1 00:00:02,480 --> 00:00:07,400 Speaker 1: Alz Media. Hello and welcome to a special exclusive episode 2 00:00:07,400 --> 00:00:21,640 Speaker 1: of Better Offline. I'm Your host ed Zitron. As a 3 00:00:21,720 --> 00:00:24,479 Speaker 1: result of discussions with sources and documents viewed of the 4 00:00:24,480 --> 00:00:27,240 Speaker 1: amounts built on Amazon Web Services, I am for the 5 00:00:27,240 --> 00:00:29,680 Speaker 1: first time in history able to disclose how much AI 6 00:00:29,720 --> 00:00:33,720 Speaker 1: firms are spending on AWS, specifically Anthropic and a coding 7 00:00:33,720 --> 00:00:37,680 Speaker 1: company Cursor, its largest customer for API services. I can 8 00:00:37,720 --> 00:00:40,920 Speaker 1: exclusively reveal today how much Anthropics spent on AWS for 9 00:00:40,920 --> 00:00:42,960 Speaker 1: the years twenty twenty four and from the beginning of 10 00:00:42,960 --> 00:00:45,520 Speaker 1: twenty twenty five through the end of September twenty twenty five, 11 00:00:45,720 --> 00:00:48,120 Speaker 1: and from what I can see, their compute spend may 12 00:00:48,200 --> 00:00:52,239 Speaker 1: vastly exceed what has previously been reported. Furthermore, I can 13 00:00:52,240 --> 00:00:54,800 Speaker 1: confirm that through the end of September twenty twenty five, 14 00:00:54,960 --> 00:00:57,320 Speaker 1: Anthropic is spent around one hundred percent of their revenue 15 00:00:57,320 --> 00:01:00,319 Speaker 1: in twenty twenty five on Amazon Web Services, spending two 16 00:01:00,320 --> 00:01:03,400 Speaker 1: point sixty six billion dollars on compute on an estimated 17 00:01:03,440 --> 00:01:06,840 Speaker 1: two point five five billion dollars in revenue. Go to 18 00:01:06,840 --> 00:01:09,080 Speaker 1: the newsletter. I source the whole goddamn thing, and if 19 00:01:09,120 --> 00:01:11,399 Speaker 1: I'm honest, this piece is the culmination of several months 20 00:01:11,400 --> 00:01:14,679 Speaker 1: of articles about how Anthropics business tactics have made be 21 00:01:14,760 --> 00:01:18,280 Speaker 1: turned the screws on their biggest customer. I can exclusively 22 00:01:18,319 --> 00:01:21,080 Speaker 1: reveal today, as well as many other numbers in the newsletter, 23 00:01:21,319 --> 00:01:24,000 Speaker 1: the curs of Amazon Web services bills doubled from six 24 00:01:24,040 --> 00:01:26,440 Speaker 1: point two million dollars in May twenty twenty five to 25 00:01:26,480 --> 00:01:28,960 Speaker 1: twelve point six million dollars in June twenty twenty five, 26 00:01:29,360 --> 00:01:31,959 Speaker 1: and have stayed inflated since Anthropic increased the costs with 27 00:01:32,000 --> 00:01:35,440 Speaker 1: the launch of priority servers, tiers and aggressive rent seeking measure. 28 00:01:35,959 --> 00:01:38,240 Speaker 1: I need to be clear I cannot one hundred percent 29 00:01:38,319 --> 00:01:40,160 Speaker 1: guarantee that's what did it. I'm going to hedge my 30 00:01:40,200 --> 00:01:43,039 Speaker 1: bets very hard on that, but it certainly Bloody Wealth 31 00:01:43,080 --> 00:01:45,560 Speaker 1: seems that way. It's my guard instinct. I'm not going 32 00:01:45,560 --> 00:01:47,640 Speaker 1: to say it declaratively, but I'm going to show you 33 00:01:47,680 --> 00:01:50,400 Speaker 1: why I believe this, And I admit I struggled with 34 00:01:50,440 --> 00:01:52,560 Speaker 1: how to turn this into an episode because the newsletter, 35 00:01:52,600 --> 00:01:54,760 Speaker 1: which is on my free feed, is a series of 36 00:01:54,840 --> 00:01:57,080 Speaker 1: numbers and analyzes that if I just read them aloud, 37 00:01:57,080 --> 00:01:59,320 Speaker 1: would sound extremely dull and at times be quite hard 38 00:01:59,320 --> 00:02:02,120 Speaker 1: to follow. It's not something that naturally plays well for radio. 39 00:02:03,320 --> 00:02:05,600 Speaker 1: So instead of giving you the audible version, I'm going 40 00:02:05,600 --> 00:02:07,040 Speaker 1: to give you the cliff notes and speak to a 41 00:02:07,080 --> 00:02:10,120 Speaker 1: degree of vindication I feel on reading these costs. So 42 00:02:10,240 --> 00:02:12,800 Speaker 1: let's start with a number. One point two two five 43 00:02:12,880 --> 00:02:15,880 Speaker 1: billion dollars. That's how much Anthropic spent on Amazon Web 44 00:02:15,919 --> 00:02:18,680 Speaker 1: Services in the third quarter of twenty twenty five. They 45 00:02:18,680 --> 00:02:20,880 Speaker 1: spent eight hundred and twenty nine point seven million in 46 00:02:20,960 --> 00:02:23,400 Speaker 1: Q two twenty twenty five, and six hundred and ten 47 00:02:23,400 --> 00:02:25,960 Speaker 1: million dollars in Q one twenty twenty five. Oh and 48 00:02:26,000 --> 00:02:28,680 Speaker 1: one other number, they spent one point three five billion 49 00:02:28,680 --> 00:02:32,360 Speaker 1: dollars on AWS and twenty twenty four. So yeah, just 50 00:02:32,600 --> 00:02:34,840 Speaker 1: in another way, talking of their twenty twenty five numbers, 51 00:02:34,880 --> 00:02:38,160 Speaker 1: anthropics spend on AWS doubled over the course of three quarters. 