SKEPTIC’S GUIDE TO INVESTING
Straight Talk for All, Nonsense for None
About - Our podcast looks to help improve investing IQ. We share 15-30 minutes on finance, market and investment ideas. We bring experience and empathy to the complex process of financial wellness. Every journey is unique, so we look for ways our insights can help listeners. Also, we want to have fun😎
Your Hosts - Meet Steve Davenport, CFA and Clem Miller, CFA as they discus the latest in news, markets and investments. They each bring over 25 years in the investment industry to their discussions. Steve brings a domestic stock and quantitative emphasis, Clem has a more fundamental and international perspective. They hope to bring experience, honesty and humility to these podcasts. There are a lot of acronyms and financial terms which confuse more than they help. There are many entertainers versus analysts promoting get rich quick ideas. Let’s cut through the nonsense with straight talk!
Disclaimer - These podcasts are not intended as investment advice. Individuals please consult your own investment, tax and legal advisors. They provide these insights for educational purposes only.
SKEPTIC’S GUIDE TO INVESTING
DeepSeek: China’s AI Disruptor (Part 1)
Please text and tell us what you like
Can a Chinese company with a modest $6 million investment truly disrupt the global AI landscape and challenge tech giants like Google and OpenAI? Join us as we explore the audacious rise of DeepSeek, founded by hedge fund manager Liang Wenfang, and its implications on international markets. We unravel the layers of this ambitious venture, probing the complexities of U.S. concerns over Chinese competition, and the real costs of AI development. Discover the intricacies of how export controls, such as NVIDIA's maneuvers in the Chinese market, play a crucial role in this unfolding narrative. The chessboard of AI competition is evolving rapidly, and DeepSeek might just be a game-changer.
As the AI world grapples with issues of transparency, secrecy, and intense rivalry, we're witnessing intriguing shifts in technology and strategies. DeepSeek's open-source approach contrasts sharply with the proprietary pathways of American giants, raising questions about transparency and financial motives. Meanwhile, as a new AI app overtakes ChatGPT in popularity on Apple devices, international scrutiny intensifies. With Italy's investigation into potential security risks, we dissect the potential implications on global data security. This episode is a thorough examination of the geopolitical and economic tensions at play in the rapidly transforming AI sector. Listen in as we unpack the interconnections of tech innovation, market strategy, and national security.
Straight Talk for All - Nonsense for None
Please check out our other podcasts:
https://skepticsguidetoinvesting.buzzsprout.com
Hello everyone and welcome to Skeptic's Guide to Investing. I'm Steve Davenport and I'm here with Clem Miller, and today we're going to seek to explain the pluses and DeepSeek of . And that may sound like a lot of gibberish, but, as we saw on Monday, the the markets reacted when this Chinese company, which many people, including us, I don't think, had ever heard of, suddenly was disrupting the ai industry. I'm going to try to give you a little bit of background, and we've divided this into two podcasts one at a high level who is deep seeking, what is it doing? And then at a lower level what are the actual bits and bytes and how will investors respond and what should you do? This is going to be a lot of fun because I think it's uh. In the last two days we've seen people getting online and going to the apple store and downloading this and, eventually they had to stop downloading it and there was an outage yesterday with a supposed cyber attack, but some people say say that the company and its computing power just couldn't handle all of the queries. So, as with most things involving a Chinese company, there's always going to be a little bit of uncertainty as to how transparent or I guess I'd say how truthful some of the claims are, so I'm going to give you, you know, the 10,000 foot view of what is DeepSeek.
Steve Davenport:Deepseek was founded by a hedge fund manager (Fund Name - High Flyer) named Liang Wenfang, and Liang is the founder of this company.
Steve Davenport:He used AI to help him manage his hedge fund called High Flyer, and so, as you can imagine, High Flyer was making decisions based on some of the models that were built on DeepSeek, and DeepSeek then said we're going to now create a company in 2023 that is going to focus on other applications besides the hedge fund, besides the hedge fund. So when we look at how quickly something like this could go, in a year and a half, into the national and international space and be a disruptor, I think under the Wikipedia definition of disruptor, we should have a little picture of DeepSeek, because I think that's really the example, and I think that we'll get into some of the questions that are around for people about how our chip bands in NVIDIA were implemented in China and how many chips they might have and how many chips they might not have. It's really a pretty interesting story. But if I was to ask you, Clem is DeepSeek for real and does it matter? Where do you stand on DeepSeek?
