Stocks lost $2 trillion. The market is wrong about AI

0:00 spk_0

I just watched $2 trillion get wiped out from the market in 40 minutes. This is the worst drawdown that we’ve seen in the market since the financial crisis of 2008, and we’re technically around 3% away from entering an official bear market. And there are 3 reasons why this is happening. One, people think that AI stocks and the AI market in general is massively overvalued. Two, people are asking for the ROI of AI and they don’t.Don’t seem to be seeing it. And 3, most recently, we had elections in the US which meant that almost 45% of planned data centers, you know, the thing that props up this entire AI bubble, is probably not going to go ahead next year, which means all the exciting numbers we’ve been seeing, all the revenue that’s been generated, all these crazy IPO evaluations that we’re expecting from the likes of Anthropic and OpenAI will likely be slowed down because a lot of it was.Based on bringing all these expensive GPUs online to serve these AI models. But there is something distinctively different about this drawdown, and that is the money isn’t being taken out of the market. It’s being rotated to companies that aren’t really affiliated with AI. It’s a flight to safety, if you want to call it that. But there’s also something weirder, which is classic tech companies which have been very heavily involved in the AI cap.Stuff such as Amazon, Google, Microsoft, whose stock has actually suffered a beating over the last couple of weeks, have rebounded, and my guess is probably because they have, you know, a very diverse and successful line of businesses that aren’t AI adjacent, that still generate billions of dollars of revenue. But I have an even hotter take. I think the market’s just wrong. I think it’s exaggerated and I think it’s over.Drawn. When you think about it, when you step back, the demand for AI has never stopped going up. In fact, it’s even accelerated going up. We’ve seen CPUs and GPUs that are 10 years old, being rented out for triple the value than when they initially debuted, and we’re seeing the increasing demand for AI models and tokens just go up only. Just take a look at any of the revenue projections that we’ve seen over the last 3 months. And unless you’reCompletely bearish about AI having any real economic impact in the world over the next 3 to 10 years. You’ll look at what’s happening in the markets, you’ll see these massive drawdowns that are happening seemingly irrationally over a very short period of time. You might think to yourself, maybe now is a good time to get involved. Now, I have to preface this with saying that this isn’t financial advice. This is just me talking about the markets, markets of which I’m involved in and stare at every single day. Unfortunately, unfortunately.And I have thoughts about it. So on today’s episode, I’m gonna walk you through everything that’s happening in the markets and a few other topics, but also give my reasoning as to why I think the market is explicitly wrong right now. So yeah, let’s get into it on today’s episode brought to you by Q Qualcomm.So let’s talk about the last 4 to 6 weeks of the market where October was basically a wipeout for the market. We seem to have reached breaking point for AI specifically, where market investors just decided that the stocks weren’t worth the value that they were, and indeed, I would be lying if I didn’t say I was thinking sort of the same. Things had just been going up successfully for the last 12 to 18 months. I would say the skepticism probably started around.Around the SpaceX IPO when that valuation went insane, and then people started to get a little bit uneasy with some of the AI memory stocks going up. And so over the last 4 weeks, we’ve seen kind of like a deflation in the bubble, but especially over the last 7 to 14 days, this has turned into a bloodbath. I’ll use the example of Intel, which has kind of been made an example of down almost 50% from literally a few.Weeks ago. And this is crazy to me because Intel has kind of been grounded as the TSMC, that being the Taiwanese company that helps Nvidia build all its very expensive and important GPUs of the US. So Intel would be that, but for the US, and the reason why this is important is this is becoming a geopolitical thing, right? You want some of the core infrastructure to be brought from offshore to onshore, and we’ve been seeing that kind of trend develop over the lastI would say a couple of years since the Trump presidency. So the idea here is Intel was kind of being proposed as a major company, especially for national security, and it went to the extent that the government had actually purchased around 10% of the entire company to.See that crash 50% over a matter of weeks, despite really good earnings reports, is shocking to me, and that sort of snowballed into a trend for that happening to pretty much any and every company out there. If you look at Google, Amazon, and Microsoft earnings that happened, I think around,I don’t know, 10 to 14 days ago, immediately after they reported record earnings, their stock crashed around 7 to 12%. So the idea here is the market, despite the good news, doesn’t really believe that number can go up for any longer. And so the straw that broke the camel’s back was what happened on Tuesday of this week, where we got the election results in the US, which is the midterm