52 00:02:38,840 --> 00:02:42,440 Speaker 1: Now a little backstory about Anthropic that's necessary to understand 53 00:02:42,440 --> 00:02:45,920 Speaker 1: this fully. Anthropic was originally invested in by both Google 54 00:02:45,960 --> 00:02:48,880 Speaker 1: and Amazon. According to The New York Times, Google owns 55 00:02:48,919 --> 00:02:51,720 Speaker 1: around fourteen percent of the company. An analyst, yes Tomate, 56 00:02:51,800 --> 00:02:55,240 Speaker 1: Amazon owned somewhere between fifteen and eighteen percent, and both have, 57 00:02:55,520 --> 00:02:58,440 Speaker 1: in not so many words, said that they're the main 58 00:02:58,600 --> 00:03:02,680 Speaker 1: or primary compute part for Aanthropic. It's unclear how much 59 00:03:02,720 --> 00:03:05,799 Speaker 1: Anthropic spends on Google Cloud, but semi analysis believes they're 60 00:03:05,800 --> 00:03:08,920 Speaker 1: a big client, and that's about as much detail as 61 00:03:08,960 --> 00:03:11,760 Speaker 1: I can get from anywhere I've really looked. In any case, 62 00:03:11,840 --> 00:03:14,840 Speaker 1: Anthropic is spending effectively every dollar they make on Amazon 63 00:03:14,840 --> 00:03:17,440 Speaker 1: Web Services, and Amazon has appears to be booking this 64 00:03:17,480 --> 00:03:20,400 Speaker 1: as revenue, though I can't directly confirm that, though I 65 00:03:20,440 --> 00:03:23,840 Speaker 1: do know these numbers are cash, they're after credits. Though 66 00:03:23,880 --> 00:03:26,919 Speaker 1: in the recent months, Anthropic has lowered the amount of 67 00:03:26,960 --> 00:03:29,359 Speaker 1: revenue they're spending on it to eighty six point two 68 00:03:29,360 --> 00:03:31,640 Speaker 1: percent in Q three twenty twenty five, which is an 69 00:03:31,680 --> 00:03:33,760 Speaker 1: improvement from Q two to twenty twenty five, where they 70 00:03:33,760 --> 00:03:36,000 Speaker 1: spent one hundred and six percent of their revenue and 71 00:03:36,120 --> 00:03:38,360 Speaker 1: Q one, where they spent one hundred and seventy five 72 00:03:38,360 --> 00:03:41,480 Speaker 1: percent of what they made on Amazon Web Services. It's 73 00:03:41,600 --> 00:03:45,040 Speaker 1: quite horrifying when you say it out loud. Now, if 74 00:03:45,040 --> 00:03:47,520 Speaker 1: you're thinking that, because these numbers are quite close, that 75 00:03:47,560 --> 00:03:51,480 Speaker 1: this might suggest that Anthropics costs are improving, think again. Anthropics. 76 00:03:51,520 --> 00:03:54,640 Speaker 1: Amazon Web services costs of a habit of massively spiking. 77 00:03:54,720 --> 00:03:57,720 Speaker 1: For example, their AWS bill led from three hundred and 78 00:03:57,800 --> 00:04:00,600 Speaker 1: eighty three point seven million dollars in August twenty twenty 79 00:04:00,640 --> 00:04:03,480 Speaker 1: five to five hundred and eighteen point nine million dollars 80 00:04:03,480 --> 00:04:06,720 Speaker 1: in September twenty twenty five. That's one hundred and thirty 81 00:04:06,720 --> 00:04:09,440 Speaker 1: five million goddamn dollars. And my hunch is it's because 82 00:04:09,480 --> 00:04:11,800 Speaker 1: they have a massive problem where clawed code users are 83 00:04:11,920 --> 00:04:14,480 Speaker 1: each costing them thousands of dollars despite only paying one 84 00:04:14,520 --> 00:04:17,279 Speaker 1: hundred or two hundred dollars a month. There's also the 85 00:04:17,360 --> 00:04:20,840 Speaker 1: nasty matter of Google Cloud. Anthropics Amazon Web Services bill 86 00:04:20,920 --> 00:04:23,000 Speaker 1: is two point sixty six billion dollars from January through 87 00:04:23,040 --> 00:04:25,080 Speaker 1: the end of September, as I said, and that is 88 00:04:25,120 --> 00:04:27,320 Speaker 1: pretty close to two point five to five billion dollars 89 00:04:27,360 --> 00:04:30,520 Speaker 1: in revenue. But if Anthropics spend on Google Cloud was 90 00:04:30,560 --> 00:04:33,039 Speaker 1: only twenty five percent of what they spent on AWS, 91 00:04:33,200 --> 00:04:35,440 Speaker 1: its compute cost would jumped to three point three three 92 00:04:35,440 --> 00:04:37,960 Speaker 1: billion dollars through the end of September, way more than 93 00:04:37,960 --> 00:04:40,800 Speaker 1: it brings in. If it's half of what they spent 94 00:04:41,320 --> 00:04:43,960 Speaker 1: on Amazon Web Services, this becomes a three point nine 95 00:04:43,960 --> 00:04:46,560 Speaker 1: to nine billion dollar compute bill, and if they spend 96 00:04:46,600 --> 00:04:49,240 Speaker 1: the same amount, the bill becomes five point three billion dollars, 97 00:04:49,279 --> 00:04:52,520 Speaker 1: and again that's just through the end of September. Another note, 98 00:04:52,680 --> 00:04:55,880 Speaker 1: cursors spend on Amazon Web services is comparatively small, but 99 00:04:55,960 --> 00:04:59,080 Speaker 1: includes some spend on Anthropics models because Amazon is allowed 100 00:04:59,080 --> 00:05:01,479 Speaker 1: to sell them. And I believe that the reason that 101 00:05:01,520 --> 00:05:04,680 Speaker 1: they do this, because they do directly pay Anthropic like 102 00:05:04,680 --> 00:05:07,560 Speaker 1: they actually send