WEBVTT - A Lineless World: AI, X, Darkness 0:00:02.720 --> 0:00:10.040 Bloomberg Audio Studios, Podcasts, Radio News. Hello and welcome to 0:00:10.039 --> 0:00:13.080 the Money Stuff Podcast. You're a weekly podcast. Where are 0:00:13.119 --> 0:00:17.439 we talking about stuff related to money? I'm Matt Levine 0:00:17.480 --> 0:00:19.880 and I write the Money Stuff column for Bloomberg Opinion. 0:00:20.160 --> 0:00:22.759 And I'm Katie Greifeld, a reporter for Bloomberg News and 0:00:22.800 --> 0:00:24.280 an anchor for Bloomberg Television. 0:00:25.880 --> 0:00:27.960 What are we talking about today, Katie? 0:00:28.120 --> 0:00:30.520 We're going to talk about how the US stock market 0:00:30.560 --> 0:00:33.760 got deep seeked. We're going to talk about that sweet 0:00:33.760 --> 0:00:36.760 Sweet ex debt, and then we're going to talk about trading. 0:00:36.479 --> 0:00:36.960 In the dark. 0:00:37.600 --> 0:00:45.720 Sounds good, deep seek Deep Seek. So I cannot believe. 0:00:46.000 --> 0:00:48.919 I cannot believe that there is a money Stuff from 0:00:49.080 --> 0:00:51.960 June that mentions deep seek because I would say ninety 0:00:52.000 --> 0:00:54.680 seven percent of the people that I talked to on 0:00:54.760 --> 0:00:57.960 Monday hadn't heard of it before this past weekend. 0:00:58.480 --> 0:00:59.520 Where did they hear of it? 0:01:00.040 --> 0:01:01.680 The people I talked to on Monday who hadn't heard 0:01:01.680 --> 0:01:05.880 of it previously, Yeah, they heard about it when the 0:01:05.920 --> 0:01:10.160 app shot up in the app store, Okay, and then 0:01:10.319 --> 0:01:13.160 you had the cell side start writing about it. There 0:01:13.160 --> 0:01:15.200 were a ton of tweets about it over the weekend. 0:01:15.520 --> 0:01:17.679 Not that Twitter is real life, but in some cases 0:01:17.720 --> 0:01:18.360 it kind of is. 0:01:18.959 --> 0:01:21.199 Yeah, Like there's one guy who wrote a long report 0:01:21.200 --> 0:01:24.959 on like his personal website and argues somewhat plausibly that 0:01:25.000 --> 0:01:28.200 he influenced a lot of the investor reaction. It's like 0:01:28.240 --> 0:01:31.520 interesting to see how like investor actions collas, right, because 0:01:31.560 --> 0:01:33.560 like the model was kind of released like at the 0:01:33.560 --> 0:01:36.880 Piniere last week. The catalyst is some combination of people 0:01:36.920 --> 0:01:38.720 getting to think about it over the weekend and like 0:01:38.800 --> 0:01:40.640 it shooting up in the app store, But at some 0:01:40.680 --> 0:01:43.199 point there's like this big shift from like people using 0:01:43.200 --> 0:01:47.920 the app to everyone having existential crisis about it. Nvidia, Yeah, 0:01:48.240 --> 0:01:50.320 this is a digression. Yes, I wrote about it in June. 0:01:50.360 --> 0:01:53.280 I'm prescient. I was like, yeah, this is going to 0:01:53.320 --> 0:01:56.080 be really bad for the future of netin. But I 0:01:56.120 --> 0:01:57.600 do love you know what I read about in June 0:01:57.680 --> 0:02:01.760 is like the founder of deep seek it kind of 0:02:01.760 --> 0:02:05.520 spun out of his quantitative hedge fund, and I love 0:02:05.560 --> 0:02:09.440 that the skill sets of hedge funds and large language 0:02:09.440 --> 0:02:11.760 models are like kind of overlapping, right. These are kind 0:02:11.800 --> 0:02:15.799 of like, you know, using machine learning techniques to predict 0:02:16.040 --> 0:02:18.799 somewhat unpredictable things, whether that's like the next word in 0:02:18.840 --> 0:02:22.440 a sentence, or the stocks that will go up. And 0:02:23.440 --> 0:02:28.520 classically people made billions of dollars, and by people I 0:02:28.560 --> 0:02:31.400 mean like renaissance technologies made billions of dollars, but like 0:02:31.440 --> 0:02:34.280 repurposing like people in the business of like natural language 0:02:34.280 --> 0:02:37.560 generation into predicting stock prices. And it's nice to see 0:02:37.560 --> 0:02:39.680 that come full circle and the people who are using 0:02:39.840 --> 0:02:42.440 machine learning to predict stock prices are now getting back 0:02:42.480 --> 0:02:45.560 into the natural language game and making billions of dollars 0:02:45.560 --> 0:02:45.880 that way. 0:02:46.080 --> 0:02:49.359 Yeah, I mean everything, everything is cyclical in that sense. 