1 00:00:03,200 --> 00:00:07,080 Speaker 1: Media. Hello and welcome back to Better Offline. I'm your 2 00:00:07,120 --> 00:00:20,880 Speaker 1: host ed ZiT Try. This is the last part of 3 00:00:20,920 --> 00:00:23,640 Speaker 1: our four part series about how the financial system and 4 00:00:23,680 --> 00:00:28,040 Speaker 1: the financial markets are the last victims of the inshitification process. 5 00:00:28,920 --> 00:00:31,400 Speaker 1: Over the past four episodes, in a twenty thousand or 6 00:00:31,400 --> 00:00:34,400 Speaker 1: so word newsletter which this series is based on, I've 7 00:00:34,440 --> 00:00:37,280 Speaker 1: explained how and in this final episode, I'm going to 8 00:00:37,320 --> 00:00:40,479 Speaker 1: explain how all this was going to fall apart and 9 00:00:40,560 --> 00:00:43,800 Speaker 1: make it clear where I'm so fucking alarmed. Now let's 10 00:00:43,800 --> 00:00:46,920 Speaker 1: get back to those data centers. We've already got some 11 00:00:47,640 --> 00:00:51,159 Speaker 1: signs of concern within the banking world around them, and 12 00:00:51,200 --> 00:00:53,920 Speaker 1: I'm kind of worried, and so are they about their exposure. 13 00:00:54,440 --> 00:00:57,240 Speaker 1: In November, the ft reported the Deutsche Bank, which back 14 00:00:57,320 --> 00:00:59,920 Speaker 1: cor we've multiple times and several data centers, was the 15 00:01:00,080 --> 00:01:03,000 Speaker 1: exploring ways the hedge it's exposure to the AI data 16 00:01:03,040 --> 00:01:06,520 Speaker 1: centers after extending billions of dollars in debt, including shortening 17 00:01:06,560 --> 00:01:09,880 Speaker 1: a basket of AI related stocks or buying default protection 18 00:01:09,920 --> 00:01:12,880 Speaker 1: on some of its debt using synthetic risk transfers, which 19 00:01:12,920 --> 00:01:15,760 Speaker 1: are when a bank sells the full or partial credit 20 00:01:15,800 --> 00:01:17,920 Speaker 1: risk of a loan or loans to another bank or 21 00:01:18,040 --> 00:01:20,080 Speaker 1: keeping the loans on their book paying a monthly fee 22 00:01:20,120 --> 00:01:23,560 Speaker 1: to investors. This is a simplification. Please don't be mad 23 00:01:23,600 --> 00:01:28,320 Speaker 1: at me. In December, Fortune reported the Morgan Stanley three 24 00:01:28,360 --> 00:01:31,720 Speaker 1: times with core Weave IPI partners, Hyperions, Soft Banks Bridge 25 00:01:31,720 --> 00:01:35,560 Speaker 1: Loan was also considering synthetic risk transfers on loans to 26 00:01:35,640 --> 00:01:39,119 Speaker 1: businesses involved in AI infrastructure. One of what those were 27 00:01:39,160 --> 00:01:43,600 Speaker 1: for now. Back in April, SMBC sold synthetic risk transfers. 28 00:01:44,640 --> 00:01:48,640 Speaker 1: Back in April, SMBC sold synthetic risk transfers tied to 29 00:01:48,680 --> 00:01:52,280 Speaker 1: private debt BDCs. And while this predates the larger data 30 00:01:52,320 --> 00:01:55,080 Speaker 1: center deals done by Blue Al, who is a BDC themselves, 31 00:01:55,440 --> 00:01:59,600 Speaker 1: smbc's oversee multiple blue ol deals in the past. In December, 32 00:01:59,800 --> 00:02:04,040 Speaker 1: s MBC closed another SRT selling off risk from Australian 33 00:02:04,080 --> 00:02:07,720 Speaker 1: and Asian project finance loans, though we can't confirm if 34 00:02:07,720 --> 00:02:10,480 Speaker 1: any of those were data center related now. In December, 35 00:02:10,520 --> 00:02:13,640 Speaker 1: Goldman Sachs pause the planned mortgage bond sale for data 36 00:02:13,639 --> 00:02:17,079 Speaker 1: Center Operate A Cyrus one, with the intent to revive 37 00:02:17,120 --> 00:02:18,919 Speaker 1: it in the first quarter of twenty twenty six. I 38 00:02:18,919 --> 00:02:22,240 Speaker 1: have heard nothing from that since Oracle's credit risk reaches 39 00:02:22,280 --> 00:02:24,880 Speaker 1: sixteen year high in the middle of December, with credit 40 00:02:24,919 --> 00:02:28,200 Speaker 1: default swaps basically betting the Oracle will default on its debt, 41 00:02:28,200 --> 00:02:31,320 Speaker 1: which is unlikely yet no longer impossible, climbing to their 42 00:02:31,400 --> 00:02:36,079 Speaker 1: highest price since the Great Financial Crisis. While Morgan Stanley 43 00:02:36,120 --> 00:02:38,720 Speaker 1: and Deutsche Banks SRTs are yet to close, it's still 44 00:02:38,760 --> 00:02:40,880 Speaker 1: notable that two of the largest players in data center 45 00:02:40,919 --> 00:02:44,600 Speaker 1: financing feel the need to hedge their bets. So what 46 00:02:44,720 --> 00:02:48,320 Speaker 1: exactly are they hedging against? Simple that tenants won't arrive 47 00:02:48,400 --> 00:02:51,440 Speaker 1: and debts won't get paid. I also believe that they're 48 00:02:51,440 --> 00:02:54,160 Speaker 1: going to need bigger hedges because I don't think there's 49 00:02:54,240 --> 00:02:58,200 Speaker 1: enough actual AI demand to meet the data center's being built, 50 00:02:58,240 --> 00:03:01,000 Speaker 1: and I think most data center loans up being underwater 51 00:03:01,040 --> 00:03:03,920 Speaker 1: within the next two years. To be clear, this isn't 52 00:03:03,919 --> 00:03:06,799 Speaker 1: a speculative chance, but rather one wrong on the chain 53 00:03:06,840 --> 00:03:09,320 Speaker 1: of pain that's been forged and looks like it's ready 54 00:03:09,360 --> 00:03:12,320 Speaker 1: to snap. Let's start by quoting my Premium newsletter from 55 00:03:12,320 --> 00:03:15,920 Speaker 1: December fifth. While many people talk about how circular the 56 00:03:15,960 --> 00:03:18,400 Speaker 1: AI bubble may or may not be, the reality is 57 00:03:18,440 --> 00:03:20,919 Speaker 1: that it's far more like a chain, a deeply vulnerable 58 00:03:20,919 --> 00:03:24,160 Speaker 1: one held together by debt and venture capital. A company 59 00:03:24,160 --> 00:03:26,799 Speaker 1: buys a GPU from Nvidia, at which point nobody is 60 00:03:26,840 --> 00:03:29,679 Speaker 1: making any profit anymore. These GPUs are purchased for the 61 00:03:29,760 --> 00:03:33,160 Speaker 1: most part using debt provided by banks or financial institutions, 62 00:03:33,520 --> 00:03:36,720 Speaker 1: while hyperscalers like Microsoft can and do fund GPUs using 63 00:03:36,760 --> 00:03:39,160 Speaker 1: cash flow even they have started to turn to debt. 64 00:03:40,400 --> 00:03:42,520 Speaker 1: At that point, the company that bought the GPUs six 65 00:03:42,680 --> 00:03:45,160 Speaker 1: hundreds of millions of dollars or even billions of dollars 66 00:03:45,240 --> 00:03:47,560 Speaker 1: to build out a data center, and once it turns on, 67 00:03:48,000 --> 00:03:51,240 Speaker 1: provides compute to a model provider such as open ai Oranthropic, 68 00:03:51,440 --> 00:03:54,320 Speaker 1: or even a smaller company, which then begins losing money 69 00:03:54,360 --> 00:03:57,600 Speaker 1: selling access to those GPUs, or of course training models 70 00:03:57,600 --> 00:04:00,920 Speaker 1: on them, which only loses money eat sample. Both open 71 00:04:00,960 --> 00:04:03,560 Speaker 1: ai and Anthropic loose billions of dollars, and both rely 72 00:04:03,680 --> 00:04:06,400 Speaker 1: on venture capital to fund their ability to continue paying 73 00:04:06,640 --> 00:04:10,960 Speaker 1: for accessing those GPUs. At that point, open ai and 74 00:04:10,960 --> 00:04:14,160 Speaker 1: Anthropic offer either subscriptions, which cost far more to offer 75 00:04:14,160 --> 00:04:16,760 Speaker 1: than the revenue they provide, or API access to their 76 00:04:16,800 --> 00:04:20,800 Speaker 1: models on a per million token basis. These rates are 77 00:04:20,839 --> 00:04:24,760 Speaker 1: often subsidized too. AI startups pay to access those models 78 00:04:24,800 --> 00:04:27,000 Speaker 1: to run their services, which end up costing more than 79 00:04:27,040 --> 00:04:29,360 Speaker 1: the revenue that they make from offering them, which means 80 00:04:29,400 --> 00:04:31,720 Speaker 1: they have to raise venture capital to continue paying to 81 00:04:31,760 --> 00:04:35,240 Speaker 1: access those models. Outside of hyperscalers paying in video for 82 00:04:35,279 --> 00:04:38,080 Speaker 1: GPUs out of cash flow, None of the AI industry 83 00:04:38,120 --> 00:04:42,400 Speaker 1: is fueled by revenue. Every single part of the industry 84 00:04:42,680 --> 00:04:45,800 Speaker 1: is fueled by a kind of subsidy, debt, or equity. 