: Yet for all the fucking hype, it comes down to 212 00:10:52,120 --> 00:10:56,360 Speaker 1: two companies, Open Ai and Nvidia, and in Video was 213 00:10:56,480 --> 00:10:59,160 Speaker 1: for a while living high on the hog. All CEO 214 00:10:59,240 --> 00:11:01,319 Speaker 1: Jensen Huang had to do every three months would say, 215 00:11:01,400 --> 00:11:03,840 Speaker 1: check out these numbers and the markets and business journalist 216 00:11:03,840 --> 00:11:06,480 Speaker 1: would squeer with glee, even as he said stuff like 217 00:11:06,679 --> 00:11:08,560 Speaker 1: the more you buy, the more you save, in part 218 00:11:08,600 --> 00:11:10,880 Speaker 1: tipping his head to the very real and sensible idea 219 00:11:10,920 --> 00:11:13,720 Speaker 1: of accelerated computing, but frame within the context of the 220 00:11:13,720 --> 00:11:17,440 Speaker 1: cash inferna that's generated AI, and it all seems kind 221 00:11:17,480 --> 00:11:21,480 Speaker 1: of fucking ludicrous. Huang showmanship worked really well for Invidia 222 00:11:21,559 --> 00:11:23,960 Speaker 1: for a while because for a while the growth was easy. 223 00:11:24,240 --> 00:11:28,200 Speaker 1: Everybody was buying GPUs, Meta, Microsoft, Amazon, Google, and to 224 00:11:28,240 --> 00:11:30,840 Speaker 1: a lesser extent, Apple and Tesla made up forty two 225 00:11:30,880 --> 00:11:33,440 Speaker 1: percent of Invidia's revenue, creating at least for the first 226 00:11:33,520 --> 00:11:36,280 Speaker 1: four a degree of shared mania where everybody justified buying 227 00:11:36,280 --> 00:11:38,400 Speaker 1: tens of billions of dollars of GPUs by saying the 228 00:11:38,440 --> 00:11:41,240 Speaker 1: other guy's doing it. This is one of the major 229 00:11:41,280 --> 00:11:44,120 Speaker 1: reasons the AI bubble is happening, because people conflated in 230 00:11:44,200 --> 00:11:47,840 Speaker 1: Vidia's incredible sales with interest in AI rather than everybody 231 00:11:47,840 --> 00:11:51,920 Speaker 1: buying GPUs at once. Don't worry, I'll explain the revenue 232 00:11:51,920 --> 00:11:54,080 Speaker 1: side a little bit later. We're here for the long haul. 233 00:11:54,120 --> 00:11:57,079 Speaker 1: Sit down, get comfort. You're gonna need to be anyway, 234 00:11:57,320 --> 00:12:00,200 Speaker 1: Invidia is now facing a big problem that the only 235 00:12:00,240 --> 00:12:04,079 Speaker 1: thing that grows forever is cancer. On September ninth, twenty 236 00:12:04,120 --> 00:12:06,360 Speaker 1: twenty five, The Wall Street Journal said that in Vidio's 237 00:12:06,400 --> 00:12:09,520 Speaker 1: wow factor was fading, going from beating analyst estimates by 238 00:12:09,520 --> 00:12:12,080 Speaker 1: nearly twenty one percent in its fiscal year Q two 239 00:12:12,200 --> 00:12:15,120 Speaker 1: twenty twenty four earnings to scraping by with a pathetic 240 00:12:15,360 --> 00:12:17,679 Speaker 1: measly one point five to two percent beat in its 241 00:12:17,679 --> 00:12:20,880 Speaker 1: most recent earnings, something that for any other company would 242 00:12:20,880 --> 00:12:23,200 Speaker 1: be a good thing because they made so much money, 243 00:12:23,240 --> 00:12:27,600 Speaker 1: But framed against the delusional expectations that generative AI has inspired, well, 244 00:12:27,640 --> 00:12:30,120 Speaker 1: the figure looks nothing short of ominous. I quote the 245 00:12:30,120 --> 00:12:33,600 Speaker 1: Wall Street Journal. Already, in Vidia's fifty six percent annual 246 00:12:33,679 --> 00:12:36,240 Speaker 1: revenue growth rate in its latest quarter was its slowest 247 00:12:36,240 --> 00:12:39,120 Speaker 1: in more than two years. If analyst projections hold, growth 248 00:12:39,160 --> 00:12:43,280 Speaker 1: will slow further in the current quarter. In any other scenario, 249 00:12:43,360 --> 00:12:45,320 Speaker 1: fifty six percent year of a year growth would lead 250 00:12:45,320 --> 00:12:48,079 Speaker 1: to an abundance of dom perignon and one signing hundreds 251 00:12:48,080 --> 00:12:50,200 Speaker 1: of boobs. But this is in video, and that's just 252 00:12:50,240 --> 00:12:52,959 Speaker 1: not good enough. Back in February twenty twenty four, in 253 00:12:53,040 --> 00:12:55,360 Speaker 1: Video was booking two hundred and sixty five percent year 254 00:12:55,400 --> 00:12:57,800 Speaker 1: of a year growth but in its February twenty twenty 255 00:12:57,840 --> 00:12:59,920 Speaker 1: five earnings in Video only