0:02:49.560 --> 0:02:51.320 I will say I wish we had talked about it 0:02:51.360 --> 0:02:53.760 on the podcast in June so that I could have 0:02:53.840 --> 0:02:56.600 shared in some of this whole it was so early. 0:02:56.800 --> 0:02:59.359 But in any case, it's fine. 0:02:59.520 --> 0:03:01.360 The thing is not that like deep Seek is an 0:03:01.400 --> 0:03:03.840 AI company in China. The thing is that the market 0:03:03.919 --> 0:03:06.359 lost the trillion dollars of market cap on Monday. I 0:03:06.400 --> 0:03:09.959 don't want to say that the US economy is based 0:03:10.040 --> 0:03:14.320 on like building empowering data centers for AI companies, but 0:03:14.400 --> 0:03:17.880 the projected incremental cash flows to the US economy, like 0:03:17.919 --> 0:03:19.400 a lot of those are like, yes, we're going to 0:03:19.400 --> 0:03:21.200 build a lot of data centers and a lot of 0:03:21.280 --> 0:03:25.720 like power plants to power them, and deep Seek arguably 0:03:26.240 --> 0:03:29.279 people think that it undermines that case, like quite significantly. 0:03:29.639 --> 0:03:32.359 And yeah, if like you thought that all of economic 0:03:32.360 --> 0:03:34.880 growth would come from like AI data centers, then now 0:03:34.920 --> 0:03:37.440 you're like, yeah, the source of economic growth is gone, 0:03:37.600 --> 0:03:39.120 which is a funny thing to think, but. 0:03:39.200 --> 0:03:41.360 Yeah, I know, I think a lot of people actually 0:03:41.600 --> 0:03:43.960 would agree with that logic because you take a look 0:03:44.000 --> 0:03:45.840 at what happened in the bond market, like there was 0:03:45.880 --> 0:03:49.160 this insane bid into bonds on Monday as well, and 0:03:49.320 --> 0:03:51.640 sort of that was bonds being a haven. 0:03:51.720 --> 0:03:54.240 Yeah, I saw Deep Seek is a macro event. 0:03:54.440 --> 0:03:58.120 Yeah. Well one of the most credible reasons I got 0:03:58.400 --> 0:04:00.440 was just because people are worried that this is going 0:04:00.480 --> 0:04:03.920 to shave that incremental bid off of GDP, because there's 0:04:03.920 --> 0:04:08.240 been plenty of risk off events where bonds didn't catch 0:04:08.240 --> 0:04:10.440 a bid, but this was the one thing that spurred 0:04:10.480 --> 0:04:13.160 like this haven bid across the treasury curve, which is 0:04:13.200 --> 0:04:13.840 pretty funny. 0:04:14.160 --> 0:04:17.320 I love AI as a as a subject matter because 0:04:17.320 --> 0:04:20.440 it is so science fictional. And you know, one thing 0:04:20.440 --> 0:04:24.320 people talk about sometimes is whether GDP is like a 0:04:24.360 --> 0:04:27.159 bad index of like human flourishing, right, And so there's 0:04:27.200 --> 0:04:31.000 an argument that like GDP measures understate GDP growth because 0:04:31.000 --> 0:04:33.760 like you're getting all these like hedonic benefits from like 0:04:33.760 --> 0:04:36.440 going on social media or whatever, and that's like captured 0:04:36.480 --> 0:04:40.080 in GDP figures or whatever. The thesis here is something like, 0:04:40.760 --> 0:04:44.080 if we have to build a lot of buildings and 0:04:44.160 --> 0:04:48.280 burn a lot of coal or natural gas to make 0:04:48.480 --> 0:04:51.160 really good AI models, then that's good for GDP. And 0:04:51.200 --> 0:04:53.880 if you can get the same AIA benefits for free, 0:04:53.960 --> 0:04:56.680 then that's bad for GDP. But like that's better, right, 0:04:56.720 --> 0:04:59.880 It's clearly better to have those benefits without burning coal, 0:05:00.080 --> 0:05:02.320 to have them, you know, and burn coal. I wrote 0:05:02.320 --> 0:05:04.000 something like, you know, if you sort of like fully 0:05:04.040 --> 0:05:06.160 believe the deep Seak thesis and that's like this magic 0:05:06.200 --> 0:05:09.800 AI thing that is free, that's clearly better for human flourishing, 0:05:09.839 --> 0:05:13.000 but it's like worse for stock market capitalization because no 0:05:13.000 --> 0:05:14.120 one can make a profit from it. 0:05:14.240 --> 0:05:16.760 And I mean, can we flourish if the stock market 0:05:16.800 --> 0:05:19.039 doesn't go higher? I don't know, Matt, I'm scared to 0:05:19.040 --> 0:05:20.039 find out the answer. 