Clem Miller:So, yes, it's consequential, there's no question about that. It's consequential in several ways, very concerned about, and companies in the United States were very concerned about, which was the emergence of Chinese competition in the AI space. And so that's what we're seeing here is we're seeing the first of what could be many Chinese competitors in the in the generative AI space or, to be even more specific, the so-called reasoning space, which is thought to be the precursor to what everybody in this field refers to as AGI, which refers to generalized intelligence, which refers to generalized intelligence, so not just sort of chat, gpt, but sort of more generalized intelligence, which I guess is a. I don't think that I've ever seen a really good definition for general intelligence, other than that it's supposed to be more human-like's. You know what when you see it, I guess Right.
Steve Davenport:Yeah, like pornography.
Clem Miller:Right, yeah, exactly so. So I think so, yeah, it's real, it's a, it's an, it's evidence of Chinese competition. I think that it shows, you know, from a geopolitical standpoint, it shows the limitations of export controls initially restricted NVIDIA's GPU chips from going into China, that NVIDIA worked on a China-specific chip that removed certain characteristics but then provided other characteristics that weren't restricted but that gave the chips it was selling to China just as much processing power as the prior chips that it sold. So you know, I think that you know, it shows the limitations of using export controls, especially if you have companies that like NVIDIA, which were honoring the letter of the law but reportedly, I guess, not the spirit of the law in terms of providing those chips into China.
Steve Davenport:Yeah, I was looking at one piece that said that they could have anywhere from 10,000 to 50,000 chips from NVIDIA the H800, that are being used in this application.
Clem Miller:Right, that's what I'm referring to. You know, the reporting out of China, which really led to the market route on Monday, was that the creation of this R1, reasoning 1 model by DeepSeek was done at a cost of $6 million, and whereas, you know, it costs an awful lot for other companies to build their models. So the question really is one of apples and oranges, right. Is this $6 million, you know, is that an apples to apples comparison or is it an apples to oranges comparison? And so let me explain that.
Clem Miller:You know that $6 million a lot of people now are thinking, well, that $6 million was just the cost, sort of the incremental cost of building the model and didn't include the cost of buying all the chips and hardware and infrastructure and so on, which were already sort of sunk costs.
Clem Miller:So when you talk about the cost of building a model for a company like Google or OpenAI or other companies here in the US, are they just counting the cost of the models or are they doing an all-in calculation that includes the cost of the semiconductors as well as well?
Clem Miller:I'm sure they're all looking at that and it may be the case that this you know this this statement about $6 million cost is really, you know, sort of misinterpreted and that the cost of deep seek, of deep seek, building this thing was actually a heck of a lot more. It's just that you know the, you know, given scaling uh, which applies to, you know, all these companies you know you're able, you know, as you build, scale, you're able to generate more and more models at cheaper and cheaper costs, and that's probably where the $6 million figure is coming from. And so, as companies and as investors began to question that $6 million number and whether it was apples and oranges. That's why we've seen some recovery not full, but some recovery in NVIDIA and some of these other stocks during the week.
Steve Davenport:Yeah, another example I heard was it's similar to drug development and that you can look and say how much did it cost you to develop drug a? But you developed 10 other versions that were cancelled somewhere along the the trial period and you have to include all 10 in your total cost to develop drug A, because you learned from each one of those what worked and what didn't work and you made adjustments, and so they probably didn't go from having no capability to having a finished product in one trial and one $6 million expenditure. So it's very hard to look at the overall development cost and you bring up chips as one example. But I also heard it's like you can't expect that the programmers didn't have versions A, b, c, d, e, f before they came to the version that they currently are at. So I think it's.
Steve Davenport:You know, it's interesting to me how, when we look at this weekend, you know, mark Andreessen, who was one of the founders for Microsoft, you know, was one of the people who said this model looks real and this thing is significant, and I just kind of wonder are we all just getting a little bit ahead of ourselves here?