money directly to them, is because Amazon 103 00:05:07,600 --> 00:05:10,719 Speaker 1: offers significant discounts in some cases for running models through 104 00:05:10,760 --> 00:05:13,479 Speaker 1: their service. I think it's their bedrock service, and my 105 00:05:13,560 --> 00:05:15,520 Speaker 1: source confirmed that this was the case, though I could 106 00:05:15,560 --> 00:05:19,000 Speaker 1: not get granular data on what exactly CURSES spend was 107 00:05:19,040 --> 00:05:21,599 Speaker 1: on Amazon, like I can't say, oh, they use this 108 00:05:21,720 --> 00:05:24,600 Speaker 1: model or that model. Now, Cursor spends most of their 109 00:05:24,640 --> 00:05:27,680 Speaker 1: compute money directly with Anthropic, as well as every other 110 00:05:27,720 --> 00:05:30,760 Speaker 1: model developer whose models they use. AWS is a small 111 00:05:30,760 --> 00:05:34,240 Speaker 1: piece of the puzzle, and while small, it's spending data 112 00:05:34,279 --> 00:05:37,080 Speaker 1: provides evidence of how much this shit actually costs, though 113 00:05:37,120 --> 00:05:39,560 Speaker 1: I also concede that some of the money CURSES spends 114 00:05:39,560 --> 00:05:41,960 Speaker 1: with AWS likely goes to the non AI part of 115 00:05:41,960 --> 00:05:46,240 Speaker 1: the business, like file hosting and other tech infrastructure. Nevertheless, 116 00:05:46,240 --> 00:05:48,920 Speaker 1: the timing of the massive jumps in CURSES AWS bill 117 00:05:48,960 --> 00:05:51,040 Speaker 1: from six point two million dollars in May to twelve 118 00:05:51,040 --> 00:05:53,880 Speaker 1: point six million dollars in June directly correlate with the 119 00:05:53,920 --> 00:05:56,719 Speaker 1: massive changes made to their product, increasing the costs on 120 00:05:56,800 --> 00:05:58,760 Speaker 1: any users that wanted to use Cursor in the way 121 00:05:58,800 --> 00:06:00,960 Speaker 1: they had in the past by making them face the 122 00:06:01,000 --> 00:06:04,080 Speaker 1: actual costs of serving models on a per million token basis. 123 00:06:04,240 --> 00:06:07,120 Speaker 1: I've written about this a lot, By the way, it's 124 00:06:07,160 --> 00:06:09,960 Speaker 1: hard to describe it in detail because it's going to 125 00:06:10,000 --> 00:06:13,200 Speaker 1: take forever. But around mid June, Cursor had to change 126 00:06:13,240 --> 00:06:17,080 Speaker 1: everything because mysteriously they had to stop spending so much 127 00:06:17,120 --> 00:06:19,599 Speaker 1: money with their customers. Their customers would burning a hole 128 00:06:19,640 --> 00:06:21,640 Speaker 1: in their pocket, and I think we can kind of 129 00:06:21,680 --> 00:06:25,120 Speaker 1: see why. Curses costs have also never come down, spiking 130 00:06:25,160 --> 00:06:27,520 Speaker 1: to a higher fifteen point five million dollars in June, 131 00:06:27,560 --> 00:06:30,200 Speaker 1: dropping to a still high nine point six million dollars 132 00:06:30,440 --> 00:06:32,640 Speaker 1: in August. I need to spike again to twelve point 133 00:06:32,720 --> 00:06:37,040 Speaker 1: nine million dollars in September, though I cannot declaratively state 134 00:06:37,080 --> 00:06:40,720 Speaker 1: that this is exactly what happened. Curses costs doubled immediately 135 00:06:40,720 --> 00:06:43,760 Speaker 1: following the addition of anthropic service tiers in late May 136 00:06:43,800 --> 00:06:47,080 Speaker 1: twenty twenty five, which require an upfront commitment of token 137 00:06:47,120 --> 00:06:50,320 Speaker 1: spend and token throughput, and when Curser announced the launch 138 00:06:50,320 --> 00:06:53,160 Speaker 1: of their two hundred dollars a month ultraplan amidst massive 139 00:06:53,160 --> 00:06:56,040 Speaker 1: product changes, they cited how it was and I quote 140 00:06:56,320 --> 00:06:59,880 Speaker 1: made possible by multi year partnerships from open Ai, Anthropic, 141 00:07:00,160 --> 00:07:03,320 Speaker 1: Google and Xai, and that their support was instrumental in 142 00:07:03,360 --> 00:07:17,760 Speaker 1: offering this volume of computer the predictable price. Now, really, 143 00:07:18,240 --> 00:07:21,080 Speaker 1: I'm being as fair as I can. Another factor might 144 00:07:21,120 --> 00:07:24,520 Speaker 1: be that the new Claude four models was significantly more expensive. 145 00:07:24,880 --> 00:07:27,640 Speaker 1: It's entirely possible that all of these things are true. 146 00:07:27,840 --> 00:07:29,640 Speaker 1: I just want to make sure I cover my basis 147 00:07:29,640 --> 00:07:32,640 Speaker 1: because I do not know for sure. But the timing, 148 00:07:33,120 --> 00:07:36,840 Speaker 1: the timing man and another thing, you know what. Anthropic 149 00:07:36,960 --> 00:07:40,440 Speaker 1: also launched a week before service Tears a competing product, 150 00:07:40,440 --> 00:07:43,000 Speaker 1: a Cursor called claud Code one that they could run 151 00:07:43,040 --> 00:07:45,080 Speaker 1: with as little restraint as they'd like to drain as 152 00:07:45,120 --> 00:07:48,200 Speaker 1: many monthly customers away from Cursor, who is also their 153 00:07:48,280 --> 00:07:52,840 Speaker 1: largest customer, buying Anthropic models through their API. Real fucking mystery, right, 154 00:07:52,920 --> 00:07:54,640 Speaker 1: If it quacks like a dark where's a T shirt 155 00:07:54,640 --> 00:07:57,080 Speaker 1: that says dark? And Claude tells you you're absolutely right, 156 00:07:57,120 --> 00:07:59,240 Speaker 1: that's a duck. When you upload a picture of it, 157 00:07:59,240 --> 00:08:01,960 Speaker 1: it's probably a fucking duck. But I obviously can't say 158 00:08:01,960 --> 00:08:05,400 Speaker 1: for sure. I need to be explicit here with what happened, though. 