0:05:20.360 --> 0:05:23.760 It's a real question. There's a Simpsons bit where Homer 0:05:23.760 --> 0:05:26.839 has an auto dialer and it calls mister burns and 0:05:26.880 --> 0:05:30.359 it says, hello, friend, would you trade one dollar for 0:05:30.440 --> 0:05:34.200 eternal happiness? And mister Burns, thinking he's talking to a person, 0:05:34.240 --> 0:05:35.680 thanks for when, and says, hmmm. 0:05:35.960 --> 0:05:38.120 One dollar for eternal happiness. 0:05:40.120 --> 0:05:42.760 May be happy with the dollar. I think about this 0:05:42.839 --> 0:05:45.000 all the time, Like I sometimes feel that way too, right, 0:05:46.040 --> 0:05:48.760 Like yeah, if you're a fron k goes down, but 0:05:48.880 --> 0:05:51.039 like all of your future needs will be met by 0:05:51.080 --> 0:05:52.880 like a little robot in your pocket, then you didn't 0:05:52.920 --> 0:05:54.280 need the four own k, but like you want the 0:05:54.320 --> 0:05:54.839 four oh and k. 0:05:55.800 --> 0:05:57.960 That's actually all I could think about on Monday was 0:05:57.960 --> 0:06:02.479 how I will ever financially recover from this trading session? 0:06:03.200 --> 0:06:05.360 Like we did not air, but I. 0:06:05.360 --> 0:06:07.320 Was like, no, I need to get to the podcast. 0:06:07.360 --> 0:06:13.440 You know, the show must go on. So you boldly 0:06:13.480 --> 0:06:16.400 asked the question in a money stuff this week, what 0:06:16.560 --> 0:06:21.240 if this Chinese quant hedge funds that you know then 0:06:21.279 --> 0:06:23.919 went on to develop deep Seek had a bunch of 0:06:23.920 --> 0:06:26.920 InVideo puts and made a bunch of money that way. 0:06:27.080 --> 0:06:30.120 And then friend of the show Bill Ackman tweeted the 0:06:30.160 --> 0:06:32.039 exact same question, which was fun. 0:06:32.480 --> 0:06:35.600 I assume that other people independently came to this idea 0:06:35.640 --> 0:06:37.960 I had several readers emailing about it before I wrote it. 0:06:38.040 --> 0:06:40.160 No, no, no, you were the first one to ever write 0:06:40.160 --> 0:06:43.440 about deep Seek, first one to ever pose the hypothetical. 0:06:43.560 --> 0:06:46.560 Sure, And I should also say that the idea of 0:06:46.600 --> 0:06:50.479 like disruptors funding their business by giving away their product 0:06:50.520 --> 0:06:53.080 for free and shorting incumbents. I learned of it from 0:06:53.160 --> 0:06:56.200 Joe Wisenthal or fellow Bloomberg podcast, like a decade ago. 0:06:56.440 --> 0:06:58.599 He was writing about it, you know, before everyone. But 0:06:58.880 --> 0:07:01.440 in any case, yeah, well it's a really funny idea. Two, 0:07:01.560 --> 0:07:04.400 it has really kind of percolated up. Apparently Howard Nick 0:07:04.480 --> 0:07:07.760 is getting questions about it in Congress. Oh my, whether 0:07:08.080 --> 0:07:11.160 that's there's no evidence for it, but it would be 0:07:11.320 --> 0:07:12.360 very funny if he did it. 0:07:12.960 --> 0:07:16.679 Also, is it illegal? I mean, is it insider trading? 0:07:17.840 --> 0:07:20.480 So here's what I'll say. One, this is not legal advice. 