85 00:04:46,680 --> 00:04:49,040 Speaker 1: As a result, the AI bubble is really a stress 86 00:04:49,080 --> 00:04:52,640 Speaker 1: test of the global venture capital, private equity, private credit, 87 00:04:52,880 --> 00:04:56,200 Speaker 1: institutional and banking system and its willingness to fund all 88 00:04:56,240 --> 00:04:58,640 Speaker 1: of this forever. Because there isn't a single generative AI 89 00:04:58,720 --> 00:05:02,560 Speaker 1: company that's got a path to profit stability. You see, 90 00:05:02,680 --> 00:05:05,200 Speaker 1: every little link in the chain of pain is necessary 91 00:05:05,200 --> 00:05:09,280 Speaker 1: to understand things. The intitified stock market pumped not by 92 00:05:09,360 --> 00:05:12,760 Speaker 1: actual cash flow or productivity, by which I mean products 93 00:05:12,760 --> 00:05:16,239 Speaker 1: that are selling Google, Microsoft, Amazon matter they don't actually 94 00:05:16,279 --> 00:05:19,520 Speaker 1: generate real revenue and definitely not profit from AI, but 95 00:05:19,560 --> 00:05:22,400 Speaker 1: by signals read by analysts and investors trained over decades 96 00:05:22,400 --> 00:05:25,080 Speaker 1: the push consumers to invest in the Magnificent seven stocks. 97 00:05:25,560 --> 00:05:27,800 Speaker 1: These stocks represent as much as forty percent of the 98 00:05:27,839 --> 00:05:30,279 Speaker 1: value of the S and P five hundred and their values, 99 00:05:30,279 --> 00:05:32,520 Speaker 1: as I said, were pumped by analysts and the media, 100 00:05:32,560 --> 00:05:37,240 Speaker 1: misleading investors into believing that revenue growth for companies like Microsoft, Meta, Amazon, 101 00:05:37,279 --> 00:05:39,880 Speaker 1: and Google have anything to do with AI versus the 102 00:05:39,920 --> 00:05:42,520 Speaker 1: growth of their regular services. The other part in the 103 00:05:42,600 --> 00:05:45,440 Speaker 1: chain is venture capital's liquidity crisis, one peaking at a 104 00:05:45,440 --> 00:05:48,360 Speaker 1: time when startups have become more capital intensive than any 105 00:05:48,400 --> 00:05:51,120 Speaker 1: other point in history, and the underlying value of their 106 00:05:51,160 --> 00:05:54,560 Speaker 1: investments is only valuable as long as these companies stay alive. 107 00:05:54,960 --> 00:05:57,599 Speaker 1: And then, of course, the ballooning centralized data center debt 108 00:05:57,640 --> 00:06:00,680 Speaker 1: bubble funded based on customer contracts, will built for demand 109 00:06:00,680 --> 00:06:03,800 Speaker 1: that does not exist, finding massive data centers of GPUs 110 00:06:03,800 --> 00:06:06,799 Speaker 1: that immediately become commoditized as a result of this hysteria. 111 00:06:07,480 --> 00:06:10,719 Speaker 1: In really simple terms, I believe that almost every investment 112 00:06:10,720 --> 00:06:13,200 Speaker 1: in a data center AI startup may go to zero. 113 00:06:14,000 --> 00:06:17,000 Speaker 1: Let me explain if we assume that half of the 114 00:06:17,040 --> 00:06:19,120 Speaker 1: one hundred and seventy one point five billion dollars of 115 00:06:19,200 --> 00:06:21,799 Speaker 1: data center debt is in GPUs, so eighty five billion 116 00:06:21,920 --> 00:06:24,360 Speaker 1: or so, that's about three point two gigawats of data 117 00:06:24,360 --> 00:06:26,960 Speaker 1: set capacity. Based on my model of in Videa's approximate 118 00:06:27,000 --> 00:06:30,719 Speaker 1: split of sales between different aigpus from a premium priest 119 00:06:30,760 --> 00:06:33,520 Speaker 1: from a few weeks ago. As a brief explainer, I 120 00:06:33,560 --> 00:06:36,039 Speaker 1: made the model by taking in videos data center revenues, 121 00:06:36,160 --> 00:06:38,920 Speaker 1: estimating how many of each GPU they sell, and dividing 122 00:06:38,960 --> 00:06:41,400 Speaker 1: those revenues by those numbers, and markeplying the result by 123 00:06:41,400 --> 00:06:43,880 Speaker 1: the power output of the GPUs. In really simple terms, 124 00:06:43,960 --> 00:06:45,880 Speaker 1: I worked out how many GPUs they sold them what 125 00:06:46,160 --> 00:06:48,040 Speaker 1: the power to run them would be. There'd be about 126 00:06:48,040 --> 00:06:51,880 Speaker 1: three point two gigawts. The likelihood of the majority of 127 00:06:51,880 --> 00:06:54,640 Speaker 1: these products being a completed within the next year and 128 00:06:54,680 --> 00:06:58,440 Speaker 1: b completed on budget is very very small. Every delay 129 00:06:58,520 --> 00:07:01,839 Speaker 1: increases the likelihood off, as each of these projects is 130 00:07:01,880 --> 00:07:05,080 Speaker 1: heavily debt based. I also want to add open Ai claims, 131 00:07:05,080 --> 00:07:07,800 Speaker 1: and I actually think they're bullshitting here that they had 132 00:07:07,839 --> 00:07:11,240 Speaker 1: two gigawatts of compute at the end of twenty twenty five, 133 00:07:12,200 --> 00:07:15,679 Speaker 1: two gigawatts. They are the only company with any real 134 00:07:15,680 --> 00:07:19,000 Speaker 1: consumer exposure. They're the only ones with any number of 135 00:07:19,040 --> 00:07:22,520 Speaker 1: real users other than Google, who has literally changed Google 136 00:07:22,520 --> 00:07:25,640 Speaker 1: a system for Gemini. I don't think there's even like 137 00:07:26,080 --> 00:07:29,440 Speaker 1: five hundred megawatts of real demand outside of people training 138 00:07:29,480 --> 00:07:33,280 Speaker 1: models or Elon Musk's s SAM Generator, and that's a 139 00:07:33,320 --> 00:07:37,280 Speaker 1: whole different story now. In general, the customers of these 140 00:07:37,280 --> 00:07:40,120 Speaker 1: projects are either hyperscalers who are only doing AI because 141 00:07:40,120 --> 00:07:42,240 Speaker 1: they have no other hyper growth ideas and because Wall 142 00:07:42,280 --> 00:07:46,040 Speaker 1: Street currently approves, or AI startups, all of whom are unprofitable. 143 00:07:47,160 --> 00:07:49,520 Speaker 1: While there are potentially hedge funds or other small companies 144 00:07:49,560 --> 00:07:52,120 Speaker 1: looking for private AI integrations. I think this is a 145 00:07:52,200 --> 00:07:55,400 Speaker 1: very very small market. On top of that, AI compute 146 00:07:55,400 --> 00:07:58,240 Speaker 1: itself may not be profitable. And because by my estima, 147 00:07:58,520 --> 00:08:01,240 Speaker 1: everybody is spent about eighty five billion dollars on filling 148 00:08:01,320 --> 00:08:04,360 Speaker 1: data centers with the same GPUs, the aggregate price of 149 00:08:04,400 --> 00:08:08,080 Speaker 1: renting out GPUs will decline. Already, the average price of 150 00:08:08,080 --> 00:08:10,600 Speaker 1: renting out Blackwell GPUs is declined to an average of 151 00:08:10,680 --> 00:08:13,200 Speaker 1: four dollars and forty five cents an hour according to 152 00:08:13,240 --> 00:08:16,280 Speaker 1: Silicon Data, and that's before the majority of black Well 153 00:08:16,320 --> 00:08:20,120 Speaker 1: powered GPUs come online. Yeah, the customer base shrinks from 154 00:08:20,160 --> 00:08:23,760 Speaker 1: there because the majority of AI startups aren't actually renting GPUs. 155 00:08:24,080 --> 00:08:26,600 Speaker 1: They build products on top of models built by Open 156 00:08:26,680 --> 00:08:29,520 Speaker 1: Ai or Anthropic, who have made it clear they're buying 157 00:08:29,520 --> 00:08:32,640 Speaker 1: capacity from either hyperscalers, or, in open AI's case, getting 158 00:08:32,679 --> 00:08:35,280 Speaker 1: Oracle or core Weave to build it for them. Why 159 00:08:35,559 --> 00:08:39,400 Speaker 1: Because building your own model is incredibly capital intensive, both 160 00:08:39,440 --> 00:08:41,160 Speaker 1: on the compute and the data you have to get, 161 00:08:41,520 --> 00:08:43,080 Speaker 1: and it's hard to tell if the results will be 162 00:08:43,080 --> 00:08:47,440 Speaker 1: worth it. Now, let's assume I don't actually believe it will, 163 00:08:47,440 --> 00:08:49,640 Speaker 1: but let's try anyway that all of that three point 164 00:08:49,640 --> 00:08:55,000 Speaker 1: two gigawatsop capacity comes online? How much of it? How much? Actually? 