159 00:08:05,920 --> 00:08:09,320 Speaker 1: Anthropics supplied access to their models to a company Cursor, 160 00:08:09,560 --> 00:08:12,200 Speaker 1: and then released a product claud Code that did exactly 161 00:08:12,240 --> 00:08:15,080 Speaker 1: the same thing as that company Cursor, turning it both 162 00:08:15,080 --> 00:08:18,240 Speaker 1: into a customer and a competitor in the process, creating 163 00:08:18,280 --> 00:08:21,360 Speaker 1: a massive conflict of interest, as not only did Anthropic 164 00:08:21,400 --> 00:08:24,320 Speaker 1: have an incentive for that customer or competitor to fail, 165 00:08:24,360 --> 00:08:26,560 Speaker 1: though they also needed their compute revenue, which is kind 166 00:08:26,560 --> 00:08:29,200 Speaker 1: of a bugger. Anthropic also had the means to make 167 00:08:29,240 --> 00:08:31,560 Speaker 1: this failure happen in the most painful and expensive way 168 00:08:31,640 --> 00:08:34,760 Speaker 1: possible by worsening the terms in which that competitor required 169 00:08:34,760 --> 00:08:38,320 Speaker 1: the compute it needed to function. Could be a coincidence, 170 00:08:38,360 --> 00:08:41,040 Speaker 1: I guess. And when I say compute, I mean tokens. 171 00:08:41,240 --> 00:08:44,439 Speaker 1: Just I'm reading a script. Okay, where us going to 172 00:08:44,480 --> 00:08:46,960 Speaker 1: eam Anyway, I'm not going to turn this into a massive, 173 00:08:47,040 --> 00:08:49,400 Speaker 1: sprawling episode about this company. I wanted to give you 174 00:08:49,480 --> 00:08:52,079 Speaker 1: the raw information so you can go and read the 175 00:08:52,120 --> 00:08:55,240 Speaker 1: detailed analysis. I did. It's free, by the way, don't worry. 176 00:08:55,880 --> 00:08:57,400 Speaker 1: But now I want to talk about how all this 177 00:08:57,559 --> 00:08:59,880 Speaker 1: made me feel, because that's what makes this show unique. 178 00:09:00,000 --> 00:09:01,800 Speaker 1: I think is the appropriate way of coming out this. 179 00:09:02,360 --> 00:09:05,120 Speaker 1: I'm going to be honest, I find what it looks like, 180 00:09:05,200 --> 00:09:08,240 Speaker 1: and I'm hedging my beds again. Anthropic did to Cursor 181 00:09:08,600 --> 00:09:12,800 Speaker 1: truly disgusting. Cursor hit five hundred million dollars in annualized 182 00:09:12,840 --> 00:09:15,360 Speaker 1: revenue in the same month that they then saw their 183 00:09:15,440 --> 00:09:19,080 Speaker 1: cost double, dramatically reducing the value of their subscription product 184 00:09:19,120 --> 00:09:22,320 Speaker 1: at the apex of their success. Yes, Cursor is an 185 00:09:22,360 --> 00:09:25,480 Speaker 1: unsustainable AI company I know, and like all of these companies, 186 00:09:25,520 --> 00:09:28,520 Speaker 1: has no part of the profitability. Anthropic should have always 187 00:09:28,640 --> 00:09:31,240 Speaker 1: charged sustainable rates, even if it meant that it wasn't 188 00:09:31,240 --> 00:09:34,400 Speaker 1: possible to build a big company based on their models. Sadly, 189 00:09:34,480 --> 00:09:36,960 Speaker 1: we don't live in that universe. And while you could 190 00:09:37,000 --> 00:09:39,240 Speaker 1: make the case that startups like Uber didn't at first 191 00:09:39,320 --> 00:09:41,679 Speaker 1: charge sustainable rates, I'd argue that the reason why its 192 00:09:41,720 --> 00:09:44,560 Speaker 1: initial rates weren't successful was because of the steep upfront 193 00:09:44,600 --> 00:09:46,840 Speaker 1: cost of customer acquisition, which is the problem that could 194 00:09:46,840 --> 00:09:49,000 Speaker 1: be solved through the lifetime of the customer and Uber 195 00:09:49,000 --> 00:09:50,920 Speaker 1: had the means to gradually ratchet up the costs of 196 00:09:51,040 --> 00:09:53,839 Speaker 1: rides or mores. Italy reduced the cut that they pay 197 00:09:53,880 --> 00:09:57,439 Speaker 1: to drivers in a way that wouldn't be immediately paying for. Furthermore, 198 00:09:57,720 --> 00:10:00,440 Speaker 1: Uber never had a fuel problem. What Anthropic has as 199 00:10:00,480 --> 00:10:02,640 Speaker 1: a fuel problem, they have a compute problem for the 200 00:10:02,720 --> 00:10:06,040 Speaker 1: amount that they're paying to run their goddamn services. Cursor 201 00:10:06,120 --> 00:10:09,160 Speaker 1: is also Anthropic's largest customer, and the timing of priority 202 00:10:09,200 --> 00:10:11,280 Speaker 1: tiers to coincide at the moment when they were growing 203 00:10:11,400 --> 00:10:15,360 Speaker 1: fastest is a suspicious and potentially