0:07:20.720 --> 0:07:23.480 Two I think in the US it's like clearly fine. 0:07:23.680 --> 0:07:25.960 The way it works in the US is like you 0:07:26.000 --> 0:07:28.080 are allowed to trade on your own information, but you're 0:07:28.080 --> 0:07:31.160 not allowed to misappropriate information from someone else. So like 0:07:31.480 --> 0:07:34.840 if he was buying in Vidia puts in his personal 0:07:34.840 --> 0:07:38.360 account while you know, running deep Seek, then that like 0:07:38.480 --> 0:07:40.960 might look bad depending on exactly what his arrangement was 0:07:41.000 --> 0:07:44.400 with deep Seek. But if like the corporate complex of 0:07:44.440 --> 0:07:47.160 like Deep Seek and the hedge fund was trading on 0:07:47.200 --> 0:07:50.600 deep Seek's own information to make a profit for that complex, 0:07:50.640 --> 0:07:52.560 and it seems fine, Like it seems like you're not 0:07:52.600 --> 0:07:55.360 misappropriating an information and you're really only trading on your 0:07:55.400 --> 0:07:57.720 own knowledge of your own business right and your own 0:07:58.120 --> 0:08:00.800 like extrapolation of what that will mean for other businesses. 0:08:00.920 --> 0:08:03.400 I think it's totally fine. Now. The two caveats that 0:08:03.480 --> 0:08:06.480 are one, I'm not sure that every jurisdiction sees it 0:08:06.520 --> 0:08:08.720 that way. The US law is much more about this 0:08:08.800 --> 0:08:11.480 like sort of misappropriation theory, whereas like in in other 0:08:11.520 --> 0:08:13.880 places it's like more common to just be like a 0:08:13.920 --> 0:08:16.280 fairness focused, like you can't trade on information that the 0:08:16.280 --> 0:08:19.480 market doesn't have. And then two, if you did this, 0:08:20.360 --> 0:08:24.040 people would get mad and they would say, oh, that's 0:08:24.080 --> 0:08:26.000 insider trading. And then you'd say, no, no, it's not inside 0:08:26.000 --> 0:08:27.840 of trading. It's fine. They're like, okay, fine, but it's 0:08:27.880 --> 0:08:31.240 market manipulation and like it's market univation because they'll find 0:08:31.240 --> 0:08:33.640 something you said wrong, right and so here in deep set, 0:08:33.640 --> 0:08:35.640 because of in a lot of controversy about like is 0:08:35.640 --> 0:08:37.719 it really the case that like its training cost was 0:08:37.760 --> 0:08:39.440 as low as it was, or was it like distilling 0:08:39.440 --> 0:08:42.160 models or rather AI companies. You know, you find something 0:08:42.160 --> 0:08:43.520 that you can seize on to be like, oh, this 0:08:43.559 --> 0:08:46.480 is a misrepresentation, and then it's like market mauntivation. You 0:08:46.480 --> 0:08:49.320 were like saying false things to bring down in video stock. 0:08:49.720 --> 0:08:52.520 So it would be a risky thing to do to 0:08:53.160 --> 0:08:55.800 tank in video stuck in such a high profile way 0:08:56.040 --> 0:08:58.120 while learning puts on it, like people would get mad 0:08:58.400 --> 0:09:00.320 and like might find a thing to try you with. 0:09:00.360 --> 0:09:01.880 But like, no, I don't think it's inside her jetty. 0:09:02.960 --> 0:09:05.640 Yeah, well, I mean again, this is all hypothetical. We 0:09:05.720 --> 0:09:07.000 have no idea if he actually did this. 0:09:07.000 --> 0:09:08.960 But I don't think he did. I just like it's fun. 0:09:09.080 --> 0:09:10.520 He should have done it. I would have done it, 0:09:11.320 --> 0:09:13.040 because like the other thing is like they don't like 0:09:13.040 --> 0:09:14.680 I don't know what their business model is, Like public 0:09:14.720 --> 0:09:17.160 reporting is like some combination of he's just you know, 0:09:17.200 --> 0:09:19.800 making money on his hedge fund, or he's doing this 0:09:19.880 --> 0:09:21.680 out of like the goodness of his heart, and like 0:09:21.720 --> 0:09:24.679 I want to wanting to contribute to AI research because 0:09:24.720 --> 0:09:27.040 it's like open source. It's cheap, how are they gonna 0:09:27.040 --> 0:09:28.520 make money? And I like get shortened in video. 0:09:29.120 --> 0:09:31.720 Yeah, I did also want to bring up that the 0:09:31.800 --> 0:09:35.000 narrative around this week has been that basically deep seek 0:09:35.120 --> 0:09:38.440 was able to recreate open ai with six million dollars 0:09:38.480 --> 0:09:41.520 and it's just been fun watching that get picked apart. 