165 00:08:55,640 --> 00:08:58,319 Speaker 1: How much compute does an AI company use? Like I said, 166 00:08:58,360 --> 00:09:00,920 Speaker 1: open Ai ended twenty twenty five to two gig what's 167 00:09:00,960 --> 00:09:03,720 Speaker 1: of capacity? And they apparently have nine hundred million weekly 168 00:09:03,760 --> 00:09:06,480 Speaker 1: active users. I don't think there are any AI companies 169 00:09:06,480 --> 00:09:08,760 Speaker 1: with even ten percent of that user base. But even 170 00:09:08,760 --> 00:09:10,840 Speaker 1: if they were, open ai spent eight point six seven 171 00:09:10,880 --> 00:09:14,160 Speaker 1: billion dollars on inference through the end of September. Who 172 00:09:14,240 --> 00:09:16,200 Speaker 1: can afford to pay even ten percent of that a 173 00:09:16,240 --> 00:09:19,840 Speaker 1: year or five percent? Yeah? In reality, open Ai is 174 00:09:19,920 --> 00:09:22,360 Speaker 1: likely more indicative of the overall compute spend of the 175 00:09:22,520 --> 00:09:25,960 Speaker 1: entire AI industry. As I've said, most companies are powered 176 00:09:25,960 --> 00:09:28,240 Speaker 1: not by their own GPU driven models, but by renting 177 00:09:28,240 --> 00:09:31,920 Speaker 1: them from model providers like open Ai. Themselves. Open Ai 178 00:09:31,960 --> 00:09:34,200 Speaker 1: and and Thropic Spenter combined eleven point three to three 179 00:09:34,200 --> 00:09:37,760 Speaker 1: billion dollars in compute on Zoro AWS, respectively through the 180 00:09:37,760 --> 00:09:41,120 Speaker 1: first nine months of last year. And that's my own reporting, 181 00:09:41,160 --> 00:09:43,640 Speaker 1: by the way, and as the two largest consumers of 182 00:09:43,679 --> 00:09:48,160 Speaker 1: AI compute, Well, I'm a little worried. These things suggest 183 00:09:48,280 --> 00:09:51,839 Speaker 1: two things. The market for AI compute is actually very 184 00:09:51,960 --> 00:09:54,800 Speaker 1: very small. If you assume that Anthropics spent the same 185 00:09:54,840 --> 00:09:57,520 Speaker 1: amount on Google Cloud as it did in AWS, that'd 186 00:09:57,559 --> 00:09:59,959 Speaker 1: be about a total of five point three two billion dollars, 187 00:10:00,000 --> 00:10:02,680 Speaker 1: and that Core Weaves revenue called it five billion dollars, 188 00:10:02,960 --> 00:10:05,839 Speaker 1: most of which was open Ai via Microsoft or in video. 189 00:10:06,160 --> 00:10:09,000 Speaker 1: That doesn't appear to be an AI compute market outside 190 00:10:09,040 --> 00:10:13,120 Speaker 1: of servicing two companies. The other problem is the market 191 00:10:13,160 --> 00:10:15,680 Speaker 1: for AI compute is not actually growing. In the last 192 00:10:15,920 --> 00:10:19,480 Speaker 1: two years, no new major customers of AI compute have emerged. 193 00:10:19,880 --> 00:10:22,840 Speaker 1: Every company that assigned a large compute deal has either 194 00:10:22,840 --> 00:10:26,920 Speaker 1: been open Ai, Anthropic or a hyperscaler. Even if Cursor 195 00:10:27,080 --> 00:10:29,840 Speaker 1: and AI Coding Company were to dump its entire two 196 00:10:29,920 --> 00:10:32,680 Speaker 1: point three billion dollars into funding AI compute, that would 197 00:10:32,679 --> 00:10:35,640 Speaker 1: still not be very much. In fact, it would take 198 00:10:35,720 --> 00:10:39,000 Speaker 1: sinking every single dollar of venture capital over two hundred 199 00:10:39,000 --> 00:10:42,280 Speaker 1: billion dollars every single year, and then some funneled into 200 00:10:42,280 --> 00:10:44,960 Speaker 1: AI compute just to produce the revenue to justify these 201 00:10:45,040 --> 00:10:48,800 Speaker 1: deals existence. In the space of a year, Microsoft Zure 202 00:10:48,840 --> 00:10:52,000 Speaker 1: made seventy five billion dollars, Google Cloud forty three billion dollars, 203 00:10:52,000 --> 00:10:56,040 Speaker 1: and Amazon Web Services one hundred billion dollars. Now that 204 00:10:56,280 --> 00:10:59,160 Speaker 1: is just add those numbers together, it's a little over 205 00:10:59,160 --> 00:11:04,319 Speaker 1: two hundred billion dollars, and that's the three largest providers 206 00:11:04,360 --> 00:11:06,640 Speaker 1: of any kind of compute, not just AI compute. Most 207 00:11:06,679 --> 00:11:09,160 Speaker 1: of that is not AI compute. In fact, they barely 208 00:11:09,160 --> 00:11:10,880 Speaker 1: make anything on that. They won't even talk about. The 209 00:11:10,920 --> 00:11:24,000 Speaker 1: AI revenue is probably because they're so stinky. Well, if 210 00:11:24,000 --> 00:11:26,400 Speaker 1: you need more proof, If you still don't believe me, 211 00:11:26,760 --> 00:11:29,560 Speaker 1: skip to page eighteen of Nvidia's most recent earnings and 212 00:11:29,600 --> 00:11:32,800 Speaker 1: I quote multi year cloud service agreement commitments as of 213 00:11:32,840 --> 00:11:35,640 Speaker 1: October twenty six, twenty twenty five, were twenty six billion dollars, 214 00:11:35,760 --> 00:11:38,520 Speaker 1: for which one billion dollars, six billion dollars, six billion dollars, 215 00:11:38,520 --> 00:11:41,079 Speaker 1: five billion dollars, four billion dollars, and four billion dollars 216 00:11:41,320 --> 00:11:43,920 Speaker 1: will we paid in fiscal years twenty twenty six, fourth quarter, 217 00:11:44,120 --> 00:11:47,040 Speaker 1: twenty twenty seven, twenty twenty eight, twenty twenty nine, twenty 218 00:11:47,080 --> 00:11:51,520 Speaker 1: thirty and twenty thirty one and thereafter respectively. Now I 219 00:11:51,559 --> 00:11:53,880 Speaker 1: know that was confusing, but just to line that up, 220 00:11:53,920 --> 00:11:57,559 Speaker 1: that means in fiscal year twenty twenty seven, which begins Oh, 221 00:11:57,600 --> 00:12:01,400 Speaker 1: I would have to look that up. Yeah, I'm literally 222 00:12:01,440 --> 00:12:03,720 Speaker 1: doing this on the air, and I'm gonna shit. So 223 00:12:03,760 --> 00:12:07,080 Speaker 1: their fiscal year begins in February, So basically twenty twenty 224 00:12:07,080 --> 00:12:11,120 Speaker 1: seven begins this February twenty twenty six. So they're going 225 00:12:11,160 --> 00:12:14,320 Speaker 1: to be spending six billion dollars on AI compute in 226 00:12:14,360 --> 00:12:16,320 Speaker 1: twenty twenty seven. This is a company that does not 227 00:12:16,440 --> 00:12:20,680 Speaker 1: rent out compute. If there's such incredible surging demand, why 228 00:12:20,720 --> 00:12:23,520 Speaker 1: is it? Why exactly is in Vidia spending six billion 229 00:12:23,559 --> 00:12:25,720 Speaker 1: fucking dollars a year in twenty twenty six and twenty 230 00:12:25,760 --> 00:12:28,400 Speaker 1: twenty seven on it. In Vidia doesn't need compute, it 231 00:12:28,520 --> 00:12:31,720 Speaker 1: just shut down. It's Amazon Web Services. Compared to those 232 00:12:31,720 --> 00:12:34,160 Speaker 1: it's called DJX cloud. It looks far more like in 233 00:12:34,240 --> 00:12:37,040 Speaker 1: video is propping up an industry with non existent demand. 