Steve Davenport:You know, on Monday I heard that this was a move by Xi to send a message to Trump that China was real and China was going to be a competitor and therefore he didn't like what he did with Stargate the week before. So when Trump makes a big appearance and says, you know, we're going to crush the rest of the world, do you think that Xi is sitting there saying, okay, everybody, let's go figure out how we can say that we're the same or better. So I, I really think that that I mean, I don't believe. Do you believe that there are people sitting in the white house and sitting in China that are sitting there going, okay, we've got to respond right now with a way to hurt the US markets and hurt these US companies, because we don't want the world to think that we're not a player?
Clem Miller:so you know, my thought on that, steve, is that is that china's main interest is in trying to develop its military and if it's going to do things to enhance its military capabilities, it's really not going to publicize them and we may not hear a lot about that uh, in um, you know, in public fora, right, I'm sure the intelligence agencies know a lot more about this, uh, but you know we're not going to hear about that. So I think that, you know, by deep seek publicizing this and coming out, I think that's, you know, I think that's that is that was an effort to embarrass the Trump administration and you know, in light of the the Stargate issue, and clearly Trump feels threatened by by deep seek because he came out I think he came out today or yesterday and said that, uh, that this might reflect sort of stolen information or stolen technology and that he's looking into that. So he clearly feels embarrassed by what the Chinese are doing. But I think, ultimately, I don't think DeepSeek itself is connected into, necessarily connected into military applications. I think that the military applications of AI within China are probably kept quite close to the vest, highly classified, you know, within the Chinese system.
Clem Miller:So that's my, that's my guess on that, but I think that I think that you know is this I mean, some people were saying, oh, this is a wake up call for US. I'm not sure that it is. I mean, think about it this way are thousands and thousands of AI scientists, computer scientists, in the US who are working on AI right now, thousands and thousands. Don't you think that some of them might have caught on to some of this? You know, ai, light technology that that deep seek has come up with. You would think so, right?
Steve Davenport:Yeah, I was going to go two places One. My first comment is that it's very unusual for US businesses to react so much. When you look at DeepSeek, it's an open source code. So here we have in the US, Google, Apple, Meta all going after their own versions of AI, and in the US we pursue all these business apps because we want one of the businesses to control it. And here, in a communist country, we've got an application that's developed open source and that has no commercial applications as one of the companies that we're threatened by.
Steve Davenport:So we're not threatened by any of the professional or the I don't want to say professional any of the corporate versions of China's AI. We're threatened by something that was developed open source, so isn't the US the? Market that's more open and China's the market that's more closed. How do I?
Clem Miller:rationalize that. Well, you got a lot in that comment, steve Allve, so all right, so so break it down for me.
Clem Miller:So so, first of all, uh, let me break out my comment into two, two pieces. First piece is um, china's a very secretive country and I don't think that we can necessarily, we can necessarily truly have faith in what might be reported out of China. Now, I say that with respect primarily to what the government reports in terms of things like GDP and so on, but also I think that you know it also applies to what companies report. You know you've had, you know, various, you know short seller companies go in and and, uh, take apart, uh, some of the reporting by, uh, by companies there. So I, you know it could be that some of these claims, especially on the cost side, as we were talking about, are overblown. So there is a secrecy element to this.
Steve Davenport:Well, I'm skeptical about anybody who says I'm a hedge fund and I want to share my software with the rest of the world, right, well, so so yeah. When he is go ahead?
Clem Miller:No, that's, but you know the question is is this the hedge fund sharing their software? Or is this the hedge fund developing a product, you know, as a sort of a project holding that it's trying to eventually market Even so? The second point I wanted because, keep in mind, this hedge fund is sort of like Renaissance Technologies in the US. They weren't going, they would never share their own technology, right, but maybe they would try to market, starting with open source, the technology of they're developing as a commercialized product, which is what I think is going on. But the second point I wanted to make, steve, before we go on, is that even open AI, as the term open implies, they have open source as well. So they provide their, you know, they provide open source code to allow.