0:09:41.640 --> 0:09:44.840 I've spoken to so many invidiables in the past couple 0:09:44.880 --> 0:09:47.120 of days who have made this into a bowl case 0:09:47.160 --> 0:09:49.559 for in video, which I find really interesting. And there 0:09:49.600 --> 0:09:52.200 is a lot of skepticism around the numbers here. And 0:09:52.320 --> 0:09:55.520 one of the news stories that came about is that 0:09:55.679 --> 0:09:59.280 Microsoft and open ai are taking a look at whether 0:10:00.080 --> 0:10:04.520 this group used open AI's API to basically get a 0:10:04.640 --> 0:10:07.880 large amount of data, which would violate open AI's terms 0:10:07.880 --> 0:10:11.800 of service. Because open Ai, as we were reminded this week, 0:10:11.840 --> 0:10:16.880 isn't actually open source. It's closed source, but Metaslama is 0:10:17.080 --> 0:10:17.720 open source. 0:10:18.440 --> 0:10:20.319 Here's how I think about this. There are a lot 0:10:20.360 --> 0:10:24.679 of businesses that have been or like look likely to 0:10:24.679 --> 0:10:28.640 be really badly disrupted by AI. Right, Like there's some 0:10:28.800 --> 0:10:33.760 margin where like if you're an accountant, let's say podcaster, podcaster, Yes, 0:10:33.800 --> 0:10:36.240 if you're a podcaster, they're coming for you. And in 0:10:36.240 --> 0:10:39.760 the next few years, an AI will be able to 0:10:39.840 --> 0:10:42.320 sit here and talk into a microphone. I mean not literally, 0:10:42.360 --> 0:10:44.800 but like we'll be able to generate voice in a 0:10:44.800 --> 0:10:46.560 way that is better than I can do for a 0:10:46.600 --> 0:10:49.839 fraction of my very high price, and therefore I will 0:10:49.880 --> 0:10:51.599 be out of work. Right, And this is true like 0:10:51.640 --> 0:10:53.960 across a range of like sort of knowledge industries, right, 0:10:54.200 --> 0:10:59.920 And so AI is like undercutting a lot of people 0:11:00.240 --> 0:11:03.400 jobs in a way that like increases abundance for like 0:11:03.400 --> 0:11:05.839 the people who want to listen to podcasts or get 0:11:05.840 --> 0:11:09.160 accounting services. But then like undercuts like the earning potential 0:11:09.160 --> 0:11:11.800 of the people who are providing it cheap. AI is 0:11:11.920 --> 0:11:15.920 to ai what AI is to humans. Right, It's like 0:11:16.679 --> 0:11:18.640 open ai is like, ah, we can do your accounting 0:11:18.679 --> 0:11:20.839 for a tenth of the price of accountants and now 0:11:20.880 --> 0:11:22.199 deep things like we can do it for a tenth 0:11:22.200 --> 0:11:24.320 of the price of open ai. It's great. But like 0:11:24.320 --> 0:11:26.360 the other interesting thing is like you know you talk 0:11:26.440 --> 0:11:29.440 about like were they using open AI's APIs to train 0:11:29.520 --> 0:11:33.079 their model to like distill open AI's model, Like get 0:11:33.120 --> 0:11:35.240 a corpus of open ai outputs and you use that 0:11:35.280 --> 0:11:37.520 to train your your model. We use it to like 0:11:37.600 --> 0:11:39.520 to refine the training of your model. That to me 0:11:39.640 --> 0:11:44.280 is a little bit analogous to the complaints that publishers 0:11:44.320 --> 0:11:48.600 have about open ai using their data to train its models. Right. 0:11:48.880 --> 0:11:52.720 Ai synthesizes the output of like all these humans and 0:11:52.760 --> 0:11:54.880 then comes up with a thing that is that they 0:11:54.880 --> 0:11:57.440 can produce that sort of output more cheaply than humans can. 0:11:57.880 --> 0:11:59.720 And they're like, wow, what we were just like learning 0:11:59.760 --> 0:12:02.520 from humans. It's fine, and like you know, deep Seek 0:12:02.559 --> 0:12:05.000 kind of did that to open Ai maybe, like argue, yeah, 0:12:05.000 --> 0:12:07.839 like people, there's a suspicion which I think violence in 0:12:07.920 --> 0:12:10.480 terms of service, but it is like a certain product 0:12:10.600 --> 0:12:11.680 justice to it. Well. 0:12:11.720 --> 0:12:14.600 Deep Seek, for its part, says that it has distilled 0:12:14.640 --> 0:12:18.400 models for R one based on other open source systems, 0:12:18.440 --> 0:12:23.440 not necessarily on open Ai. But again, met Islama is 0:12:23.480 --> 0:12:27.280 open source. It's freely available for use, and I don't know, 0:12:27.320 --> 0:12:30.080 maybe that's not a bad thing. They're all just building 0:12:30.080 --> 0:12:32.160 on each other. That doesn't seem terrible, right. 