234 00:12:37,920 --> 00:12:41,120 Speaker 1: I'm afraid there is no secret Amazon Web Services size 235 00:12:41,160 --> 00:12:43,480 Speaker 1: spend waiting in the wings for the right moment to pounce. 236 00:12:43,800 --> 00:12:46,360 Speaker 1: There is no secret demand wave, nor is there any 237 00:12:46,360 --> 00:12:50,480 Speaker 1: capacity crunch that is holding back incredible swarths of revenue oracles. 238 00:12:50,480 --> 00:12:52,880 Speaker 1: Five hundred and twenty three billion dollars in remaining performance 239 00:12:52,920 --> 00:12:56,720 Speaker 1: obligations are made up of OpenAI meta and you'll never 240 00:12:56,800 --> 00:13:03,560 Speaker 1: guess who in video fucking in VideA. The AI bubble 241 00:13:03,679 --> 00:13:06,240 Speaker 1: is just in Vidia being handed money and then handing 242 00:13:06,280 --> 00:13:08,560 Speaker 1: it to people. They're very profitable as well. They do 243 00:13:08,640 --> 00:13:12,960 Speaker 1: grame everyone else not so much. For AI data center 244 00:13:13,040 --> 00:13:15,280 Speaker 1: deals to make sense, most startups would have to start 245 00:13:15,320 --> 00:13:18,199 Speaker 1: becoming direct users of AI compute while also spending more 246 00:13:18,240 --> 00:13:22,079 Speaker 1: on cloud compute services than they've ever spent. The largest 247 00:13:22,080 --> 00:13:26,640 Speaker 1: consumers of AI compute are both unprofitable unsustainable monstrosities dependent 248 00:13:26,720 --> 00:13:30,880 Speaker 1: on venture capital. Eventually, reality will dawn on one or 249 00:13:30,920 --> 00:13:33,680 Speaker 1: more of these banks. Projects will get delayed thanks to 250 00:13:33,720 --> 00:13:36,959 Speaker 1: weather or budgetary issues, or when customers walk away, as 251 00:13:37,040 --> 00:13:40,720 Speaker 1: just happened to a data center ariit called FERMI, and 252 00:13:40,840 --> 00:13:44,679 Speaker 1: loan payments will start going unpaid elsewhere. AI startups will 253 00:13:44,679 --> 00:13:47,440 Speaker 1: start asking for money again and again, and for a 254 00:13:47,440 --> 00:13:49,960 Speaker 1: while they'll keep raising until the evaluations get too high, 255 00:13:50,040 --> 00:13:52,839 Speaker 1: or VC coffers get too low. You're probably going to 256 00:13:52,840 --> 00:13:54,880 Speaker 1: say at this point that Open AI or Anthropic might 257 00:13:54,920 --> 00:13:57,520 Speaker 1: go public, which will infuse capital into the system, and 258 00:13:57,559 --> 00:13:59,520 Speaker 1: I want to give you a preview of what that 259 00:13:59,640 --> 00:14:03,480 Speaker 1: might look like. Courtesy of AI Labs, Minimax, and Zipu, 260 00:14:03,520 --> 00:14:06,840 Speaker 1: as reported by the Information which just filed to go Bubblica. 261 00:14:07,760 --> 00:14:09,400 Speaker 1: I'm a little scared of These are the first real 262 00:14:09,480 --> 00:14:11,440 Speaker 1: numbers about the AI BUBBLA. Hope, I'm not going to 263 00:14:11,440 --> 00:14:14,960 Speaker 1: get embarrassed. Everybody's saying that behind the scenes these companies 264 00:14:14,960 --> 00:14:19,200 Speaker 1: are secretly great. Let's take a look. Oh my god, 265 00:14:19,480 --> 00:14:22,760 Speaker 1: these numbers aren't great at all. These numbers are terrible. 266 00:14:23,160 --> 00:14:25,480 Speaker 1: In the first half of this of twenty twenty five, 267 00:14:25,680 --> 00:14:27,760 Speaker 1: Zipu had a net loss of three hundred and thirty 268 00:14:27,760 --> 00:14:31,920 Speaker 1: four million dollars on twenty seven million dollars of revenue. Meanwhile, 269 00:14:31,960 --> 00:14:34,720 Speaker 1: Minimax made fifty three point four million dollars in revenue 270 00:14:34,720 --> 00:14:36,800 Speaker 1: in the first nine months of twenty twenty five and 271 00:14:36,840 --> 00:14:40,080 Speaker 1: bun two hundred and eleven million dollars to earn it. 272 00:14:40,080 --> 00:14:42,840 Speaker 1: It's time to wake up. These are the real life 273 00:14:42,880 --> 00:14:46,359 Speaker 1: costs of running an AI company, Open Ai and Anthropic 274 00:14:46,400 --> 00:14:49,560 Speaker 1: are going to be even worse than this. This is 275 00:14:49,600 --> 00:14:52,600 Speaker 1: why nobody wants to take AI company's public This is 276 00:14:52,640 --> 00:14:54,680 Speaker 1: why nobody wants to talk about the actual costs of 277 00:14:54,760 --> 00:14:58,000 Speaker 1: running AI. This is why nobody wants you to know 278 00:14:58,040 --> 00:15:00,680 Speaker 1: the hourly cost of running a GPU, And this is 279 00:15:00,720 --> 00:15:04,000 Speaker 1: why open Ai and Anthropic burn billions and billions of dollars. 280 00:15:04,240 --> 00:15:07,880 Speaker 1: The margins fucking stink. Every product is unprofitable and none 281 00:15:07,920 --> 00:15:10,520 Speaker 1: of these companies can afford their bills based on their 282 00:15:10,560 --> 00:15:14,360 Speaker 1: actual cash flow. Generative AI is not a functional industry, 283 00:15:14,400 --> 00:15:18,960 Speaker 1: and once the money works that out, everything burns. Though 284 00:15:19,000 --> 00:15:22,400 Speaker 1: many AI data centers boast of having tendency agreements, remember 285 00:15:22,560 --> 00:15:24,720 Speaker 1: these agreements are either with AI startups that will run 286 00:15:24,720 --> 00:15:27,360 Speaker 1: out of money who really shouldn't be renting GPUs by 287 00:15:27,400 --> 00:15:30,680 Speaker 1: the way, or hyperscalers with legal teams numbering in the thousands. 288 00:15:31,160 --> 00:15:35,080 Speaker 1: Every single deal that Microsoft, Amazon, Meta, Googler, and Video 289 00:15:35,200 --> 00:15:39,640 Speaker 1: signs is riddled withouts specifically hedging against this scenario, and 290 00:15:39,680 --> 00:15:41,920 Speaker 1: there won't be a damn thing that anybody can do 291 00:15:42,000 --> 00:15:45,400 Speaker 1: about it. If Hyperscalers decide to walk away before then, 292 00:15:45,560 --> 00:15:48,360 Speaker 1: in Video's bubble is likely to burst. As I discussed 293 00:15:48,360 --> 00:15:50,400 Speaker 1: a few weeks ago, and Vidia claims to have shipped 294 00:15:50,480 --> 00:15:52,960 Speaker 1: six million Blackwell GPUs, and while it may be employing 295 00:15:53,080 --> 00:15:55,960 Speaker 1: very dodgy maths claiming that each GPU is actually to 296 00:15:56,360 --> 00:16:01,920 Speaker 1: chips Jensen Jensen, I've been saying. My modeling of its 297 00:16:02,040 --> 00:16:04,640 Speaker 1: last three quarters suggests that in videos shipped around five 298 00:16:04,680 --> 00:16:07,640 Speaker 1: point three to three gigawatts worth of GPUs, and based 299 00:16:07,680 --> 00:16:10,520 Speaker 1: on reading about every single data center deal I can find, 300 00:16:10,800 --> 00:16:13,480 Speaker 1: it doesn't appear that many have been built and powered on, 301 00:16:14,400 --> 00:16:18,640 Speaker 1: or even started building in many cases. Worse still, in 302 00:16:18,720 --> 00:16:22,040 Speaker 1: Video's diversified revenue is collapsing. In the first quarter of 303 00:16:22,040 --> 00:16:24,960 Speaker 1: its fiscal year twenty twenty six, I'm sorry, you're just 304 00:16:25,000 --> 00:16:27,040 Speaker 1: going to have to ride with me on this one, 305 00:16:27,160 --> 00:16:30,400 Speaker 1: two customers represented sixteen percent and fourteen percent in revenue. 