Clem Miller:You know, they start off, at least as a company that was sort of dedicated to advancing AI in general with the idea that you know eventually, by getting people hyped up on AI, that they would be able to make money in the long term. And so some of their products, some of the open AI products, were not commercialized, rather they were open source. But as they moved along and they became more commercialized, you could have more and more comprehensive versions of the OpenAI Comprehensive, deeper, more sophisticated let's call it that way deeper, more sophisticated let's call it that way more sophisticated versions of the OpenAI models that they could commercialize and sell. So it's almost like you know, when you have software where there's a free version and there's a premium version, right, that's what OpenAI has and maybe that's what DeepSeq was you know was trying to create, but you know.
Steve Davenport:Yeah, I go back to something a little more cynical or a little more skeptical, which is I'm a hedge fund manager and I'm trying to generate returns and on the one hand, I have an asset which is my model for coming up with trades, and that model has given me some value. Then I say here's this code and others can use it for other applications, which doesn't seem like something a hedge fund manager would do. But when I look at this and I see the reaction on Monday, my question to you, clem, is do they have any shorts on NVIDIA, asml, taiwan, semi? I'm not sure how transparent hedge funds are in China. I know how transparent they are in the US. So did High Flyer make money on Monday on shorts in the US companies? I have no idea. Or are they just going down the wrong path?
Clem Miller:Because it's not about the money, Steve.
Clem Miller:It's about the technology. I think you have an excellent point. I think it's a good question. I don't have an answer to that. You know, in terms of shorts, I don't have access to who's shorting who's shorting what. I just know what the shorts are. Short interest is with and with a two week lag, which is the the normal lag in the reporting of shorts, because I think, as as our listeners know, I use short interest as a as a tool in my decision making with respect to stocks and a lot of these, a lot of AI related. Well, some, some AI related companies, like NVIDIA, have pretty low short interest. Others have somewhat higher short interest, right.
Steve Davenport:Others have somewhat higher short interests, right. But my only point being one we've got an issue with a Chinese hedge fund, which I don't put a scale together that measures the transparency and regulatory diligence of hedge fund A in the US to hedge fund B in China, in the US to hedge fund B in China. But I'm just going to try to make a wild, a wild ass guess here that says you know, I think that there's probably a little less supervision going on in the Chinese hedge fund than there are in the US hedge fund.
Clem Miller:So is that fair? That is, I mean I would. I have to believe, I have to believe that high flyer, which is the name of this hedge fund, made money on what happened Monday.
Steve Davenport:Okay, yeah. So my question to you would be if you're going to make money from that trade, would you set it up such that you get an endorsement? The other thing is, clem, I find this whole wow, I just learned about this this is so shocking. You know, like the guy in Casablanca, I'm shocked that gambling is going on here. This software, in May, when they released their first version that had some reasoning in it, that had some reasoning in it, was rated by the University of Waterloo in Toronto as the seventh best chat GPT, you know, chat model in the world. So somebody in May had gone through and evaluated all of these different models and theirs came out seventh models and theirs came out seventh. So something tells me that you know the people at Meta and the people at Google.
Steve Davenport:I know you talk about OpenAI, but our listeners should understand OpenAI is one participant in this space, it is not the whole space. All of these companies have already been using and implementing. I mean the numbers from Facebook were just enormous because people believe that, hey, they were able to have more advertising and more revenues and about 30% less cost. 30% less cost. And how does that happen to a company like this? Well, if they start to use AI and automate some of the decision making that goes on, then they can operate a lot more effectively. So we are already seeing AI being used by commercial companies in the US to improve their operations and improve their throughput.
Steve Davenport:So I'm sitting there saying it seems like we've had something going on here for a while. But I mean, is it like Silicon Valley Bank, where analysts and everyone follow these companies and nobody mentions DeepSeek? And? And then all of a sudden, deepseek is a major competitor? How does the analyst community decide what to talk about or not talk about? And I'd hate to say this, glenn, but would analysts not cover it because they believe that China isn't worth worrying about as a competitor? I mean, is there that kind of arrogance in the US technology space? I mean, if they have a product, why are we just hearing about it now?
Clem Miller:Does that ever-? So, Steve, let me suggest that's a great question and there's a lot to that.