0:12:32.280 --> 0:12:34.560 It's like like there's this notion that if deep seak 0:12:34.640 --> 0:12:36.439 is like piggybacking on the work of other models, then 0:12:36.520 --> 0:12:42.240 like then somehow the like baircase against like us AI 0:12:42.280 --> 0:12:45.800 infrastructure spending is like weaker, But I'm not sure that's true. Right, 0:12:45.880 --> 0:12:47.720 It's possible that it's just like the answer is that 0:12:47.760 --> 0:12:51.040 you've made it cheaper to scale AI because like you 0:12:51.120 --> 0:12:54.760 can build impressively on prior work and so you don't 0:12:54.800 --> 0:12:56.360 need all those data centers, don't it. 0:12:56.720 --> 0:13:00.320 I am curious what this means for this date of 0:13:00.360 --> 0:13:02.760 CAPEX spending when it comes to all these big tech giants, 0:13:02.840 --> 0:13:04.920 And we do have a lot of tech giants reporting 0:13:05.320 --> 0:13:07.280 this week, and you know it's going to come up 0:13:07.280 --> 0:13:09.920 on the earnings calls, which haven't happened yet as we're 0:13:09.920 --> 0:13:12.120 recording this podcast, but we're like. 0:13:12.040 --> 0:13:13.920 Two days out from like the US government like we're 0:13:13.920 --> 0:13:16.240 gonna spend a trillion dollars on AI Capex. 0:13:16.320 --> 0:13:18.480 Right, It's like you think about last week. 0:13:18.559 --> 0:13:20.120 I don't know, people are saying news driven like you 0:13:20.120 --> 0:13:22.439 could imagine everyone being like, ah, we're just getting out 0:13:22.440 --> 0:13:25.320 all the capex, Like can you imagine a really hard 0:13:25.360 --> 0:13:27.600 pivot in the next week. It just that doesn't seem 0:13:27.600 --> 0:13:28.280 that there's. 0:13:28.080 --> 0:13:31.920 A lot of awkward timing here because you think about Stargate, 0:13:31.920 --> 0:13:34.760 which was announced last week with open AI with soft 0:13:34.800 --> 0:13:38.160 Bank one hundred billion dollars Meta last week announced it 0:13:38.160 --> 0:13:41.560 was boosting its capex to up to sixty five billion dollars. 0:13:41.960 --> 0:13:44.840 Microsoft is spending eighty billion dollars. It's just felt like 0:13:44.840 --> 0:13:46.880 this arms race to see who can spend the most 0:13:46.920 --> 0:13:50.640 money on this and then to have everyone again get 0:13:50.720 --> 0:13:55.360 deep seaked on Monday was as someone with a four 0:13:55.360 --> 0:13:58.120 toh one K brutal but pretty fun to watch. 0:13:58.559 --> 0:14:00.720 I'm sorry this is rude, but it's such a soft 0:14:00.720 --> 0:14:02.880 thanks story. So I think announcing like we're gonna been 0:14:02.880 --> 0:14:06.440 ae hundred billion dollars on data centers like ten minutes before, 0:14:06.520 --> 0:14:08.920 like that becomes an obsolete thesis, Like yeah, that's. 0:14:08.760 --> 0:14:25.040 Like, that's a good story, a good story. 0:14:27.680 --> 0:14:29.520 Speaking of old times traits. 0:14:30.200 --> 0:14:33.680 Man, let's talk about X debt. It seems like it's 0:14:33.680 --> 0:14:36.520 getting a lot sweeter potentially. Well, this is also weird. 0:14:36.800 --> 0:14:38.240 I never know how to interpret. 0:14:37.920 --> 0:14:39.400 This because is this sweetener? 