elections, and it went overwhelmingly blue. Now, without getting political here, the point of that.A lot of the new leaders that have been elected in these new townships, cities and states are basically very anti-AI. They’re very anti-AI data centers specifically. So all those hundreds of billions of dollars’ worth of GPUs that we’ve been standing up or trying to stand up are likely going to get paused and delayed. And it’s worth pointing out that this isn’t necessarily a political thing. This is representative of what the entire country was thinking at the time or currently, right? So they were voting.On the basis of thinking that AI would do more harm than good, I think the stat was like 67% of people that were polled thought that. But it goes further than that where an overwhelming majority of individuals also thought that having an AI data center in their town just wouldn’t help them, it would increase their electricity bills, increase their water bills, so they didn’t want to get involved in that. So we’ve reached sort of this breaking point where it’s like, OK, enough with the AI stuff, enough with paying hundreds of dollars worth of chat GBT or.To subscriptions. I just want to slow things down. I want to stop laying people off, and it’s time to go back into real companies that don’t rely so much on such a heavily inflated tech sector. That’s what I’m guessing the opinions were. But if you’ve watched this show for any length of time, you’ll know that overall, I lean extremely bullish and optimistic as to where this tech is going. I think fundamentally, AI is actually a very unique technology becauseYou can kind of take the technology and throw it at literally any sector, from science, medical fields to back office, back-end admin systems, and it’ll be able to conjure up something pretty magical. We’ve seen it over recent stories with it’s solving crazy mathematical problems and also automating a bunch of accounting work. So if the directional bet is that AI is going to change these industries for the better, whether it’s inflated evaluations over the short-term,Then you would probably think that some of these drawdowns are starting to get a little bit exaggerated, and listen, don’t take my word for it, let me walk you through my kind of thoughts on some of these companies. If we pull up this screenshot from Bull Theory, which is basically like representative of what the bloodbath looks like right now, let’s walk through some of these companies. Start off with Nvidia. Nvidia used to be one of the most valuable companies in the world. Now I think it’s like #2 or maybe even #3. Apple has.Seated it, right, and I’ll get to Apple in a second, but the idea here is that it’s down, I think roughly, I don’t know, 5 to 7% over the last week, and Nvidia’s typically been more stable or more positive in general, right? Because it’s the bedrock of this entire AI sector. If you don’t have Nvidia producing these GPUs, you don’t have anthropic, you don’t have OpenAir, you don’t have any of the hyperscalers making the most money that they’ve ever made every single quarter because you need GP.Used to train, inference and build the models in the first place. So the idea here is Nvidia’s kind of the bedrock and seeing it draw down that much, it may seem like a small percentage, but at their market cap, that’s a lot of money being wiped out. Doesn’t make sense to me because if you’re optimistic and your long-term bet is that AI is going to proliferate and change things, maybe not as rapidly as we projected or maybe not as impactful as we originally estimated.You would still bet that over time this will eventually happen, maybe just a little more slowly. So you would bet that, OK, we’ll still need some of these GPUs. And actually if you look at the demand for these GPUs, there’s two data points that are completely irrefutable, especially if you’re taking a bearish view of this. Number one is the value of these GPUs. I’m talking about theGPUs, which a lot of people say are going to depreciate really quickly. There’s 6 to 10 year old GPUs that are currently worth 3X, 2 to 3X more than they originally did when they debuted. This is the cost to rent them out and the price to rent these things out have gone up on order of around, I think it’s like 60 to 70% since July.So, prior to July, we were seeing that trend, but people were kind of like, including myself, were like, hm, I don’t know if this is sustainable. We’re still seeing it, despite the crashes that we’re seeing around us. People still want these GPUs. But the second data point is the most telling, which is the cost of tokens. So tokens are the things that you,Need to, they’re kind of like pennies in a slot machine, you need it to kind of get the AI going to use the AI models. And so you look at the value of those tokens and if it crashes, you could argue, hmm, you’re not gonna earn as much money as you want for AI products and GPUs in general. It’s not what we’ve seen at all. The cost of these tokens have plummeted to around I think 58 to 60%.Despite that, these GPU rental prices are still going up. So the point I’m trying to make here is, if you look at a very fair market valuation, like not what Nvidia prices their GPUs at, not what anyone else sells their GPUs at, just a fair market, secondary market, re-rental for these GPUs, old GPUs, demand still massively outpaces it, and this is happening during