grew by a measly per 256 00:13:00,080 --> 00:13:03,679 Speaker 1: thetic disgusting seventy eight percent year of a year. I'm 257 00:13:03,720 --> 00:13:06,599 Speaker 1: being sarcastic, of course. It isn't so much that in 258 00:13:06,720 --> 00:13:09,520 Speaker 1: Vidia isn't growing, but to grow year over year at 259 00:13:09,559 --> 00:13:12,920 Speaker 1: the rates that people expect is insane. Life was a 260 00:13:12,920 --> 00:13:15,640 Speaker 1: lot easier when in Video went from six point oh 261 00:13:15,679 --> 00:13:18,319 Speaker 1: five billion dollars in revenue in Q four fiscal year 262 00:13:18,360 --> 00:13:21,200 Speaker 1: twenty twenty five to twenty two billion dollars in revenue 263 00:13:21,200 --> 00:13:24,000 Speaker 1: in Q four fiscal year twenty twenty four. But for 264 00:13:24,080 --> 00:13:26,000 Speaker 1: it to grow even fifty five percent year of a 265 00:13:26,080 --> 00:13:28,360 Speaker 1: year from Q two FY twenty twenty six, I'm just 266 00:13:28,400 --> 00:13:30,880 Speaker 1: going to truncate that now, which is forty six point 267 00:13:30,880 --> 00:13:33,960 Speaker 1: seven billion dollars to Q two twenty twenty seven. That 268 00:13:34,000 --> 00:13:36,080 Speaker 1: would require them to make seventy two point three eight 269 00:13:36,160 --> 00:13:38,560 Speaker 1: five billion dollars in revenue in the space of three months, 270 00:13:38,559 --> 00:13:41,000 Speaker 1: mostly from selling GPUs, which make up about eighty eight 271 00:13:41,000 --> 00:13:43,400 Speaker 1: percent of its revenue. Just want to be clear that 272 00:13:44,360 --> 00:13:47,080 Speaker 1: in a year they would have to make seventy two 273 00:13:47,160 --> 00:13:51,080 Speaker 1: billion dollars just selling pretty much GPUs and the associated 274 00:13:51,080 --> 00:13:55,320 Speaker 1: hardware in the space of three months. It's it's insane. 275 00:13:55,360 --> 00:13:57,959 Speaker 1: This is this is really that it's too much. It's 276 00:13:58,000 --> 00:14:00,400 Speaker 1: too much to expect. And this, by the way, we 277 00:14:00,400 --> 00:14:02,400 Speaker 1: would put in video in the ballpark of Microsoft, who 278 00:14:02,400 --> 00:14:04,720 Speaker 1: made seventy six billion dollars in their last quarterly earnings, 279 00:14:04,760 --> 00:14:06,839 Speaker 1: and within the neighborhood of Apple, who made ninety four 280 00:14:06,880 --> 00:14:09,160 Speaker 1: billion dollars in their last quarter of earnings. And they 281 00:14:09,160 --> 00:14:11,760 Speaker 1: would do this predominantly making money in an industry that 282 00:14:11,800 --> 00:14:13,760 Speaker 1: a year and a half ago barely made the company 283 00:14:13,800 --> 00:14:17,120 Speaker 1: six billion dollars in a quarter. And the market needs 284 00:14:17,160 --> 00:14:19,640 Speaker 1: in Vidia to perform. They must, they must, as the 285 00:14:19,680 --> 00:14:21,760 Speaker 1: company makes up seven to eight percent of the value 286 00:14:21,800 --> 00:14:24,240 Speaker 1: of the S and P five hundred. It's not enough 287 00:14:24,280 --> 00:14:26,440 Speaker 1: for Invidia to be wildly profitable, or to having a 288 00:14:26,480 --> 00:14:29,240 Speaker 1: monopsony on selling GPUs, or for it to have effectively 289 00:14:29,280 --> 00:14:33,840 Speaker 1: ten x their stock in a few years. No, no, no, more, more, more, 290 00:14:34,000 --> 00:14:36,960 Speaker 1: always more. Number must go up. It must continue to 291 00:14:37,000 --> 00:14:39,920 Speaker 1: grow at the fastest rate of anything ever, making more 292 00:14:39,960 --> 00:14:42,160 Speaker 1: and more money, selling more and more of these GPUs 293 00:14:42,200 --> 00:14:45,920 Speaker 1: to a small group of companies that immediately start losing money. 294 00:14:46,000 --> 00:14:49,400 Speaker 1: The moment they plugged them in. It's not brilliant, is it. 295 00:14:49,880 --> 00:14:51,760 Speaker 1: While a few members of the Magnificent Seven could be 296 00:14:51,800 --> 00:14:53,960 Speaker 1: depended on to funnel tens of billions of dollars into 297 00:14:53,960 --> 00:14:56,960 Speaker 1: a furnace each quarter, there were limits even for companies 298 00:14:57,000 --> 00:14:59,160 Speaker 1: like Microsoft, which had brought over four hundred and eighty 299 00:14:59,160 --> 00:15:02,160 Speaker 1: five thousand g us in twenty twenty four alone. To 300 00:15:02,200 --> 00:15:04,680 Speaker 1: take a step back about how people actually make money 301 