0:14:39.640 --> 0:14:43.600 So X twitter X, the company formerly known as Twitter 0:14:43.800 --> 0:14:47.360 Elion Mask bought it in twenty twenty two, and when 0:14:47.360 --> 0:14:49.400 he signed his deal to buy it, for a variety 0:14:49.400 --> 0:14:51.400 of reasons, it looked like a better deal than it 0:14:51.440 --> 0:14:53.760 turned out to be like months later, and so all 0:14:53.760 --> 0:14:55.960 these banks, led by Morgan Stanley agreed to provide them 0:14:55.960 --> 0:14:59.400 thirteen billion dollars of debt to buy Twitter, and by 0:14:59.440 --> 0:15:02.040 the time that closed, you know, ordinarily like they would 0:15:02.120 --> 0:15:04.680 sell that debt to investors before the deal closed, but 0:15:04.760 --> 0:15:07.680 for a variety of reasons, mainly that he was suing 0:15:07.680 --> 0:15:09.040 to try to stop the deal from closing to the 0:15:09.080 --> 0:15:12.120 last part and so wouldn't help out on the dead cell. 0:15:12.520 --> 0:15:14.120 For a variety of reasons, they didn't sell the debt 0:15:14.120 --> 0:15:15.960 before the deal closed. And by the time the deal closed, 0:15:16.360 --> 0:15:19.880 it looked bad, both because Twitter's business had deteriorated and 0:15:19.920 --> 0:15:23.080 because Elon Musk deteriorated it further, and because he was 0:15:23.120 --> 0:15:25.640 like talking smack about Twitter for the whole time he 0:15:25.760 --> 0:15:28.280 was trying not to buy it, and so like no 0:15:28.320 --> 0:15:32.400 one wanted the debt, they couldn't sell the debt, And 0:15:33.360 --> 0:15:37.600 there were occasional news stories being like they tried to 0:15:37.680 --> 0:15:40.520 sell the debt at like sixty cents on the dollar 0:15:41.040 --> 0:15:44.160 and they couldn't or whatever, like they got bids at six. Yeah, 0:15:44.360 --> 0:15:47.000 so like numbers like sixty cents on the dollar were 0:15:47.000 --> 0:15:49.800 floating around and now they're apparently outloading some of the 0:15:49.840 --> 0:15:52.240 debt and like numbers like ninety to ninety five cents 0:15:52.240 --> 0:15:55.440 on the dollar are floating around so pretty good. That's 0:15:55.440 --> 0:15:57.600 like not necessarily apples to apples, because it's like they're 0:15:57.680 --> 0:16:00.400 varying levels of seniority and it's possible that they couldn't 0:16:00.440 --> 0:16:02.400 sell the worst that at sixty and now they're signing 0:16:02.440 --> 0:16:03.280 the best that at ninety. 0:16:03.320 --> 0:16:07.600 But still so this Xai stake that Twitter has, I mean, 0:16:07.640 --> 0:16:10.520 I have questions. I was already questioning, you know who 0:16:10.600 --> 0:16:12.440 knows on pricing, but is that worth you know, up 0:16:12.440 --> 0:16:15.840 to thirty cents per bond or whatever, and whether it 0:16:16.000 --> 0:16:20.280 still is after of course deep Seek came about on Monday. 0:16:20.440 --> 0:16:23.440 After Elon Musk bought Twitter. The next thing quickly became 0:16:23.520 --> 0:16:25.400 large language models. Like when he signed the deal to 0:16:25.400 --> 0:16:28.000 by Twitter, like no one was talking about open ai 0:16:28.160 --> 0:16:31.840 or whatever. But shortly after the closing, like large language 0:16:31.880 --> 0:16:33.960 models were the thing. And so he started an AI 0:16:34.040 --> 0:16:37.720 company called Xai. It has the same first letter as 0:16:37.880 --> 0:16:41.560 Twitter does now, which is X and it clearly shared 0:16:41.600 --> 0:16:44.160 some resources with x and you know who like made 0:16:44.280 --> 0:16:48.920 enough noise about doing AI out of x that the 0:16:49.280 --> 0:16:53.240 upshot is that Xai is not just owned by him personally, 0:16:53.280 --> 0:16:55.440 and like people who put money into Xai and he's 0:16:55.480 --> 0:16:57.920 raised a lot of money for It's also partly owned 0:16:57.920 --> 0:17:01.800 by x by Twitter, So that company that he bought 0:17:02.200 --> 0:17:06.600 one of its assets is apparently a six billion dollar 0:17:06.640 --> 0:17:10.000 stake in Xai that's measured it it's like most recent evaluation, 0:17:10.040 --> 0:17:12.320 it's like fifty billion dollars. Now maybe that valuation has 0:17:12.359 --> 0:17:16.280 gone way down since Deepsek was released. There's not