disgraceful move. While you 204 00:10:15,400 --> 00:10:17,760 Speaker 1: could describe it as a necessary step in the direction 205 00:10:17,840 --> 00:10:21,280 Speaker 1: of sustainability, that plausible excuse is undercut by the overall 206 00:10:21,360 --> 00:10:24,319 Speaker 1: timing of the move. One cannot ignore how close the 207 00:10:24,400 --> 00:10:26,600 Speaker 1: launch of these tiers were to the launch of anthropics 208 00:10:26,679 --> 00:10:29,680 Speaker 1: clawed Code, a product that lacks curses, flashy front end, 209 00:10:29,840 --> 00:10:33,319 Speaker 1: but performs similar functions, all subsidized by anthropics massive hordes 210 00:10:33,360 --> 00:10:36,880 Speaker 1: of venture capital, and its chummy relationships with hyperscalers like 211 00:10:37,000 --> 00:10:40,040 Speaker 1: Amazon and Google. The thing is, even with these moves, 212 00:10:40,080 --> 00:10:42,840 Speaker 1: Anthropics still spent a dollar and four cents on Amazon 213 00:10:42,880 --> 00:10:45,240 Speaker 1: Web services for every dollar they made through the end 214 00:10:45,280 --> 00:10:48,160 Speaker 1: of September twenty twenty five, and that's for just twenty 215 00:10:48,240 --> 00:10:51,040 Speaker 1: twenty five. By the way, their costs increase nearly with 216 00:10:51,160 --> 00:10:53,679 Speaker 1: their revenue. And while they've improved, when they spent a 217 00:10:53,720 --> 00:10:56,200 Speaker 1: remarkable two hundred and twenty seven percent of their revenue 218 00:10:56,200 --> 00:10:59,000 Speaker 1: on AWS in January, they still spent eighty eight point 219 00:10:59,080 --> 00:11:02,520 Speaker 1: nine percent of it on a western September. Now, if 220 00:11:02,559 --> 00:11:05,120 Speaker 1: you're worried hearing how close these numbers before, like I said, 221 00:11:05,400 --> 00:11:09,920 Speaker 1: means there's somehow approaching profitability. Good lord, No, I'm repeating myself. 222 00:11:09,960 --> 00:11:11,680 Speaker 1: I realized, But I really need you to come away 223 00:11:11,720 --> 00:11:14,760 Speaker 1: with this with reality in your brain. These digital mister 224 00:11:14,840 --> 00:11:18,320 Speaker 1: beans very likely spend comparable sums on Google Cloud unlikely 225 00:11:18,360 --> 00:11:20,760 Speaker 1: another billion or two one salaries data And I don't 226 00:11:20,800 --> 00:11:23,080 Speaker 1: know that one point five billion dollar settlement with all 227 00:11:23,080 --> 00:11:26,000 Speaker 1: the authors that they just agreed to. This company absolutely 228 00:11:26,040 --> 00:11:28,680 Speaker 1: fucking sucks. I don't care if you like Claude Sonnet 229 00:11:28,760 --> 00:11:30,880 Speaker 1: or claud Opus. I don't give a fuck. Claude Opus 230 00:11:30,920 --> 00:11:33,280 Speaker 1: and Claudes on It are not worth burning billions of 231 00:11:33,360 --> 00:11:36,640 Speaker 1: dollars a year in cloud costs, fueling an environmentally destructive 232 00:11:36,720 --> 00:11:39,719 Speaker 1: plagiarism charge pseudo company that would roll over and die 233 00:11:39,800 --> 00:11:42,319 Speaker 1: within months if it didn't constantly get fed billions of 234 00:11:42,360 --> 00:11:44,800 Speaker 1: dollars a year. What are you gonna tell me they're 235 00:11:44,800 --> 00:11:47,000 Speaker 1: gonna turn this ship around. They're gonna make some sort 236 00:11:47,040 --> 00:11:50,680 Speaker 1: of autonomous AI coder. You know that's bullshit. Every goddamn 237 00:11:50,760 --> 00:11:53,600 Speaker 1: one of you boosters knows that total bullshit. I'm sure 238 00:11:53,720 --> 00:11:56,360 Speaker 1: son at four point five is somewhat better than Sonnet four, 239 00:11:56,559 --> 00:12:00,400 Speaker 1: but what does that actually mean? Anthropic raised twenty billion 240 00:12:00,400 --> 00:12:03,000 Speaker 1: dollars this year? Do we give them more next year? 241 00:12:03,840 --> 00:12:06,680 Speaker 1: I've heard reports that they're actually targeting twenty billion dollars 242 00:12:06,720 --> 00:12:09,280 Speaker 1: in anialized revenue, so one point six to seven billion 243 00:12:09,320 --> 00:12:12,679 Speaker 1: dollars a month in revenue by the end of next year. 244 00:12:12,679 --> 00:12:16,040 Speaker 1: It's an absolute fucking joke. But the only thing funnier 245 00:12:16,120 --> 00:12:18,160 Speaker 1: than that joke is that it will likely cost them 246 00:12:18,160 --> 00:12:20,880 Speaker 1: twenty five billion dollars to make that fictional money. And 247 00:12:20,960 --> 00:12:25,040 Speaker 1: where preytell is that coming from? And why? Why? What 248 00:12:25,240 --> 00:12:27,360 Speaker 1: is so remarkable about this company that gives them a 249 00:12:27,400 --> 00:12:29,760 Speaker 1: free paster burn two point sixty six billion dollars in 250 00:12:29,800 --> 00:12:33,600 Speaker 1: AWS in fucking nine months. I'm not talking about your 251 00:12:33,640 --> 00:12:36,640 Speaker 1: cynical oh, Amazon is booking at is revenue crony capitalisms? 