306 00:16:30,760 --> 00:16:32,880 Speaker 1: In the second quarter of fiscal year twenty six, two 307 00:16:32,920 --> 00:16:36,240 Speaker 1: customers represented twenty three percent and sixteen percent, and in 308 00:16:36,280 --> 00:16:40,160 Speaker 1: the third quarter, four customers represented twenty two percent, fifteen percent, 309 00:16:40,240 --> 00:16:43,200 Speaker 1: thirteen percent, and eleven percent of total revenue, with all 310 00:16:43,280 --> 00:16:46,080 Speaker 1: of that money going towards GPUs or networking gear. In 311 00:16:46,120 --> 00:16:48,600 Speaker 1: simpler terms, in Vidia's revenue is no longer coming from 312 00:16:48,640 --> 00:16:51,760 Speaker 1: a diverse swarth of customers. In its first quarter fiscal 313 00:16:51,840 --> 00:16:54,000 Speaker 1: year twenty twenty six, it had thirty point eight four 314 00:16:54,040 --> 00:16:57,200 Speaker 1: billion dollars of diversified revenue. That's from a customer that 315 00:16:57,520 --> 00:16:59,680 Speaker 1: from customers that do not number more than ten percent 316 00:16:59,680 --> 00:17:02,640 Speaker 1: of its revenue, and in the second quarter of FY 317 00:17:02,720 --> 00:17:05,640 Speaker 1: twenty six that was twenty eight point five billion Q three. 318 00:17:05,640 --> 00:17:09,600 Speaker 1: It's twenty two point twenty three billion. In simpler terms, 319 00:17:10,320 --> 00:17:13,280 Speaker 1: there are only a few companies that can buy GPUs, 320 00:17:13,359 --> 00:17:16,080 Speaker 1: and other companies are no longer buying GPUs at the 321 00:17:16,080 --> 00:17:18,600 Speaker 1: same rate. I doubt this number is going to increase. 322 00:17:19,640 --> 00:17:23,160 Speaker 1: You see in video, GPUs are astronomically expensive four point 323 00:17:23,200 --> 00:17:25,320 Speaker 1: five million dollars for a GB three hundred radc of 324 00:17:25,600 --> 00:17:28,520 Speaker 1: seventy two B three hundred GPUs, for example, and feeling 325 00:17:28,600 --> 00:17:31,639 Speaker 1: data centers full of them requires debt unless you're a hyperscaler. Well, 326 00:17:31,680 --> 00:17:34,320 Speaker 1: I can't say for sure. I believe in videos, diversified 327 00:17:34,320 --> 00:17:36,959 Speaker 1: revenue collapses a sign that smaller data center projects are 328 00:17:36,960 --> 00:17:40,399 Speaker 1: starting to have issues getting funded and or hyperscalers are 329 00:17:40,400 --> 00:17:43,320 Speaker 1: pulling back on the GPU purchases from those Taiwanese companies. 330 00:17:43,359 --> 00:17:46,640 Speaker 1: I was mentioning, Now, let me look through the eyes 331 00:17:46,680 --> 00:17:49,720 Speaker 1: of an AI boosted for a second. Okay, okay, everything 332 00:17:49,800 --> 00:17:52,440 Speaker 1: is blue and yellow as usual. Okay. One might say 333 00:17:52,480 --> 00:17:55,080 Speaker 1: that these big customers are covering the loss of revenue, 334 00:17:55,160 --> 00:17:57,280 Speaker 1: but the reality is that these big projects are run 335 00:17:57,320 --> 00:17:59,920 Speaker 1: on debt issued by banks are becoming increasingly worried about 336 00:18:00,080 --> 00:18:04,640 Speaker 1: nobody paying them back. The mistake that every investor, commentator, analyst, 337 00:18:04,680 --> 00:18:06,879 Speaker 1: and member of the media makes about Nvidia is believing 338 00:18:06,880 --> 00:18:09,520 Speaker 1: that its sales are an expression of demand for AI compume, 339 00:18:09,880 --> 00:18:12,159 Speaker 1: when in reality it's more of a statement about the 340 00:18:12,200 --> 00:18:16,000 Speaker 1: availability of debt from banks in private credit. Similarly, the 341 00:18:16,000 --> 00:18:18,720 Speaker 1: continued existence of AI startups is an expression of the 342 00:18:18,760 --> 00:18:21,960 Speaker 1: desperation of venture capital and the continuing flow of massive 343 00:18:22,000 --> 00:18:24,360 Speaker 1: funding rounds as assigned that they see no other revenues 344 00:18:24,400 --> 00:18:28,480 Speaker 1: for growth and that their startups can't wipe their own assholes. Eventually, 345 00:18:28,600 --> 00:18:30,760 Speaker 1: data centers are going to go unbuilt and data center 346 00:18:30,840 --> 00:18:34,240 Speaker 1: debt packages will begin to fall apart. Remember, Oracle's thirty 347 00:18:34,240 --> 00:18:36,680 Speaker 1: eight billion dollar data center deal is actually yet to close, 348 00:18:37,040 --> 00:18:39,960 Speaker 1: much like Stargate New Mexico is yet to close, and 349 00:18:40,040 --> 00:18:43,040 Speaker 1: much like Stargate Michigan is yet to close. These deals 350 00:18:43,080 --> 00:18:46,640 Speaker 1: while seeming like they're trending positively, are all incredibly important 351 00:18:46,640 --> 00:18:49,040 Speaker 1: to the future of the AI bubble, and any failure 352 00:18:49,160 --> 00:18:53,119 Speaker 1: will spooken already notice mark them only one link in 353 00:18:53,200 --> 00:18:57,200 Speaker 1: this chain of pain needs to break every single part 354 00:18:57,320 --> 00:19:01,400 Speaker 1: of the AI bubble. This fucking charade is unprofitable save 355 00:19:01,480 --> 00:19:04,120 Speaker 1: for in Video and the construction firms erecting future laser 356 00:19:04,200 --> 00:19:08,320 Speaker 1: tag arenas full of negative margin GPUs tell me what 357 00:19:08,520 --> 00:19:11,560 Speaker 1: happens if the debt stops flowing data centers? How will 358 00:19:11,600 --> 00:19:14,840 Speaker 1: in Vidia sell their so called twenty million Blackwell and 359 00:19:14,960 --> 00:19:17,919 Speaker 1: veried Ruben GPUs their claiming they'll ship by the end 360 00:19:17,960 --> 00:19:21,360 Speaker 1: of twenty twenty six. What happens if venture capitalists start 361 00:19:21,440 --> 00:19:24,320 Speaker 1: running low on funds and can't keep feeding hundreds of 362 00:19:24,359 --> 00:19:26,879 Speaker 1: millions of dollars to AI startups so that they can 363 00:19:27,000 --> 00:19:30,439 Speaker 1: feed them directly to open Ai or Anthropic. What happens 364 00:19:30,480 --> 00:19:33,639 Speaker 1: to open Ai and Anthropic and they're already negative margin 365 00:19:33,720 --> 00:19:37,280 Speaker 1: businesses when their customers can't pay them. What happens to 366 00:19:37,359 --> 00:19:40,080 Speaker 1: Oracle or core weaves, work in progress data centers if 367 00:19:40,119 --> 00:19:43,359 Speaker 1: open ai can't pay their bills? What happens to Anthropics? 368 00:19:43,400 --> 00:19:46,240 Speaker 1: Twenty one billion dollars of Broadcom orders or tens of 369 00:19:46,280 --> 00:19:49,959 Speaker 1: billions of dollars of Google cloud spend. And what if 370 00:19:50,000 --> 00:19:53,119 Speaker 1: I'm right about everything I've said, not just in this series, 371 00:19:53,359 --> 00:19:57,159 Speaker 1: but in the episodes and newsletters before him. In the 372 00:19:57,280 --> 00:20:00,040 Speaker 1: last year, I estimate I've been asked the question what 373 00:20:00,160 --> 00:20:03,440 Speaker 1: if You're wrong? Over twenty five times. Every single time 374 00:20:03,520 --> 00:20:06,680 Speaker 1: the question comes with this undercurrent venom, the suggestion that 375 00:20:06,760 --> 00:20:09,280 Speaker 1: I'm being an asshole for daring to question the wondrous 376 00:20:09,320 --> 00:20:12,919 Speaker 1: AI bubble. Every single person who's asked me this has 377 00:20:12,960 --> 00:20:14,960 Speaker 1: been poorly read, both in terms of my work and 378 00:20:15,000 --> 00:20:18,479 Speaker 1: the surrounding economics and technological possibilities of large language models 379 00:20:18,640 --> 00:20:21,639 Speaker 1: and believes they're defending technology, when in reality they're defending 380 00:20:21,720 --> 00:20:24,240 Speaker 1: growth in the rock economy's growth at all cost mindset. 