a 0:17:16.320 --> 0:17:19.880 lot of public market comps for like pure play AI companies, 0:17:20.080 --> 0:17:22.840 so like, who knows what the valuation is of Xai 0:17:23.040 --> 0:17:25.720 or open Ai or anything else. But you know, last 0:17:25.720 --> 0:17:28.159 time we checked, that stake was worth six billion dollars, 0:17:28.440 --> 0:17:29.879 which sure is worth a lot of money to the 0:17:29.880 --> 0:17:31.760 debt because the debt is like thirteen billion dollars, and 0:17:31.920 --> 0:17:35.159 like you know, if they have collateral, then you know 0:17:35.200 --> 0:17:37.160 that's worth you know, forty five cents on the dollar. 0:17:37.720 --> 0:17:41.040 So I think that like there was reporting that X's 0:17:41.160 --> 0:17:46.240 banks were shopping the debt, specifically telling people, hey, you'd 0:17:46.240 --> 0:17:49.320 have a priority claim on all these Xai shares. Isn't 0:17:49.359 --> 0:17:51.480 that nice? And what I wrote about this is like 0:17:52.119 --> 0:17:54.560 it's very hard to do a credit analysis of x 0:17:55.280 --> 0:17:59.240 Twitter because if Twitter doesn't pay interest on its debt, 0:17:59.560 --> 0:18:02.520 like the lenders can foreclose, but like what good does 0:18:02.600 --> 0:18:04.600 that do them? Right, Like Elon Musk can kind of 0:18:04.640 --> 0:18:07.560 trash Twitter on the way out the door, and Twitter 0:18:07.600 --> 0:18:09.480 seems like a hard like seems like a hard company 0:18:09.480 --> 0:18:11.119 to run before Elon Musk bought it, and now it's 0:18:11.119 --> 0:18:13.440 gotten even harder. So it's a company that has a 0:18:13.480 --> 0:18:16.040 lot of potential upside, particularly for Elon Musk, who can 0:18:16.119 --> 0:18:18.480 use it to like get presidents elected. But in terms 0:18:18.520 --> 0:18:20.760 of like downside production for lenders, it's like, I don't 0:18:20.760 --> 0:18:24.000 know whereas like you know, again a week ago, saying 0:18:24.280 --> 0:18:26.080 we own a big stake in an AI company seems 0:18:26.080 --> 0:18:29.760 like really great downside production because one AA companies are valuable. 0:18:29.960 --> 0:18:32.240 Two Elon Musk is like kind of unlikely to walk 0:18:32.240 --> 0:18:33.840 away from an AI company in a way that he 0:18:33.880 --> 0:18:36.480 tried to walk away from Twitter. And three people do 0:18:36.560 --> 0:18:39.640 credit analysis of AA companies where they did where they're like, oh, look, 0:18:39.680 --> 0:18:42.520 it has like all these like Nvidia chips. Even if 0:18:42.560 --> 0:18:44.920 like something goes horribly wrong at this company, those chips 0:18:44.920 --> 0:18:47.359 are super valuable because there's such an AI gold rush, 0:18:47.600 --> 0:18:50.439 So like there's really good collateral here right again, like 0:18:50.760 --> 0:18:52.960 that has been undermined by the events of the last week. 0:18:53.200 --> 0:18:55.480 But it does feel like the Xai stake was pretty 0:18:55.480 --> 0:18:57.480 credit enhancing for the Twitter bonds. 0:18:57.800 --> 0:19:00.440 Yeah, and I mean you say that el Musk is 0:19:00.520 --> 0:19:02.639 less likely to walk away from the Ai company in 0:19:02.680 --> 0:19:05.440 the same way he might be tempted to walk away 0:19:05.440 --> 0:19:08.359 from Twitter. I mean, how confident are you when you 0:19:08.359 --> 0:19:11.680 say that? Because who really knows, Matt. 0:19:11.880 --> 0:19:15.560 Yeah, nobody knows anything. But like Elon Musk is kind 0:19:15.560 --> 0:19:18.760 of always sad, including when he was buying it, that 0:19:19.359 --> 0:19:22.119 Twitter was not a great business decision, but yeah, he 0:19:22.160 --> 0:19:23.680 thought it was important for the future of the world. 0:19:23.680 --> 0:19:25.240 Blah blah, blah blah. And like that, by the way, 0:19:25.280 --> 0:19:27.560 turned out to be right. Like he's had a ton 0:19:27.600 --> 0:19:30.320 of political influence with Twitter without it necessarily making him 0:19:30.320 --> 0:19:32.680