one of the worst market drawdowns ever, so if you look at Nvidia, I’m just kind of looking at that and I’m like, well.It might be a buy here. Now, let’s take some other companies here. Uh, let’s see, let’s, let’s go with memory stocks. I think this is a good one to kind of like level my thinking here. Memory stocks in general have seen a crazy price increase over the last 6 to 9 months, especially when people started to realize that there’s only really 2 to 3 major manufacturers that will be able to supply high bandwidth memory or HPM.For short, that is needed to build these GPUs. If you ask or have a conversation with Jensen Huang, he’ll tell you, yeah, the most costly material that I need to buy to build my GPUs. The reason why the prices of the GPUs are going up quite a bit is because memory costs so much. Memory, I think 3 months ago cost like 50% of the bill of materials to build a GPU, which is absolutely insane becauseLiterally 6 to 9 months ago, it was like a fraction of that. So the point being is memory has become a very scarce commodity. But it’s not gonna remain scarce for long, and that’s what investors are seeing here, which is like, listen, the memory stocks can’t go up much more because we’ve already priced in crazy revenue for the next 3 years, and I’m not exaggerating, by the way. If you talk to any of these memory manufacturers, they’re just gonna tell you.Yeah, we’ve sold out supply till the end of 2029. OK, so let me put that into my equation of valuing these stocks, and that’s why we’ve seen such a meteoric rise in these stocks. But the point that’s being made here is maybe it’s a little too high right now. Where do we go from three years from now? We can’t even project that far ahead, and what? What if some form of new AI architecture gets released over the next, I don’t know, year or so, that doesn’t require that much memory, or if the memory supplies are able to somehow increase their.Capacity of memory and flood the market with more supply and the price of memory goes down. What happens then? That is kind of like what I see when I look at the market reaction to things like Micron, SK Hein and Samsung. Now let’s talk about Intel and AMD. These are the ones that annoy me the most because in my opinion, it doesn’t make sense. Now, picture yourself as the US that wants to own AI and so far we have been the leaders or the Western hemisphere in general.Been the leading sector here for AI. One thing has gone absolutely viral over the last, I wanna say, 6 to maybe 8 months, and that is something known as AI agents. And there’s been a bunch of products that have gone super viral recently, such as Metamuse, you’ve got Claude Duett, and you’ve also got OpenAI Dots, and of course, Crockbot. Now if you’re listening to this and you haven’t used any of those things, don’t worry. The point that I make around my whole thesis around personal super.Agents is that you will have an AI agent that will be personal to you. It’ll be kind of like Jarvis from Iron Man. It’ll be able to go out there and make your life more productive or free you up to go do things that you want to do that you don’t need to explain to an AI. It gets rid of all the things that you just don’t want to focus on and do. It automates all the boring stuff away. Now, directionally, if you bet that every single person could deem or gain value from something like this, which I think is a veryConservative estimate or take, then in order to power these agents, you don’t just need GPUs, you need something known as a CPU. And let me tell you, there’s only two major manufacturers of CPUs, AMD.And Intel, yeah, that same Intel whose stock has crashed over 50%. And what’s crazy about this is if you look at the ratio between number of GPUs you need for the number of CPUs that you need to just, you know,Let all of these AI models function, you’ll see a massive balance in favor of CPUs recently to the point where it’s currently 1 to 1, but it’s soon to be 2 to 12 CPUs for every GPU. So, without getting technical, the point I’m making here is you’re gonna need a lot more CPUs. And so that revenue line has strictly not been taken into account here.You look at Intel and AMD. Listen, I’m not saying that they weren’t overvalued, but a 50% decline and AMD, I’m looking at the screen right now, crashing around 13% just seems very unfeasible to me, especially when both of those companies are US based and we have like the momentum going right now. But for the skeptics out there that are still saying, Ejaz, I think you’re taking too much of an optimistic view.What’s actually happening is token cost is going down, fewer people are gonna be using these models, or they’re gonna be more careful about how they use these models, and they’re not necessarily going to be using anthropic cloud or chat GBT of which the IPOs are dependent on this. It’s just all a marketing thing. People are gonna be going to more open source, they’re gonna be running models that are more private, etc. etc. That’s not gonna generate as much.Revenue. OK, well then, why is Palantir and other such companies who have been building products that would allow all these companies to do exactly that also crashing? Like when I looked at Palantir’s recent quarterly earnings, they had, you guessed it, a record quarter, but their stock still crashed, and some of the case studies that they used was that I think it was consumer, yeah, it