00:15:04,800 --> 00:15:08,440 Speaker 1: from buying these GPUs, companies like Microsoft, Google, and Amazon 302 00:15:08,480 --> 00:15:11,280 Speaker 1: make their money by either selling access to large language 303 00:15:11,320 --> 00:15:14,120 Speaker 1: models that people incorporate into their products, or by renting 304 00:15:14,120 --> 00:15:17,080 Speaker 1: out servers full of those GPUs to run influence the thing, 305 00:15:17,200 --> 00:15:20,360 Speaker 1: to generate the output, or train AI models for companies 306 00:15:20,400 --> 00:15:23,600 Speaker 1: that develop and market their models themselves, namely Anthropic and 307 00:15:23,640 --> 00:15:26,480 Speaker 1: Open AI, with some smaller competitors that don't really matter 308 00:15:27,200 --> 00:15:30,680 Speaker 1: that latter revenue stream. Renting out GPUs is where Jensen 309 00:15:30,720 --> 00:15:33,880 Speaker 1: Wong found a solution to that horrible eternal growth problem 310 00:15:34,400 --> 00:15:39,760 Speaker 1: the neo cloud, namely companies like core Weave, Lambda, and Nebius. Now, 311 00:15:39,800 --> 00:15:43,080 Speaker 1: these businesses are fairly straightforward. They own or lease data 312 00:15:43,080 --> 00:15:45,520 Speaker 1: centers that they then feel full of servers that are 313 00:15:45,560 --> 00:15:47,960 Speaker 1: full of Nvidio GPUs, which they then rent out on 314 00:15:48,000 --> 00:15:51,120 Speaker 1: an hourly basis to customers either on a per GPU basis, 315 00:15:51,120 --> 00:15:53,760 Speaker 1: are in large badges for large customers who guarantee they'll 316 00:15:53,800 --> 00:15:55,920 Speaker 1: use a certain amount of compute and sign up for 317 00:15:55,920 --> 00:15:58,640 Speaker 1: a long term agreement for moderan an era a time 318 00:15:58,720 --> 00:16:03,000 Speaker 1: couple years perhaps these larger commitments. A neocloud is a 319 00:16:03,040 --> 00:16:06,440 Speaker 1: specialist cloud compute company that exists only to provide access 320 00:16:06,440 --> 00:16:10,560 Speaker 1: to GPUs for AI, Unlike Amazon Web Services, Microsoft Azure, 321 00:16:10,640 --> 00:16:13,080 Speaker 1: and Google Cloud, all of which to have healthy businesses 322 00:16:13,120 --> 00:16:15,440 Speaker 1: selling other kinds of compute with AI, as I'll get 323 00:16:15,480 --> 00:16:18,120 Speaker 1: into later, failing to provide much of a return on 324 00:16:18,160 --> 00:16:21,000 Speaker 1: investment at all. It's not just the fact that these 325 00:16:21,000 --> 00:16:24,440 Speaker 1: companies are more specialized than say AWS or Azure, as 326 00:16:24,480 --> 00:16:27,040 Speaker 1: you've gathered from the name. These are new, young, and 327 00:16:27,120 --> 00:16:30,680 Speaker 1: in almost all cases incredibly precarious businesses, each with financial 328 00:16:30,720 --> 00:16:34,400 Speaker 1: circumstances that will make a Greek finance minister blush. That's 329 00:16:34,440 --> 00:16:37,520 Speaker 1: because setting up a neocloud is expensive, even if the 330 00:16:37,520 --> 00:16:40,320 Speaker 1: company in question already has data centers as core We've 331 00:16:40,320 --> 00:16:44,240 Speaker 1: did with its cryptocurrency mining operation, AI requires, as I said, 332 00:16:44,280 --> 00:16:47,600 Speaker 1: completely new data center infrastructure to run and call the GPUs, 333 00:16:47,960 --> 00:16:50,160 Speaker 1: and those GPUs also need paying for. And then there's 334 00:16:50,200 --> 00:16:52,400 Speaker 1: the other stuff I mentioned earlier like power, water and 335 00:16:52,440 --> 00:16:54,520 Speaker 1: the other bits of the computer. CPU are the bah 336 00:16:54,640 --> 00:16:57,520 Speaker 1: blah blah blah blah blah. As a result, these neoclouds 337 00:16:57,520 --> 00:16:59,720 Speaker 1: are forced to raise billions of dollars in debt, which 338 00:16:59,720 --> 00:17:03,000 Speaker 1: they can lateralized using the GPUs they already have, along 339 00:17:03,000 --> 00:17:05,520 Speaker 1: with contracts from customers, which they then use to buy 340 00:17:05,640 --> 00:17:10,040 Speaker 1: more GPUs. That's right, they buy GPUs from Nvidia. They 341 00:17:10,440 --> 00:17:13,000 Speaker 1: raised dead on those GPUs and then they used that 342 00:17:13,040 --> 00:17:16,040 Speaker 1: debt to my more GPUs from Nvidia. It's enough to 343 00:17:16,080 --> 00:17:19,200 Speaker 1: drive a man insane. Corwave, for example, has twenty five 344 00:17:19,200 --> 00:17:22,040 Speaker 1: billion dollars in debt on an estimated five point three 345 00:17:22,160 --> 00:17:24,520 Speaker 1: five billion dollars of revenue in twenty twenty five, losing 346 00:17:24,680 --> 00:17:27,600 Speaker 1: hundreds of billions of dollars per quarter. Now you know 347 00:17:27,680 --> 00:17:31,480 Speaker 1: who also invests in these neoclouds. You'll never guess. It's 348 00:17:31,520 --> 00:17:35,160 Speaker 1: in video. Invidia is also one of Corweed's largest customers, 349 00:17:35,160 --> 00:17:37,800 Speaker 1: accounting for fifteen percent of its revenue in twenty twenty four, 350 00:17:38,000 --> 00:17:40,160 Speaker 1: and just signed a deal to buy six point three 351 00:17:40,200 --> 00:17:43,240 Speaker 1: billion dollars of any capacity that core Weave can't otherwise 352 00:17:43,280 --> 00:17:46,480 Speaker 1: sell to someone else through twenty thirty two, an extension 353 00:17:46,480 --> 00:17:48,560 Speaker 1: of a one point three billion dollar twenty twenty three 354 00:17:48,640 --> 00:17:51,440 Speaker 1: deal reported by the Information. It was also the anchor 355 00:17:51,480 --> 00:17:55,480 Speaker 1: investment in Corewey's IPO about two hundred and fifty million dollars. 356 00:17:55,840 --> 00:17:58,000 Speaker 1: In Vidia is currently doing the same thing with Lambda, 357 00:17:58,080 --> 00:18:01,000 Speaker 1: another neocloud that Invidia invested in, which also plans to 358 00:18:01,000 --> 00:18:03,520 Speaker 1: go public next year. In Vidia is also one of 359 00:18:03,640 --> 00:18:06,520 Speaker 1: Landa's largest customers, signing a deal with it this summer 360 00:18:06,560 --> 00:18:09,520 Speaker 1: to rent ten thousand GPUs for one point three billion 361 00:18:09,520 --> 00:18:13,239 Speaker 1: dollars over four years in the UK. And Video has 362 00:18:13,240 --> 00:18:16,200 Speaker 1: also just invested seven hundred million dollars in n Scale, 363 00:18:16,320 --> 00:18:19,119 Speaker 1: a former cryptomner that has never built an AI data 364 00:18:19,160 --> 00:18:22,840 Speaker 1: center that that has despite having no experience, committed one 365 00:18:22,920 --> 00:18:26,320 Speaker 1: billion dollars and or one hundred thousand GPUs to an 366 00:18:26,320 --> 00:18:30,080 Speaker 1: open AI data center in Norway. On Thursday, September twenty fifth, 367 00:18:30,480 --> 00:18:32,960 Speaker 1: Nscale announced that it closed another funding round within Video 368 00:18:33,040 --> 00:18:35,200 Speaker 1: listed as the main backer. Although it's unclear how much 369 00:18:35,200 --> 00:18:37,320 Speaker 1: money it put in, it would be a safe to 370 00:18:37,400 --> 00:18:40,520 Speaker 1: assume it's probably at least one hundred million dollars. In 371 00:18:40,640 --> 00:18:44,280 Speaker 1: Vidia also invested in Nebius, an outgrowth of Russian conglomerate Yandex, 372 00:18:44,280 --> 00:18:46,800 Speaker 1: and Nebbeus provides, through their partnership with Nvidia, tens of 373 00:18:46,840 --> 00:18:49,399 Speaker 1: thousands of dollars of compute credits to companies in Videa's 374 00:18:49,480 --> 00:18:53,800 Speaker 1: Inception startup program. Look, in Vidia's plan is simple, fund 375 00:18:53,800 --> 00:18:56,680 Speaker 1: these neoclouds. Let these neoclouds load themselves up with debt, 376 00:18:56,840 --> 00:18:59,399 Speaker 1: with which point they buy bunches of GPUs from Video, 377 00:18:59,560 --> 00:19:01,760 Speaker 1: which can be used as collateral for loans along with 378 00:19:01,800 --> 00:19:04,440 Speaker 1: contracts and customers, allowing the neoclouds to buy even more 379 00:19:04,480 --> 00:19:10,320 Speaker 1: GPUs from Nvidia. It is just that simple. It's infinite money, right, 380 00:19:10,440 --> 00:19:14,480 Speaker 1: just money, me money. Now, you fund the company, the 381 00:19:14,520 --> 00:19:16,680 Speaker 1: company buys from you, you fund them again. They've used 382 00:19:16,720 --> 00:19:18,480 Speaker 1: the thing they bought to buy from more from you, 383 00:19:18,960 --> 00:19:23,560 Speaker 1: unlimited money. Except that is for ones more problem. These 384 00:19:23,560 --> 00:19:27,000 Speaker 1: companies don't They don't really appear to have that many customers, 385 00:19:27,000 --> 00:19:43,840 Speaker 1: and they don't appear to be making much money. As 386 00:19:43,840 --> 00:19:46,760 Speaker 1: I went into a recent premium newsletter, and Video funds 387 00:19:46,760 --> 00:19:49,720 Speaker 1: and sustains neoclouds