381 00:20:25,200 --> 00:20:28,280 Speaker 1: In many cases, they are not excited about technology at all, 382 00:20:28,520 --> 00:20:30,480 Speaker 1: but the prospects of being the first in line to 383 00:20:30,560 --> 00:20:34,240 Speaker 1: lick an already sparkling boot. This has never ever been 384 00:20:34,280 --> 00:20:37,840 Speaker 1: about progress or productivity. If it was, we'd seen actual 385 00:20:37,880 --> 00:20:42,560 Speaker 1: progress or productivity boosts. So anything other than the frothiest debt, 386 00:20:42,960 --> 00:20:46,240 Speaker 1: the frothiest venture capital markets, and the most annoying and 387 00:20:46,400 --> 00:20:51,159 Speaker 1: incorrect headlines I've ever read in the news. Large language 388 00:20:51,200 --> 00:20:54,919 Speaker 1: models do not create novel concepts. They are inconsistent and unreliable, 389 00:20:55,080 --> 00:20:57,320 Speaker 1: and even the things that people like them for, like 390 00:20:57,480 --> 00:21:00,320 Speaker 1: coding very wildly thanks to the dramatic very tents of 391 00:21:00,359 --> 00:21:05,040 Speaker 1: a giant probability machine l alams, are not good enough 392 00:21:05,080 --> 00:21:07,919 Speaker 1: for people to pay regular software prices at any scale, 393 00:21:08,119 --> 00:21:10,359 Speaker 1: and the consequences of this will be that every single 394 00:21:10,440 --> 00:21:13,440 Speaker 1: dollar spent on GPUs has been for exactly one point 395 00:21:14,280 --> 00:21:19,800 Speaker 1: manipulating the value of stocks by making hyperscalers look good 396 00:21:20,080 --> 00:21:22,800 Speaker 1: so that they can keep pretending that they create anything. 397 00:21:24,359 --> 00:21:27,040 Speaker 1: AI does not have the business returns and may indeed 398 00:21:27,119 --> 00:21:32,760 Speaker 1: have negative gross margins across the board. It is inconsistent, ugly, unreliable, 399 00:21:33,040 --> 00:21:37,639 Speaker 1: expensive and environmentally ruinous, pissing off a large chunk of consumers, 400 00:21:37,880 --> 00:21:41,000 Speaker 1: and underwhelming most of the rest, other than those convinced 401 00:21:41,000 --> 00:21:43,240 Speaker 1: they're smart for using it, or those who have resigned 402 00:21:43,240 --> 00:21:45,160 Speaker 1: to giving up at the site of a confidence game 403 00:21:45,280 --> 00:21:48,200 Speaker 1: sold by a tech industry that stopped making products primarily 404 00:21:48,240 --> 00:21:51,240 Speaker 1: focused on solving the problems of consumers or businesses a 405 00:21:51,359 --> 00:21:55,760 Speaker 1: long time ago. You may say I'm wrong because Google, Microsoft, 406 00:21:55,840 --> 00:21:58,879 Speaker 1: Meta and Amazon continue to have healthy net revenues and 407 00:21:58,960 --> 00:22:02,159 Speaker 1: revenue growth. And as I've previously said, these companies are 408 00:22:02,200 --> 00:22:05,280 Speaker 1: not sharing AI revenues, and their existing businesses are still 409 00:22:05,400 --> 00:22:08,800 Speaker 1: growing due to the massive monopolies they've built. And I 410 00:22:08,960 --> 00:22:12,040 Speaker 1: want to plead to the AI boosters and bullish analysts alike, 411 00:22:12,520 --> 00:22:18,720 Speaker 1: you are being had. Sacha Adela, Sam Altman, Dario Amade, 412 00:22:18,880 --> 00:22:23,680 Speaker 1: Jensen Huang, Mark Zuckerberg, Larry Ellison, saffracats, Elon Musk, Klay mcguk, 413 00:22:24,119 --> 00:22:28,200 Speaker 1: Mike Cecilia, Mike truell Aravincerivenus, all of them are laughing 414 00:22:28,240 --> 00:22:30,160 Speaker 1: at you behind your back because they know that you're 415 00:22:30,240 --> 00:22:32,520 Speaker 1: never going to ask the obvious questions that would defeat 416 00:22:32,560 --> 00:22:34,840 Speaker 1: my arguments, and know that you will never ever push 417 00:22:34,920 --> 00:22:38,200 Speaker 1: back on them. They know the truth, they just don't 418 00:22:38,240 --> 00:22:41,760 Speaker 1: want to tell you because you don't care to argue. 419 00:22:58,480 --> 00:23:02,200 Speaker 1: The inceritification of the share has the downstream effect of 420 00:23:02,240 --> 00:23:05,320 Speaker 1: an intitification of the media and the Wall Street analysts 421 00:23:05,359 --> 00:23:09,959 Speaker 1: writ large analysts media. These companies own you, they own 422 00:23:10,000 --> 00:23:13,080 Speaker 1: your asses. They treat you with disdain and condescension because 423 00:23:13,119 --> 00:23:16,120 Speaker 1: they know you'll let them. They know that no Cell 424 00:23:16,240 --> 00:23:19,240 Speaker 1: side analysts will ever ask them when will you be profitable? 425 00:23:19,359 --> 00:23:21,640 Speaker 1: Or how much you spending? Or if you do ask, 426 00:23:21,720 --> 00:23:24,760 Speaker 1: they know you will experience temporary amnesia and forget whatever 427 00:23:24,880 --> 00:23:27,879 Speaker 1: answer they give, because these are the incentives of an 428 00:23:27,960 --> 00:23:31,879 Speaker 1: inshitified stock market, where stocks are not extrapolations of shareholder 429 00:23:32,000 --> 00:23:34,600 Speaker 1: value but chips in a fucking casino where the house 430 00:23:34,640 --> 00:23:39,000 Speaker 1: always wins and changes the rules every three months. These 431 00:23:39,080 --> 00:23:41,480 Speaker 1: companies have changed the meaning of the word stock to 432 00:23:41,560 --> 00:23:45,000 Speaker 1: mean whatever the market will reward, and when you allow 433 00:23:45,160 --> 00:23:48,240 Speaker 1: companies to start dictating the terms of what will be rewarded. 434 00:23:48,520 --> 00:23:53,840 Speaker 1: As neoliberalism, Freedman, Reagan, Nixon, NAFTA, and every other fucked 435 00:23:53,920 --> 00:23:58,800 Speaker 1: up growth focused policy has orienting everything exclusively around growth, 436 00:23:59,119 --> 00:24:02,120 Speaker 1: companies eventually cut off any powers that may curtail any 437 00:24:02,160 --> 00:24:05,639 Speaker 1: reevaluation of the fundamental terms of capitalism and the incentives 438 00:24:05,720 --> 00:24:10,760 Speaker 1: with him focusing on growth at all costs thinking naturally encourages, enables, 439 00:24:10,800 --> 00:24:13,840 Speaker 1: and empowers grifters, because all they ever have to promise 440 00:24:13,960 --> 00:24:17,000 Speaker 1: is more more users, more debt, more venture, more features, 441 00:24:17,359 --> 00:24:22,800 Speaker 1: more everything forever. From Adam Becker by it Today, the 442 00:24:22,960 --> 00:24:25,920 Speaker 1: very institutions that are meant to hold companies accountable. Analysts 443 00:24:25,960 --> 00:24:28,560 Speaker 1: and the media are far more desperate to trade scoops 444 00:24:28,600 --> 00:24:30,960 Speaker 1: for interviews, to pull punches, to find ways to explain 445 00:24:31,000 --> 00:24:33,399 Speaker 1: why a company is right, rather than understand what the 446 00:24:33,480 --> 00:24:35,840 Speaker 1: company is doing. And this is something pushed not by 447 00:24:35,880 --> 00:24:37,959 Speaker 1: writers but by editors that want to make sure they 448 00:24:38,000 --> 00:24:40,920 Speaker 1: stay on the right side of the largest companies. And goddamn, 449 00:24:40,960 --> 00:24:43,160 Speaker 1: do I know a few stories there that one day 450 00:24:43,200 --> 00:24:46,679 Speaker 1: I might even tell. And if I'm right, open AI's 451 00:24:46,720 --> 00:24:49,600 Speaker 1: death will kill off most, if not all, AI startups 452 00:24:49,680 --> 00:24:53,640 Speaker 1: Anthropic included. Every investor that invested in AI will take 453 00:24:53,720 --> 00:24:56,520 Speaker 1: massive losses. Every startup that builds on the back of 454 00:24:56,560 --> 00:24:58,760 Speaker 1: their models will see their companies fold if it hasn't 455 00:24:58,800 --> 00:25:02,040 Speaker 1: already done so to the massive costs an upcoming price 456 00:25:02,119 --> 00:25:05,240 Speaker 1: increase we've already seen signs of with priority processing with 457 00:25:05,280 --> 00:25:07,640 Speaker 1: both Anthropic and open Ai in the middle of last year. 458 00:25:08,280 --> 00:25:11,000 Speaker 1: The majority of GPU based data centers, which really have 459 00:25:11,119 --> 00:25:14,280 Speaker 1: no other revenue stream, will be left inert, likely powered down, 460 00:25:14,320 --> 00:25:16,200 Speaker 1: waiting for the day that someone works it all out, 461 00:25:16,240 --> 00:25:18,840 Speaker 1: which they won't because literally everybody has these things