252 00:12:36,679 --> 00:12:39,080 Speaker 1: Here an'tswer it. I'm not, I'm not. I'm talking about 253 00:12:39,120 --> 00:12:42,920 Speaker 1: the scientific or technological reasoning for keeping Anthropic alive. And yes, 254 00:12:43,040 --> 00:12:46,280 Speaker 1: I feel exactly the same way about Open AI. What 255 00:12:46,520 --> 00:12:50,080 Speaker 1: possible achievement does Anthropic have that warrants this needless, endless, 256 00:12:50,200 --> 00:12:54,120 Speaker 1: sprawling financial destruction. Why are we rewarding a company with 257 00:12:54,240 --> 00:12:57,079 Speaker 1: bad business practices for making a product that loses more 258 00:12:57,120 --> 00:13:00,240 Speaker 1: money the more money it makes. I'll even try and 259 00:13:00,360 --> 00:13:03,160 Speaker 1: see this through the eyes of an AI booster. Damn, 260 00:13:03,200 --> 00:13:05,520 Speaker 1: all I'm seeing is blue and yellow anyway, And even 261 00:13:05,640 --> 00:13:08,439 Speaker 1: from here, the only reason to keep Anthropic alive is 262 00:13:08,480 --> 00:13:11,560 Speaker 1: because you see these companies as sports teams. You see 263 00:13:11,640 --> 00:13:15,760 Speaker 1: Dario Amiday as the equivalent of Dan Campbell or Greg Popovich. 264 00:13:15,800 --> 00:13:17,800 Speaker 1: You root for them and their causes because you think 265 00:13:17,880 --> 00:13:20,760 Speaker 1: that if they win, you as a fan will be rewarded. 266 00:13:21,120 --> 00:13:22,839 Speaker 1: You don't think too hard about what it is that 267 00:13:22,920 --> 00:13:25,959 Speaker 1: Claude Sonnet or Claude Opus do, and you find enough 268 00:13:26,000 --> 00:13:28,640 Speaker 1: ways that this is somewhat kind of useful to you, 269 00:13:28,880 --> 00:13:31,400 Speaker 1: and you use those reasons to justify the proliferation of 270 00:13:31,440 --> 00:13:36,840 Speaker 1: a wasteful and destructive technology. What exactly happens here? Anthropics 271 00:13:36,880 --> 00:13:40,240 Speaker 1: AWS bills are not really going down. They've normalized in 272 00:13:40,280 --> 00:13:42,439 Speaker 1: an eighty eight to ninety five percent range and they're 273 00:13:42,480 --> 00:13:44,679 Speaker 1: clearly going to stay there. And if your argument is 274 00:13:45,000 --> 00:13:47,760 Speaker 1: they'll go down, your argument is quite literally no, eh. 275 00:13:48,480 --> 00:13:51,000 Speaker 1: Go read semi analysis for seventeen hours and come up 276 00:13:51,040 --> 00:13:55,160 Speaker 1: with some demented GPU based argument about inference smax scores, 277 00:13:55,360 --> 00:13:57,400 Speaker 1: pretend like you give a shit, come up with a 278 00:13:57,480 --> 00:14:00,200 Speaker 1: real argument against mine, because I am working hard, are 279 00:14:00,240 --> 00:14:02,240 Speaker 1: at this than you are, And if you believe otherwise, 280 00:14:02,360 --> 00:14:04,320 Speaker 1: you should ask yourself why the guy who said Sam 281 00:14:04,400 --> 00:14:07,439 Speaker 1: Altman's no it loads refused cash dump in a premium 282 00:14:07,480 --> 00:14:11,079 Speaker 1: newsletter got this scoop and you did not. But that 283 00:14:11,160 --> 00:14:14,120 Speaker 1: actually leads me to a key question. How long do 284 00:14:14,280 --> 00:14:17,760 Speaker 1: we hand Anthropic and by extension, open AI billions of dollars? 285 00:14:18,400 --> 00:14:21,080 Speaker 1: And for the first time in your goddamn life, it's 286 00:14:21,160 --> 00:14:24,400 Speaker 1: time to ask, what if I'm right? What if these 287 00:14:24,480 --> 00:14:28,160 Speaker 1: companies are incapable of becoming profitable? What if there really 288 00:14:28,280 --> 00:14:31,960 Speaker 1: is no massive demand for generative AI? Do you really 289 00:14:32,000 --> 00:14:34,720 Speaker 1: think Anthropic will make one point six billion dollars a month? 290 00:14:34,800 --> 00:14:37,840 Speaker 1: Sometime in twenty twenty six, do you really think that? 291 00:14:38,520 --> 00:14:42,120 Speaker 1: And even for Amazon, it's kind of shit wow to 292 00:14:42,840 --> 00:14:45,880 Speaker 1: do a couple billion on one hundred and five billion 293 00:14:45,920 --> 00:14:47,760 Speaker 1: dollars of capex. I might have even said this later 294 00:14:47,800 --> 00:14:50,760 Speaker 1: in the script, but just thinking about it makes me 295 00:14:50,840 --> 00:14:54,240 Speaker 1: feel a little crazy. And look, I get there's a 296 00:14:54,320 --> 00:14:56,840 Speaker 1: middle ground here where people say that there's some sort 297 00:14:56,920 --> 00:14:59,320 Speaker 1: of use case that sort of works for AI, where 298 00:14:59,360 --> 00:15:01,440 Speaker 1: you hit it hard enough to write good enough prompts 299 00:15:01,480 --> 00:15:03,320 Speaker 1: or whatever that you like it for search, that your 300 00:15:03,360 --> 00:15:05,680 Speaker 1: brainstorm with it, they helped you pick out a hat 301 00:15:05,760 --> 00:15:07,840 Speaker 1: that you used it to solve some sort of problem. Once, 302 00:15:08,200 --> 00:15:10,680 Speaker 1: I just want to ask you how much of those 303 00:15:10,720 --> 00:15:14,240 Speaker 1: anecdotes really worth to you? How impressed with these things 304 00:15:14,360 --> 00:15:18,360 Speaker 1: are you? Would you pay double, triple, quadruple? Would you 305 00:15:18,440 --> 00:15:20,600 