as a way of funneling revenue to itself, 388 00:19:49,920 --> 00:19:52,480 Speaker 1: as well as partners like super Micro and Dell resellers 389 00:19:52,560 --> 00:19:55,200 Speaker 1: that take in video GPUs like I mentioned and put 390 00:19:55,240 --> 00:19:57,480 Speaker 1: them in service to sell pre built to customers. These 391 00:19:57,480 --> 00:19:59,920 Speaker 1: two companies made up thirty nine percent of in video 392 00:20:00,119 --> 00:20:05,600 Speaker 1: revenues last quarter. Yet when you remove hyperscalar revenue Microsoft, Amazon, Google, 393 00:20:05,640 --> 00:20:08,400 Speaker 1: Open Ai, and n Video from the revenues of these neoclouds, 394 00:20:08,480 --> 00:20:11,240 Speaker 1: there's barely one billion dollars in revenue can bind across 395 00:20:11,280 --> 00:20:15,120 Speaker 1: core Weave, Nebius and Lambda. Corwev's five point three five 396 00:20:15,200 --> 00:20:17,600 Speaker 1: billion dollars in revenue is predominantly made up with its 397 00:20:17,600 --> 00:20:20,919 Speaker 1: contracts with Nvidia, Microsoft, who are offering that compute to 398 00:20:20,960 --> 00:20:24,320 Speaker 1: open Ai, Google who have hired Corewave to offer compute 399 00:20:24,359 --> 00:20:26,680 Speaker 1: to open Ai. And I'm not kidding, and of course 400 00:20:26,720 --> 00:20:29,600 Speaker 1: open Ai itself, which has now promised core Weave twenty 401 00:20:29,600 --> 00:20:31,960 Speaker 1: two point four billion dollars in business over the next 402 00:20:32,040 --> 00:20:34,959 Speaker 1: five years. This is all a lot of stuff, So 403 00:20:35,000 --> 00:20:37,280 Speaker 1: I'll make it really simple. There's no real money in 404 00:20:37,320 --> 00:20:41,480 Speaker 1: offering AI compute, but that isn't Jensen Huong's problem, So 405 00:20:41,520 --> 00:20:44,439 Speaker 1: we simply will force in Nvidia to hand money to 406 00:20:44,480 --> 00:20:47,119 Speaker 1: these companies that they have contracts to point at so 407 00:20:47,160 --> 00:20:49,760 Speaker 1: they can raise debt to buy more of those GPUs, 408 00:20:49,960 --> 00:20:51,959 Speaker 1: so that in Vidia can give them more contracts they 409 00:20:52,000 --> 00:20:56,480 Speaker 1: can use that to raise more money. It's it's really bad, 410 00:20:56,640 --> 00:20:58,560 Speaker 1: all right, It's really bad. When I read this stuff 411 00:20:58,560 --> 00:21:00,879 Speaker 1: out loud, I feel a little crazy because it's so 412 00:21:01,119 --> 00:21:06,080 Speaker 1: obviously unsustainable. Neil clouds are effectively giant private equity vehicles 413 00:21:06,080 --> 00:21:08,680 Speaker 1: that exist to raise money to buy GPUs from Nvidia 414 00:21:08,960 --> 00:21:11,120 Speaker 1: or for hyperscalers to move money around so they don't 415 00:21:11,119 --> 00:21:14,160 Speaker 1: have to increase their capital expenditures and can, as Microsoft 416 00:21:14,200 --> 00:21:16,600 Speaker 1: did earlier in the year, simply walk away from deals 417 00:21:16,640 --> 00:21:18,520 Speaker 1: they don't like, with the masses of data center at 418 00:21:18,600 --> 00:21:22,000 Speaker 1: least as they walk from. Nebius recently signed a seventeen 419 00:21:22,040 --> 00:21:25,600 Speaker 1: point four billion dollar deal with Microsoft, which even included 420 00:21:25,640 --> 00:21:28,159 Speaker 1: the clause in its six K filing and official filing 421 00:21:28,160 --> 00:21:30,240 Speaker 1: with the government the Microsoft can terminate the deal in 422 00:21:30,280 --> 00:21:32,600 Speaker 1: the event that the capacity isn't built by the delivery dates. 423 00:21:32,760 --> 00:21:36,280 Speaker 1: And by the way, ah Nebius already used the contract 424 00:21:36,280 --> 00:21:38,520 Speaker 1: that Microsoft gave them to raise three billion dollars to 425 00:21:39,640 --> 00:21:42,000 Speaker 1: I'm not shitting you here, build the data center to 426 00:21:42,080 --> 00:21:45,679 Speaker 1: actually to actually provide the compute for the for the contract. 