Steve Davenport:So let me suggest that we cover that in our second part of our of our discussion um, so this other questions I have is how could such a small competitor have such a big influence? Yeah, you know, yesterday we saw 500 and or monday we saw $593 billion get taken out of the major AI players in the US market. We saw a trillion dollar correction on Monday, so a trillion dollars of market cap went away from a small open source Chinese company. I mean, do we believe that the 400 or 500 names that are going to use AI in the US marketplace let's just say the S&P 500, is an S&P 500 company going to take the software from China and base their system on it?
Clem Miller:Of course not.
Steve Davenport:Well, I mean, well then, let me go to the next point. Then, Clem, if you answer that quickly, Well then, why did the market react?
Clem Miller:the way it did.
Steve Davenport:How do we have that kind of reaction if a reasonable mind says this is a small company, this isn't going to be able to translate over into the US market?
Clem Miller:Because people in these companies, in US companies, especially those that are pursuing AI and the power infrastructure companies that are involved in providing power for data centers, see a cheap model, a cheap to run model, cheap to develop, cheap to run model as a threat to their business modes. And we can go more into that in our second episode.
Steve Davenport:So we look at again. I'm looking at different ideas here about potential things that aren't being said as well as are being said. And on Friday, meta invested $65 billion in ordering more chips. And so if there's a cheaper version out there that uses one-tenth the power and has a cost to develop of one-tenth or one-fifteenth the similar development in the US, why would Meta go about trying to get more chips if they can get by with one-tenth using the algorithms from DeepSea?
Steve Davenport:Wouldn't the demand for chips go down? No, because of something called the Jevons paradox.
Clem Miller:Oh, that sounds interesting. Is that a recent technology development? No, it actually dates back to the steam engine actually. But what the Jevons paradox tells us is that as a particular technology gets cheaper, more efficient, there will be greater demand for that particular technology. So you might think, okay, well, you know, if steam engines get more efficient, then what will happen is there'll be less demand for steam engines.
Clem Miller:At least some might think that. But when you look back, you realize that as steam engine technology got cheaper, there was more and more demand for steam engines, and those who were producing steam engines were able to sell more steam engines. So the cost of each steam engine, the price of each steam engine, might have been lower, but the volume of steam engines they were able to produce was a lot higher. And so that's what we're dealing with here, with regard to both semiconductors as well as with demand for AI models, in that as development costs, as chip costs, get lower, as model development costs get lower, then there's going to be more demand for chips, more demand for models, and this will lead to higher revenues for the companies involved in these industries, even though the prices they may charge will be a lot lower.
Steve Davenport:Yeah, I mean I think that when we think about you know how much we can see the opportunity here. I mean I liked the line that said that this is an AI 400 companies, that now something like this makes it available to the top 4,000 companies. It's about the floor, not the ceiling, and so by this raising the floor and the access to the floor, everybody can potentially benefit and therefore there can be a lot more usage of AI, because it's not going to be this huge technology, energy and GPU hog that makes only the biggest players able to play on the field. So if we have more players on the field, then are we going to have more chaos or more well-organized implementation?
Clem Miller:So, again, a lot of answers to that question.
Steve Davenport:Okay, so I guess, we're getting close to the end of the first why?
Clem Miller:don't we move on to the second episode and a lot more about all of this stuff?
Steve Davenport:Sure, my my one thing that's at a national level is.
Steve Davenport:It's interesting is that we're already talking about whether this can be controlled, like TikTok, and whether it will be banned. It was. It just passed in the three or four days chat GPT and downloads on Apple, so more people downloaded it than downloaded chat GPT. So Italy has already said they're doing an investigation as to whether this should be banned in their country because of the questions of security, and whether this should be banned in their country because of the questions of security, and whether China will be able to take anybody who downloads it and potentially use the information from those downloaded individuals to build its database and build the strength of its model. So I think that when we look at strengthening China's models in AI, I'm not sure that's on any regulators to-do list. So I appreciate everybody listening and we look forward to talking more about this in our second episode and everyone talking more about this in our second episode. And everyone please follow, please share and please let others know about what we're doing here at Skeptic's Guide to Investing.