Speaker 1: pay on a meted basis where those little flights of 306 00:15:20,680 --> 00:15:23,000 Speaker 1: fancy cost you a few cents, then ten cents, then 307 00:15:23,040 --> 00:15:25,600 Speaker 1: a dollar, because that's how much it costs to provide 308 00:15:25,640 --> 00:15:27,320 Speaker 1: these services, and at some point you're going to be 309 00:15:27,360 --> 00:15:43,280 Speaker 1: made to pay for it one way or another. Advertising 310 00:15:43,320 --> 00:15:45,840 Speaker 1: won't be the answer. By the way, the literal only 311 00:15:45,920 --> 00:15:49,080 Speaker 1: company to try advertising in large language models is an 312 00:15:49,120 --> 00:15:52,840 Speaker 1: AI search engine company called Perplexity, and they just paused 313 00:15:52,880 --> 00:15:56,280 Speaker 1: accepting new advertisers too, and I quote ad week rethink 314 00:15:56,320 --> 00:16:00,520 Speaker 1: how ads fit into its AI search experience. They made 315 00:16:00,600 --> 00:16:04,160 Speaker 1: twenty thousand dollars in twenty twenty four in advertising revenue. 316 00:16:04,560 --> 00:16:07,520 Speaker 1: Are we supposed to be impressed that Perplexity made enough 317 00:16:07,600 --> 00:16:11,080 Speaker 1: revenue to buy a second hand Toyota Corolla. There are 318 00:16:11,120 --> 00:16:15,000 Speaker 1: people making more money than that's slinging fucking Herbalife. And 319 00:16:15,120 --> 00:16:18,680 Speaker 1: this is literally the exact company that should have succeeded 320 00:16:19,000 --> 00:16:22,080 Speaker 1: based on any kind of ads will fix everything argument. 321 00:16:22,480 --> 00:16:25,440 Speaker 1: And they couldn't even buy courtside tickets at the NBA playoffs. 322 00:16:26,480 --> 00:16:29,720 Speaker 1: The costs are increasing linearly with revenue, and I've fucking 323 00:16:29,880 --> 00:16:32,640 Speaker 1: proved it. I am open to any compelling arguments that 324 00:16:32,720 --> 00:16:35,640 Speaker 1: can explain how this ever changes. And my god, if 325 00:16:35,720 --> 00:16:38,920 Speaker 1: you say trainium, I will absolutely lose my shit. Chips 326 00:16:38,920 --> 00:16:43,400 Speaker 1: aren't fixing this. By the way, if your answer is 327 00:16:43,440 --> 00:16:46,400 Speaker 1: the anthropic will make some sort of theoretical ultrapowerful large 328 00:16:46,440 --> 00:16:49,080 Speaker 1: language model or invent agi, you are a goddamn mark. 329 00:16:49,160 --> 00:16:55,480 Speaker 1: You are being conned. Look join me. I'm serious. There's 330 00:16:55,560 --> 00:16:58,320 Speaker 1: no harm in being wrong I've been wrong tons of 331 00:16:58,360 --> 00:17:01,400 Speaker 1: times in my life. Being wrong and admitting you're wrong 332 00:17:01,520 --> 00:17:04,000 Speaker 1: is an act of bravery. Shit, actually kind of get it. 333 00:17:04,480 --> 00:17:06,640 Speaker 1: This stuff feels if you let it, like it's doing 334 00:17:06,760 --> 00:17:09,119 Speaker 1: something for you, even though interacting with it is actually 335 00:17:09,200 --> 00:17:11,560 Speaker 1: draining you because you're constantly having to find ways to 336 00:17:11,640 --> 00:17:13,240 Speaker 1: make it do what you want it to do, to 337 00:17:13,359 --> 00:17:15,159 Speaker 1: the point that when it actually does something for the 338 00:17:15,240 --> 00:17:19,120 Speaker 1: first time, it almost feels magical. You feel very powerful, 339 00:17:19,560 --> 00:17:21,800 Speaker 1: despite the fact that you have been put to work 340 00:17:22,119 --> 00:17:25,320 Speaker 1: to make automation work. That's not how automation is meant 341 00:17:25,400 --> 00:17:29,119 Speaker 1: to work. And sure, there are software engineers out there 342 00:17:29,119 --> 00:17:31,600 Speaker 1: who have, like any good software engineer, found a way 343 00:17:31,600 --> 00:17:33,800 Speaker 1: to take the useful parts of llms and use them to, 344 00:17:34,000 --> 00:17:36,239 Speaker 1: to quote Carl Brown of the Internet at bugs, make 345 00:17:36,320 --> 00:17:39,000 Speaker 1: the easy things easier. Then there are the ones that 346 00:17:39,040 --> 00:17:41,679 Speaker 1: are spending more time than they were building software, prompting 347 00:17:41,760 --> 00:17:44,800 Speaker 1: l elms and rewriting claude dot md files and thinking 348 00:17:44,840 --> 00:17:47,320 Speaker 1: that because things sort of worked off they hit enter 349 00:17:47,359 --> 00:17:50,639 Speaker 1: that they're privy to a great becoming. And there are 350 00:17:50,680 --> 00:17:54,119 Speaker 1: the victims, of course, of vibe coding startups companies that 351 00:17:54,240 --> 00:17:56,840 Speaker 1: sell the outright lie that somebody who cannot read or 352 00:17:56,840 --> 00:18:01,760 Speaker 1: write software can write secure, effective and functional software. Look, 353 00:18:02,320 --> 00:18:04,960 Speaker 1: I'm serious, join me. If you're an AI booster, I 354 00:18:05,080 --> 00:18:09,280 Speaker 1: don't care. Everybody is welcome. In reality, I don't care 355 00:18:09,320 --> 00:18:10,879 Speaker 1: who you are. I don't care if I've called you 356 00:18:10,960 --> 00:18:13,639 Speaker 1: a booster and given you a verbal swirly one hundred times. 