427 00:21:45,720 --> 00:21:49,120 Speaker 1: They don't have it. Yeah, now they don't have them. 428 00:21:49,280 --> 00:21:52,359 Speaker 1: They don't have the fucking compute. They they haven't fucking 429 00:21:52,359 --> 00:21:55,199 Speaker 1: built up. No one built up. They haven't got the compute. Mate, 430 00:21:56,840 --> 00:22:00,800 Speaker 1: These fucking companies are right anyway. Anyway, sorry, sorry, I'll 431 00:22:00,800 --> 00:22:03,760 Speaker 1: stop spiraling. Let me just break down these numbers. Let's 432 00:22:03,760 --> 00:22:06,760 Speaker 1: look at cort we First, Microsoft, they're sixty percent of 433 00:22:06,800 --> 00:22:09,520 Speaker 1: their revenue twenty twenty four and they're providing compute mostly 434 00:22:09,520 --> 00:22:12,280 Speaker 1: for open Ai. Fifteen percent of their revenue last year 435 00:22:12,280 --> 00:22:15,280 Speaker 1: was in VideA, and then the rest was Meta and 436 00:22:15,320 --> 00:22:18,159 Speaker 1: then open Ai and then Google. Lambda half of their 437 00:22:18,160 --> 00:22:20,440 Speaker 1: revenue comes from Amazon and Microsoft, and now one point 438 00:22:20,440 --> 00:22:22,600 Speaker 1: five billion dollars of their revenue comes from in VideA, 439 00:22:23,080 --> 00:22:25,119 Speaker 1: which are their current revenue by the way, and that 440 00:22:25,240 --> 00:22:27,679 Speaker 1: one point five billion dollars over four years. So the 441 00:22:27,720 --> 00:22:30,760 Speaker 1: current revenue, it's two hundred and fifty billion dollars. Well, 442 00:22:30,760 --> 00:22:33,040 Speaker 1: that would make in video the largest customer. I realize 443 00:22:33,040 --> 00:22:36,680 Speaker 1: I'm just saying numbers here, but for real, with that contract, 444 00:22:36,760 --> 00:22:39,480 Speaker 1: because Lambda only made two hundred and fifty million dollars 445 00:22:39,520 --> 00:22:41,400 Speaker 1: in the first half of this year, and in Video 446 00:22:41,520 --> 00:22:43,840 Speaker 1: is spreading one point five billion dollars across four years. 447 00:22:43,920 --> 00:22:46,560 Speaker 1: In Video is the largest customer now Now Nebutus has 448 00:22:46,560 --> 00:22:49,639 Speaker 1: got similar revenue to Lambda, but their largest customer is 449 00:22:49,680 --> 00:22:55,120 Speaker 1: now is It's it's fucking Microsoft. It's just they don't 450 00:22:55,160 --> 00:22:59,040 Speaker 1: have real customers. They just have hyperscalers or in VideA themselves. 451 00:23:00,160 --> 00:23:02,520 Speaker 1: And from my analysis, it appears that Core we've despited 452 00:23:02,560 --> 00:23:05,400 Speaker 1: expectations to make that five point three five billion dollars 453 00:23:05,440 --> 00:23:08,480 Speaker 1: this year, has only around five hundred million dollars of 454 00:23:08,520 --> 00:23:11,800 Speaker 1: non Magnificent seven or open AI revenue in twenty twenty five, 455 00:23:11,840 --> 00:23:14,560 Speaker 1: with Lambda estimated to have maybe a round of one 456 00:23:14,640 --> 00:23:17,560 Speaker 1: hundred million dollars in AI revenue. Otherwise, Nebius only around 457 00:23:17,600 --> 00:23:21,400 Speaker 1: two hundred and fifty million dollars, and that's being generous. 458 00:23:21,440 --> 00:23:24,800 Speaker 1: In much simpler terms, the Magnificent seven is the AI bubble, 459 00:23:24,800 --> 00:23:27,359 Speaker 1: and the AI bubble exists to buy more GPUs because, 460 00:23:27,400 --> 00:23:30,040 Speaker 1: as I'll talk about, there's no real money or growth 461 00:23:30,040 --> 00:23:32,040 Speaker 1: coming out of this other than the amount that private 462 00:23:32,119 --> 00:23:34,840 Speaker 1: credit is investing. And this really is quite worrying. By 463 00:23:34,880 --> 00:23:36,920 Speaker 1: the way, I had a quote here for analyst that 464 00:23:37,000 --> 00:23:39,639 Speaker 1: says is about fifty billion dollars a quarter for the 465 00:23:39,640 --> 00:23:42,960 Speaker 1: low end for the past three quarters. So why is 466 00:23:43,000 --> 00:23:43,480 Speaker 1: this bad? 467 00:23:44,160 --> 00:23:44,560 Speaker 2: All right? 468 00:23:44,760 --> 00:23:47,919 Speaker 1: I don't know. Let's start simple. Fifty billion dollars a 469 00:23:47,960 --> 00:23:50,600 Speaker 1: quarter of data center funding is going into an industry 470 00:23:50,600 --> 00:23:53,680 Speaker 1: that has less revenue than three to play mobile game 471 00:23:53,720 --> 00:23:56,919 Speaker 1: gensin impact. That feels pretty bad. Who's going to use 472 00:23:56,960 --> 00:23:59,000 Speaker 1: these data centers? How are they even going to make 473 00:23:59,040 --> 00:24:02,080 Speaker 1: money on them? Firms don't typically hold on to assets. 