now, 462 00:25:18,920 --> 00:25:22,000 Speaker 1: and I truly believe they've tried everything. And I'm gonna 463 00:25:22,040 --> 00:25:24,920 Speaker 1: hear from someone that says us, just like the Duck combo, 464 00:25:25,040 --> 00:25:27,560 Speaker 1: they're fiber optic cable. We had so much fiber optic 465 00:25:27,600 --> 00:25:29,800 Speaker 1: cable and then we put internet in them. This is 466 00:25:29,880 --> 00:25:32,879 Speaker 1: not GPUs are not fiber optic cable. They're not I 467 00:25:33,000 --> 00:25:34,879 Speaker 1: just did a premium piece on this. I'm probably going 468 00:25:34,960 --> 00:25:36,880 Speaker 1: to have to do an episode on it. But they 469 00:25:36,920 --> 00:25:39,320 Speaker 1: are not the same fucking thing. And people that keep 470 00:25:39,359 --> 00:25:41,840 Speaker 1: saying this are flippant about the damage that's going to 471 00:25:41,880 --> 00:25:44,320 Speaker 1: be caused and just flat out wrong. I really do 472 00:25:44,600 --> 00:25:47,840 Speaker 1: challenge you to actually read history before you make any 473 00:25:48,200 --> 00:25:50,680 Speaker 1: fucking statements like that. This is not for the people 474 00:25:50,720 --> 00:25:52,600 Speaker 1: who are saying this just because they haven't looked and 475 00:25:52,680 --> 00:25:56,119 Speaker 1: they just assumed that the world wouldn't be stupid. I 476 00:25:56,200 --> 00:25:58,640 Speaker 1: don't think many people saying that or off that camp. 477 00:25:58,720 --> 00:26:01,080 Speaker 1: I think they are people that come up with comfortable 478 00:26:02,000 --> 00:26:06,520 Speaker 1: reasons to validate bad behavior. And I must be clear, 479 00:26:06,920 --> 00:26:11,920 Speaker 1: I'm watching I'm watching everyone. I'm watching how everyone handles this, 480 00:26:12,160 --> 00:26:13,840 Speaker 1: and I'm going to be watching extra hard on the 481 00:26:13,880 --> 00:26:16,920 Speaker 1: way down because anybody who has been a booster that 482 00:26:17,080 --> 00:26:20,199 Speaker 1: tries to change their target and tries to change their 483 00:26:20,280 --> 00:26:26,000 Speaker 1: direction without a direct acceptance and real contrition about their role. 484 00:26:27,160 --> 00:26:32,679 Speaker 1: I have taken such detailed notes, but I will add something. 485 00:26:33,480 --> 00:26:35,960 Speaker 1: I don't hate on AI because I'm a hater. I 486 00:26:36,040 --> 00:26:38,560 Speaker 1: hate on it because it fucking sucks. And what I'm 487 00:26:38,600 --> 00:26:41,840 Speaker 1: worried about happening seems to be happening. The tech industry 488 00:26:41,880 --> 00:26:44,640 Speaker 1: has run out of hypergrowth ideas, and in its desperation, 489 00:26:44,800 --> 00:26:47,800 Speaker 1: hitched itself to the least profitable hardware and software in history. 490 00:26:48,040 --> 00:26:51,160 Speaker 1: Then spent three straight years lying about what was possible 491 00:26:51,200 --> 00:26:54,600 Speaker 1: to the media, alice, and shareholders. And they were allowed 492 00:26:54,640 --> 00:26:57,560 Speaker 1: to lie because everybody lapped it the fuck up. They 493 00:26:57,600 --> 00:27:01,359 Speaker 1: didn't need to worry about convincing Anybodyncers, editors, analysts, and 494 00:27:01,440 --> 00:27:04,080 Speaker 1: investors were already drafting reasons why they were so excited 495 00:27:04,080 --> 00:27:06,880 Speaker 1: about something they didn't really understand or believe in, other 496 00:27:07,000 --> 00:27:10,919 Speaker 1: than the fact it promised more. This is what happens 497 00:27:11,000 --> 00:27:15,200 Speaker 1: when you make everything about growth. Everybody becomes stupid, ready 498 00:27:15,240 --> 00:27:17,120 Speaker 1: to be conned, ready to hear what the next big 499 00:27:17,200 --> 00:27:20,919 Speaker 1: growth thing is because asking nasty questions gets your fucking fired. 500 00:27:22,200 --> 00:27:24,560 Speaker 1: And what's left is a tech industry that doesn't build 501 00:27:24,600 --> 00:27:28,600 Speaker 1: technology but growth focused startups. Look at silicon value. Do 502 00:27:28,640 --> 00:27:30,680 Speaker 1: you see these fucking people ever building a new kind 503 00:27:30,680 --> 00:27:32,920 Speaker 1: of computer? Do you believe these men are fit to 504 00:27:33,000 --> 00:27:36,280 Speaker 1: even imagine a future? These men care about the status quo. 505 00:27:36,560 --> 00:27:38,840 Speaker 1: They want to always have more software to sell in 506 00:27:39,000 --> 00:27:41,359 Speaker 1: what or ways to increase advertising revenue so that the 507 00:27:41,400 --> 00:27:43,480 Speaker 1: stock number goes up, so they receive more money in 508 00:27:43,520 --> 00:27:46,440 Speaker 1: the form of stock compensation. They are concerned with neither 509 00:27:46,520 --> 00:27:50,920 Speaker 1: actual business value on this exchange of value or societal value. 510 00:27:51,119 --> 00:27:54,000 Speaker 1: Their existence is only in shareholder value, which is how 511 00:27:54,040 --> 00:27:56,720 Speaker 1: they are incentivized by their board of directors. And right 512 00:27:56,840 --> 00:28:02,480 Speaker 1: now they're in shitifying being a shareholder. But really, if 513 00:28:02,520 --> 00:28:06,639 Speaker 1: you're still defending AI, does it matter to any of 514 00:28:06,720 --> 00:28:10,960 Speaker 1: you that this software fucking sucks? If you think it's good, 515 00:28:11,000 --> 00:28:14,360 Speaker 1: you don't know much about software. Sure, your coding tool 516 00:28:14,480 --> 00:28:16,720 Speaker 1: is useful in whatever way it can fart out and 517 00:28:16,880 --> 00:28:19,960 Speaker 1: language simula or whatever. Oh, it can do some of 518 00:28:20,000 --> 00:28:24,119 Speaker 1: the easy coding things. I'm sorry, I'm sorry, man, I 519 00:28:24,240 --> 00:28:26,760 Speaker 1: can't get excited about that. We're hundreds of billions of 520 00:28:26,840 --> 00:28:29,440 Speaker 1: dollars in the hole. I need something more than that. 521 00:28:29,960 --> 00:28:33,560 Speaker 1: I need so much more than that. And if I'm honest, 522 00:28:33,640 --> 00:28:37,080 Speaker 1: the problem with large language models, it's not good software. 523 00:28:37,240 --> 00:28:39,840 Speaker 1: It doesn't respond precisely at any point to a user 524 00:28:39,920 --> 00:28:43,080 Speaker 1: or a programmers intent. That's bad software. And I don't 525 00:28:43,160 --> 00:28:45,080 Speaker 1: care that you've heard developers really like it, because that 526 00:28:45,160 --> 00:28:48,080 Speaker 1: doesn't fix the underlying economic and social poison in AI. 527 00:28:48,680 --> 00:28:50,840 Speaker 1: I don't care that it's sort of replaced search for you. 528 00:28:51,000 --> 00:28:52,840 Speaker 1: I don't care if you know a team of engineers 529 00:28:52,880 --> 00:28:56,000 Speaker 1: that use it. Every single AI app is subsidized. Its 530 00:28:56,040 --> 00:28:58,000 Speaker 1: price is fake. You are being laid to, and none 531 00:28:58,040 --> 00:29:02,400 Speaker 1: of this is real. I also, and this is a 532 00:29:02,520 --> 00:29:04,520 Speaker 1: side a little bit of a side quest, I think 533 00:29:04,560 --> 00:29:06,960 Speaker 1: AI psychosis is so much bigger than people think. I 534 00:29:07,000 --> 00:29:09,880 Speaker 1: feel like I might have mentioned this on the recent monologue, 535 00:29:09,880 --> 00:29:12,959 Speaker 1: but it's just it's really bothering me. I think there 536 00:29:13,040 --> 00:29:15,920 Speaker 1: is something that happens when you play with these things 537 00:29:16,360 --> 00:29:18,840 Speaker 1: and you make them work, versus just being like, oh, 538 00:29:18,880 --> 00:29:21,800 Speaker 1: they don't work. I think what happens is it convinces 539 00:29:21,880 --> 00:29:24,520 Speaker 1: you you're smart, when what you really are is being 540 00:29:24,920 --> 00:29:27,680 Speaker 1: a slave to the software yourself. You are being tricked. 