357 00:18:14,119 --> 00:18:16,200 Speaker 1: Now is the time to accept that this software is 358 00:18:16,240 --> 00:18:19,960 Speaker 1: too expensive, too destructive, and too wasteful to continue backing it. 359 00:18:20,359 --> 00:18:22,240 Speaker 1: I'm not even saying you have to say fuck AI 360 00:18:22,440 --> 00:18:25,119 Speaker 1: or shun chat GPT like you're an armish teenager that 361 00:18:25,200 --> 00:18:28,000 Speaker 1: looked at porno. But it's time to be loud and 362 00:18:28,119 --> 00:18:30,440 Speaker 1: direct that these products are not worth the egregious and 363 00:18:30,440 --> 00:18:34,879 Speaker 1: perpetual annihilation of billions of dollars every fucking year. I 364 00:18:34,960 --> 00:18:36,560 Speaker 1: don't even know if this means you have to stop 365 00:18:36,680 --> 00:18:39,440 Speaker 1: using them. I don't want you to, but I don't like. 366 00:18:39,840 --> 00:18:42,520 Speaker 1: What are we gonna do. These things are not gonna 367 00:18:42,560 --> 00:18:45,119 Speaker 1: go away because you stopped using Claude. They're gonna go 368 00:18:45,200 --> 00:18:48,359 Speaker 1: away because you stop talking about them. They're gonna go 369 00:18:48,440 --> 00:18:51,760 Speaker 1: away because they cost too much in their pay pigs 370 00:18:51,800 --> 00:18:55,840 Speaker 1: stop paying them. What I am advocating for is for 371 00:18:56,000 --> 00:18:58,639 Speaker 1: everybody to openly discuss that the amount of money it 372 00:18:58,720 --> 00:19:01,040 Speaker 1: costs to run these companies is that odds with what 373 00:19:01,160 --> 00:19:03,600 Speaker 1: they have built, are building and will build in the future. 374 00:19:04,280 --> 00:19:07,439 Speaker 1: Nothing they are building is moving towards superintelligence or AGI. 375 00:19:07,920 --> 00:19:11,199 Speaker 1: No combination of Amazon Tranium or Google TPUs is going 376 00:19:11,280 --> 00:19:14,159 Speaker 1: to usher in the birth of the machine guard. The 377 00:19:14,280 --> 00:19:17,640 Speaker 1: products they make are, at best, and in inconsistent moments, 378 00:19:17,880 --> 00:19:21,600 Speaker 1: kind of cool, but one hundred times more often mediocre, unreliable, 379 00:19:21,640 --> 00:19:26,600 Speaker 1: and outright ridiculous, even if you really get a lot 380 00:19:26,800 --> 00:19:29,560 Speaker 1: out of these models. Do you think that these companies 381 00:19:29,600 --> 00:19:31,920 Speaker 1: should be allowed to burn billions of dollars every year? 382 00:19:32,760 --> 00:19:34,560 Speaker 1: How much do you think they should be allowed to burn? 383 00:19:35,240 --> 00:19:38,520 Speaker 1: And how much is too much for you? It's time 384 00:19:38,600 --> 00:19:42,280 Speaker 1: to start having this conversation and having it publicly, especially 385 00:19:42,320 --> 00:19:45,200 Speaker 1: as Clammy Sam Mortman Bloviate's about building two hundred and 386 00:19:45,240 --> 00:19:48,879 Speaker 1: fifty gigawatts of data centers in seven goddamn years at 387 00:19:48,880 --> 00:19:51,680 Speaker 1: the cost of one third of America's entire fucking economic 388 00:19:51,760 --> 00:19:54,680 Speaker 1: output in twenty twenty four. Anyway, this has been a 389 00:19:54,720 --> 00:19:56,280 Speaker 1: big day for me, so I'm going to leave it there. 390 00:19:56,480 --> 00:19:59,280 Speaker 1: It's a huge scoop. I'm grateful that I get to 391 00:19:59,320 --> 00:20:01,879 Speaker 1: do this every day, grateful for you listening, and grateful 392 00:20:02,000 --> 00:20:05,159 Speaker 1: for your reading. I hope you've enjoyed this episode and 393 00:20:05,440 --> 00:20:15,680 Speaker 1: thank you as eva for supporting my work. Thank you 394 00:20:15,720 --> 00:20:18,440 Speaker 1: for listening to Better Offline. The editor and composer of 395 00:20:18,480 --> 00:20:21,440 Speaker 1: the Better Offline theme song is Matasowski. You can check 396 00:20:21,480 --> 00:20:24,400 Speaker 1: out more of his music and audio projects at Matasowski 397 00:20:24,440 --> 00:20:27,879 Speaker 1: dot com, M A T T O. S O w 398 00:20:28,280 --> 00:20:31,800 Speaker 1: Ski dot com. You can email me at easy at 399 00:20:31,840 --> 00:20:34,439 Speaker 1: Better Offline dot com or visit Better Offline dot com 400 00:20:34,520 --> 00:20:36,920 Speaker 1: to find more podcast links and of course, my newsletter. 401 00:20:37,359 --> 00:20:39,840 Speaker 1: I also really recommend you go to chat dot Where's 402 00:20:39,880 --> 00:20:42,320 Speaker 1: Youreed dot at to visit the discord, and go to 403 00:20:42,400 --> 00:20:45,760 Speaker 1: our slash Better Offline to check out our reddit. Thank 404 00:20:45,800 --> 00:20:49,200 Speaker 1: you so much for listening. Better Offline is a production 405 00:20:49,320 --> 00:20:51,720 Speaker 1: of cool Zone Media. For more from cool Zone Media, 406 00:20:52,080 --> 00:20:55,199 Speaker 1: visit our website cool Zonemedia dot com, or check us 407 00:20:55,240 --> 00:20:58,159 Speaker 1: out on the iHeartRadio app, Apple Podcasts, or wherever you 408 00:20:58,280 --> 00:20:59,080 Speaker 1: get your podcasts.