474 00:24:02,200 --> 00:24:04,160 Speaker 1: They sell them or they take them public. That doesn't 475 00:24:04,160 --> 00:24:07,320 Speaker 1: seem great to me anyway. If AI was truly the 476 00:24:07,359 --> 00:24:10,640 Speaker 1: next big growth vehicle, neoclouds would be swimming in diverse 477 00:24:10,680 --> 00:24:14,440 Speaker 1: global revenue streams. Instead, they're heavily centralized around the same 478 00:24:14,480 --> 00:24:17,399 Speaker 1: few names, one of which in Vidia, directly benefits from 479 00:24:17,440 --> 00:24:20,119 Speaker 1: their existence not as a company doing business, but as 480 00:24:20,160 --> 00:24:22,960 Speaker 1: an entity that can accrue debt and spend money on GPUs. 481 00:24:23,520 --> 00:24:26,439 Speaker 1: These neoclouds are entirely dependent on a continual flow of 482 00:24:26,440 --> 00:24:28,800 Speaker 1: private credit from firms like Gold and Sachs, Who's becked, 483 00:24:28,800 --> 00:24:33,640 Speaker 1: Nebbia's Corweave and Lambda, JP Morgan, Lambda Crusoe Building Applene, 484 00:24:33,640 --> 00:24:36,200 Speaker 1: Texas As Open AI Data Center, and of course Corwave 485 00:24:36,359 --> 00:24:39,240 Speaker 1: and Blackstone Lambda and corwav, who have in a very 486 00:24:39,280 --> 00:24:42,360 Speaker 1: real sense created an entirely debt based infrastructure to feed 487 00:24:42,400 --> 00:24:45,440 Speaker 1: billions of dollars directly to Nvidia, all in the name 488 00:24:45,480 --> 00:24:49,160 Speaker 1: of an AI revolution that's yet to arrive. The fact 489 00:24:49,160 --> 00:24:51,960 Speaker 1: that the rest of the neoclo revenue stream is effectively 490 00:24:51,960 --> 00:24:55,920 Speaker 1: either a hyperscaler or open ai is also concerning. Hyperscalers 491 00:24:55,920 --> 00:24:58,359 Speaker 1: are at this point the majority of data center capital 492 00:24:58,359 --> 00:25:00,399 Speaker 1: expenditures and have yet to prove an any kind of 493 00:25:00,480 --> 00:25:03,760 Speaker 1: success from building out this capacity, outside of course, Microsoft's 494 00:25:03,800 --> 00:25:07,000 Speaker 1: investment in open Ai, which has succeeded in generating revenue 495 00:25:07,040 --> 00:25:10,120 Speaker 1: while burning billions of dollars of revenue on well, I mean, 496 00:25:10,160 --> 00:25:12,919 Speaker 1: it's not really any profit, is they have just burning money. 497 00:25:13,560 --> 00:25:16,280 Speaker 1: It's also insane when you say this stuff. I've got 498 00:25:16,280 --> 00:25:18,960 Speaker 1: two more goddamn episodes of this and when I read 499 00:25:19,000 --> 00:25:22,280 Speaker 1: these scripts, I'm just like, how is nobody else more 500 00:25:22,359 --> 00:25:27,840 Speaker 1: freaked out? Oh well, hyperscaler revenue is also capricious. But 501 00:25:27,920 --> 00:25:30,800 Speaker 1: even if it isn't, why are there no other major customers? 502 00:25:30,800 --> 00:25:33,160 Speaker 1: Why across all of these companies does there not seem 503 00:25:33,200 --> 00:25:36,680 Speaker 1: to be one major customer who isn't open Ai Well, 504 00:25:36,720 --> 00:25:39,440 Speaker 1: the answers are quite obvious. Nobody that wants it can 505 00:25:39,440 --> 00:25:41,800 Speaker 1: afford it, and those that can afford it don't need it. 506 00:25:41,800 --> 00:25:44,959 Speaker 1: It's also unclear what exactly hyperscalers are doing with this compute, 507 00:25:44,960 --> 00:25:48,000 Speaker 1: because it sure isn't making money. While Microsoft makes ten 508 00:25:48,040 --> 00:25:50,400 Speaker 1: billion dollars in revenue from renting compute to open Ai 509 00:25:50,520 --> 00:25:53,320 Speaker 1: via their Microsoft as Your Cloud, it does so at cost, 510 00:25:54,200 --> 00:25:56,840 Speaker 1: and was charging open Ai one dollars and thirty cents 511 00:25:56,880 --> 00:26:00,639 Speaker 1: per hour for each a one hundred AIGPU at a 512 00:26:00,680 --> 00:26:03,560 Speaker 1: loss of two point two dollars an hour per GPU, 513 00:26:03,880 --> 00:26:06,360 Speaker 1: meaning that it is likely losing money on this compute, 514 00:26:06,440 --> 00:26:09,200 Speaker 1: especially as semi analysis as the total cost per hour 515 00:26:09,240 --> 00:26:12,040 Speaker 1: per GPU around one dollars and forty six cents with 516 00:26:12,080 --> 00:26:14,880 Speaker 1: the cost of capital and debt associated for a hyperscaler, 517 00:26:15,200 --> 00:26:17,680 Speaker 1: though it's unclear that whether that's for an H one 518 00:26:17,720 --> 00:26:21,040 Speaker 1: hundred or an A one hundred GPU. In any case, 519 00:26:21,480 --> 00:26:23,560 Speaker 1: how do these neoclouds pay for their debt if the 520 00:26:23,600 --> 00:26:26,520 Speaker 1: hyperscalers give up or in video