541 00:29:27,920 --> 00:29:30,000 Speaker 1: You are being convinced that the software is doing something 542 00:29:30,000 --> 00:29:32,000 Speaker 1: because you are bonking it on the head until it 543 00:29:32,080 --> 00:29:36,240 Speaker 1: does something you need it to. Really, none of this 544 00:29:36,440 --> 00:29:40,240 Speaker 1: is real, none of it. I'm so scared. I'm so 545 00:29:40,360 --> 00:29:44,440 Speaker 1: scared because on one hand, I kind of look like 546 00:29:44,480 --> 00:29:46,240 Speaker 1: I'm about to be proven right in a way that's 547 00:29:46,240 --> 00:29:50,080 Speaker 1: going to be good for business, I guess. But being 548 00:29:50,240 --> 00:29:55,080 Speaker 1: right here means watching something really calamitous and when the 549 00:29:55,160 --> 00:29:58,400 Speaker 1: collapse happens. Do not let a single person that waved 550 00:29:58,400 --> 00:30:01,280 Speaker 1: off the economics have a moment's piece. Do not let 551 00:30:01,400 --> 00:30:03,960 Speaker 1: anybody who sat in front of Dario Amadey or Sam 552 00:30:04,080 --> 00:30:06,040 Speaker 1: Ortman such as I don't know on a New York 553 00:30:06,120 --> 00:30:10,280 Speaker 1: Times podcast and squealed with delight at whatever vacuous talking 554 00:30:10,360 --> 00:30:12,719 Speaker 1: points they burped out. Forget that they didn't push them, 555 00:30:12,800 --> 00:30:15,240 Speaker 1: They didn't ask the hard questions. They didn't worry or 556 00:30:15,320 --> 00:30:18,520 Speaker 1: wonder or feel any concerns for investors of the general public. 557 00:30:18,680 --> 00:30:21,040 Speaker 1: They only cared about getting rock hard listening to whatever 558 00:30:21,240 --> 00:30:24,880 Speaker 1: bullshit the latest large Language model asshole has to say. 559 00:30:25,600 --> 00:30:29,000 Speaker 1: Do not let a single analyst that called AI skeptics, loodites, 560 00:30:29,120 --> 00:30:33,000 Speaker 1: Daniel Newman futuream group, or equate them to flat earthers 561 00:30:33,240 --> 00:30:35,760 Speaker 1: hear the end of it. Do not let anybody who 562 00:30:35,840 --> 00:30:38,120 Speaker 1: claimed that we lost control of AI or that it 563 00:30:38,240 --> 00:30:42,080 Speaker 1: blackmailed developers go without their complimentary fell for it again, badge, 564 00:30:42,240 --> 00:30:46,400 Speaker 1: because no, these systems didn't blackmail anyone, they were literally 565 00:30:46,560 --> 00:30:49,200 Speaker 1: being prompted to do it. You fell for it, and 566 00:30:49,360 --> 00:30:51,959 Speaker 1: you will You'll No one is ever going to hear 567 00:30:52,120 --> 00:30:54,520 Speaker 1: the end of this from me, And when it happens, 568 00:30:54,600 --> 00:30:57,480 Speaker 1: I promise I won't be too insufferable, but I will 569 00:30:57,560 --> 00:31:01,320 Speaker 1: be calling for accountability for anybody who boosted AI twenty 570 00:31:01,400 --> 00:31:03,760 Speaker 1: twenty seven, who sat in front of Sam Ortman of 571 00:31:03,880 --> 00:31:07,040 Speaker 1: Dariyama Day and once again refused to ask real questions 572 00:31:07,160 --> 00:31:11,400 Speaker 1: or allowed bullshit answers, And for anyone anyone who collected 573 00:31:11,480 --> 00:31:15,240 Speaker 1: anything resembling detailed notes about me or any other AI 574 00:31:15,360 --> 00:31:19,840 Speaker 1: skeptic threatening independent reporters because I didn't clap for Daddy's 575 00:31:19,960 --> 00:31:25,160 Speaker 1: venture capital. Bullshit is insufferable, poisonous, and will never ever 576 00:31:25,240 --> 00:31:28,320 Speaker 1: be forgotten. And if you think I'm talking about you, 577 00:31:28,480 --> 00:31:31,760 Speaker 1: I probably am. And I have a question. Why didn't 578 00:31:31,800 --> 00:31:34,640 Speaker 1: you approach the AI companies with as much skepticism as 579 00:31:34,680 --> 00:31:39,080 Speaker 1: you did the skeptics. Why were you so concerned about 580 00:31:39,680 --> 00:31:43,160 Speaker 1: minor details when you could have cared about the major 581 00:31:43,240 --> 00:31:46,840 Speaker 1: details that will very likely fuck our markets up lead 582 00:31:46,880 --> 00:31:51,320 Speaker 1: to an apocalyptic state of venture capital and startups, and genuinely, 583 00:31:51,440 --> 00:31:53,280 Speaker 1: I don't even know if a recession will be it. 584 00:31:54,840 --> 00:31:59,000 Speaker 1: I also genuinely promise you if I'm wrong, I'll happily 585 00:31:59,080 --> 00:32:01,680 Speaker 1: explain how and why, and I'll do so at length 586 00:32:01,760 --> 00:32:05,760 Speaker 1: to I'll have links, I'll have citations, I'll do podcast episodes. 587 00:32:06,120 --> 00:32:08,920 Speaker 1: I will make a good faith effort to explain every 588 00:32:09,160 --> 00:32:12,000 Speaker 1: single failing because my concern is the truth and I 589 00:32:12,040 --> 00:32:15,400 Speaker 1: would love everybody else to follow suit. Do you think 590 00:32:15,440 --> 00:32:18,120 Speaker 1: any AI booster will have the same courtesy? Do you 591 00:32:18,160 --> 00:32:20,320 Speaker 1: think they care about the truth or do they want 592 00:32:20,360 --> 00:32:23,600 Speaker 1: their fish biscuit from Sam Ormon or Jensen Joan. It's 593 00:32:23,640 --> 00:32:27,320 Speaker 1: absolutely pathetic, and as we enter into twenty twenty six, 594 00:32:27,640 --> 00:32:30,480 Speaker 1: I'm going to fucking floor it with this stuff. I've 595 00:32:30,560 --> 00:32:33,000 Speaker 1: taught myself a great deal about numbers in the last year. 596 00:32:33,600 --> 00:32:36,400 Speaker 1: I've really got a good head for reading these ten cues, 597 00:32:36,720 --> 00:32:39,080 Speaker 1: and there's already stuff I'm seeing that is spelling the 598 00:32:39,200 --> 00:32:41,680 Speaker 1: end that I'm yet to talk about. This has been 599 00:32:41,760 --> 00:32:44,400 Speaker 1: a long episode, and it's been a long four parter. 600 00:32:45,320 --> 00:32:48,000 Speaker 1: I really hope you've enjoyed it, because I've loved reading it. 601 00:32:49,200 --> 00:32:51,640 Speaker 1: More to come next week I'll be joined by mister 602 00:32:51,720 --> 00:32:55,120 Speaker 1: Matt Rosoff, wonderful editor in chief over at the Register. 603 00:32:55,200 --> 00:32:57,960 Speaker 1: We're going to talk about the dot com bubble. Just folks, 604 00:32:58,280 --> 00:33:00,640 Speaker 1: hit me up on the round app, slub me on Goop. 605 00:33:09,000 --> 00:33:11,360 Speaker 1: Thank you for listening to Better Offline. The editor and 606 00:33:11,440 --> 00:33:14,560 Speaker 1: composer of the Better Offline theme song is Matasowski. You 607 00:33:14,640 --> 00:33:16,800 Speaker 1: can check out more of his music and audio projects 608 00:33:17,040 --> 00:33:20,480 Speaker 1: at Matasowski dot com, M A T T O S 609 00:33:20,600 --> 00:33:24,640 Speaker 1: O W s ki dot com. You can email me 610 00:33:24,720 --> 00:33:27,280 Speaker 1: at easy at Better Offline dot com or visit Better 611 00:33:27,320 --> 00:33:29,480 Speaker 1: Offline dot com to find more podcast links and of 612 00:33:29,560 --> 00:33:32,640 Speaker 1: course my newsletter. I also really recommend you go to 613 00:33:32,760 --> 00:33:35,200 Speaker 1: chat dot Where's Youreed dot at to visit the discord, 614 00:33:35,480 --> 00:33:37,760 Speaker 1: and go to our slash Better Offline to check out 615 00:33:37,800 --> 00:33:42,160 Speaker 1: our reddit. Thank you so much for listening. Better Offline 616 00:33:42,280 --> 00:33:44,440 Speaker 1: is a production of cool Zone Media. For more from 617 00:33:44,480 --> 00:33:47,760 Speaker 1: cool Zone Media, visit our website cool Zonemedia dot com, 618 00:33:48,280 --> 00:33:51,080 Speaker 1: or check us out on the iHeartRadio app, Apple Podcasts, 619 00:33:51,240 --> 00:33:52,640 Speaker 1: or wherever you get your podcasts.