doesn't send them money, 521 00:26:26,600 --> 00:26:29,640 Speaker 1: or more likely private credit begins to notice that there's 522 00:26:29,680 --> 00:26:32,200 Speaker 1: no real revenue growth outside of circular compute deals with 523 00:26:32,280 --> 00:26:35,840 Speaker 1: neocloud's largest suppliers, investors and customers. Don't know why I 524 00:26:35,880 --> 00:26:38,720 Speaker 1: said plural there, because it's just one in video, And 525 00:26:38,800 --> 00:26:41,080 Speaker 1: the answer is they don't. In fact, I have serious 526 00:26:41,119 --> 00:26:44,639 Speaker 1: concerns that they can't even build the capacity necessary to 527 00:26:44,680 --> 00:26:47,080 Speaker 1: fulfill these deals, but nobody seems to worry or think 528 00:26:47,119 --> 00:26:49,720 Speaker 1: about them. But really, though it appears to be taking 529 00:26:49,760 --> 00:26:52,800 Speaker 1: Oracle and Cruso around two point five years per gigawatt 530 00:26:52,880 --> 00:26:56,800 Speaker 1: of compute capacity, how exactly are any of these neoclouds, 531 00:26:56,880 --> 00:27:00,000 Speaker 1: or indeed Oracle itself able to expand to capture this revenue. 532 00:27:00,560 --> 00:27:02,760 Speaker 1: Who knows, but I assume somebody is going to say 533 00:27:02,800 --> 00:27:06,040 Speaker 1: open Ai. Here's an insane statistic for you. By the way, 534 00:27:06,240 --> 00:27:08,760 Speaker 1: open ai will account for in both its revenue projected 535 00:27:08,800 --> 00:27:12,560 Speaker 1: thirteen billion dollars and in its own compute cost ten dollars, 536 00:27:12,600 --> 00:27:14,919 Speaker 1: somewhere in the region of forty to fifty percent of 537 00:27:14,960 --> 00:27:18,000 Speaker 1: all AI revenues in twenty twenty five. As a reminder, 538 00:27:18,080 --> 00:27:19,879 Speaker 1: open ai is leaked that it will burn one hundred 539 00:27:19,880 --> 00:27:21,800 Speaker 1: and fifteen billion dollars in the next four years, and 540 00:27:21,880 --> 00:27:24,800 Speaker 1: based on my estimates, it actually needs to raise I 541 00:27:24,840 --> 00:27:27,080 Speaker 1: mean upwards are four hundred billion dollars in the next 542 00:27:27,080 --> 00:27:29,120 Speaker 1: four years based on its three hundred billion dollar deal 543 00:27:29,200 --> 00:27:32,520 Speaker 1: with Oracle and some recently announced one hundred billion dollar 544 00:27:32,560 --> 00:27:36,680 Speaker 1: compute purchases for backup. And that alone is a very 545 00:27:36,760 --> 00:27:39,960 Speaker 1: bad sign, very very bad indeed. Especially it's with three 546 00:27:40,040 --> 00:27:42,560 Speaker 1: years and five hundred billion dollars or more into this 547 00:27:42,640 --> 00:27:45,760 Speaker 1: hype cycle, with few signs of life outside of well 548 00:27:45,920 --> 00:27:49,720 Speaker 1: open AI promising people money, and that's not healthy or 549 00:27:49,960 --> 00:27:53,520 Speaker 1: sane or normal, and it's certainly not stable, and it's 550 00:27:53,560 --> 00:28:05,720 Speaker 1: going to get bad real fast. Gatget tomorrow. Thank you 551 00:28:05,720 --> 00:28:08,440 Speaker 1: for listening to Better Offline. The editor and composer of 552 00:28:08,480 --> 00:28:11,440 Speaker 1: the Better Offline theme song is Matasowski. You can check 553 00:28:11,480 --> 00:28:14,400 Speaker 1: out more of his music and audio projects at Matasowski 554 00:28:14,440 --> 00:28:17,960 Speaker 1: dot com, M A T T O S O W 555 00:28:18,280 --> 00:28:21,600 Speaker 1: s ki dot com. You can email me at easy 556 00:28:21,680 --> 00:28:24,320 Speaker 1: at Better offline dot com or visit Better Offline dot 557 00:28:24,320 --> 00:28:26,960 Speaker 1: com to find more podcast links and of course my newsletter. 558 00:28:27,359 --> 00:28:29,840 Speaker 1: I also really recommend you go to chat dot where's 559 00:28:29,880 --> 00:28:32,359 Speaker 1: youreaed dot at to visit the discord, and go to 560 00:28:32,400 --> 00:28:35,800 Speaker 1: our slash Better Offline to check out our reddit. Thank 561 00:28:35,800 --> 00:28:38,040 Speaker 1: you so much for listening better. 562 00:28:38,080 --> 00:28:40,720 Speaker 2: Offline is a production of cool Zone Media. For more 563 00:28:40,760 --> 00:28:44,640 Speaker 2: from cool Zone Media, visit our website cool zonemedia dot com, 564 00:28:44,720 --> 00:28:47,600 Speaker 2: or check us out on the iHeartRadio app, Apple Podcasts, 565 00:28:47,680 --> 00:29:02,200 Speaker 2: or wherever you get your podcasts. FI