was.for the first time, so non kind of government contract type based work for Palantir, which is the bulk of their revenue, finally surpassed or increased massively. It became a big chunk of revenue earnings for them. And the reason why that was the case is because Palantir has been kind of building this version of a product, which allows you to agnostically use whatever AI model you want and not overspend.And so companies have been really leaning into that and it’s been generating a lot of revenue for them. And the final sector I want to call out, which again makes no sense, is when I look at companies like CrowdStrike or any other cybersecurity platform, and they don’t necessarily need to be uh in ownership of an AI model that can prevent hacks, it’s just like companies that are focused on cybersecurity.If you look at the last 2 to 3 weeks, we have seen some of the worst hacks in history. If you look at the last 6 months with OpenAir’s hugging face attack where models internally broke out and hacked a public company and we’ve seen multiple versions of this happen in various different ways over the last couple of months, you might be worried that we’ve reached a point where AI models or any kind of person with access to AI models can now be.Competent enough to write a few words or write an extensive prompt and go out and hack certain things, and certainly if you look over the fence at Europe and what’s happening there, even on home ground in the US, you’ll notice that a bunch of devices this week, known as fridges, yes, the thing where you literally store your groceries in, have been hacked to the point where food has been going spoiled, or in one serious case recently I saw a blood bank in France got hacked.And whoever hacked it turned the temperature up of the test tubes of this blood to around 100 °C, which would have killed or eradicated the entire thing. So the idea here is we’re in a world where it’s a free for all. If you have access to an open model or a frontier model, it doesn’t really matter at this point the open models are in many ways, from a cybersecurity point, as good as the frontier centralized models, you can just hack your way through a bunch of different things.At a point in time like that, you probably want to be increasing your defense and security, and so you might argue that it was bullish for those stocks, but nope, red on the day as well. And for those of you who think that this is just a US or Western specific problem, no, this is being seen globally when I talk about the demand that is out there for.AI, I mean, look at this tweet over here from Shay where he goes, Chinese AI lab Moonshot is reportedly looking to rent more than 8000 Nvidia Blackwell GPUs from European neoclouds with one proposal offering a 40% upfront premium. Now, China has not been able to buy Nvidia GPUs and Jensen.He tried to politically lobby to enable this to be the case, and he’s kind of got some allowance where he can sell like a certain amount, but it’s nowhere near as much to satiate the demand. And actually, the Chinese government themselves have gone out of their way to restrict Chinese labs, saying we need to be dependent on our own infrastructure. We need to bring our chip level caliber up to the likes of Nvidia.But even they are still trying to circumvent the law in China to get access to these chips because they’re so desperate to stay in the AI race and the AI competition to the extent where they’re willing to pay 40% upfront or a premium on top of the prices that US labs are paying, just insane to see. And if IMy long-term lens on, I can’t help but think that we’ve barely cracked into what AI models today that exist today, that I’m not talking about the new ones that are gonna be released in a few weeks’ time, which are also another story to talk about later. But the AI models that exist today haven’t even cracked the surface of what could possibly be done.Uh, if you look at Anthropic S1, for example, right, and we’re talking about Q3 Financials or the Q2 and Q3 financials, one crazy stat from there is that of all the Fortune 500 companies that are in contracts or use Cloud, they are spending less than 0.1% of their entire budget.Entire annual budget. So that’s barely a fraction. Can you imagine if that increases to 1, which is a 10x or even 5%? That would result in so much revenue generation if we’re taking into account the economics that we see today. But that’s not even the biggest market there. There’s all the other markets that are about to be encroached by AI. Now listen, over the last 12 months, a really good example is coding, right? Software engineering just got pillaged.By the brand new AI models that were doing a fantastic job in not only automating all the boring parts of software engineering, but being creative in ways that software engineers didn’t really think of, and also being able to work 24/7 nonstop for a fraction of the price so that you can pick out bugs and discover things that the human brain just hasn’t been able to comprehend. Like, imagine having 1000 cybersecurity experts on your doorstep working 24/7.For $5 right? Like that kind of economics margins we haven’t been able to play with. That has happened to software engineering and it’s gonna continue happening, but that’s about to happen with cybersecurity. It’s going to happen with mathematics, it’s going to happen with science, it’s going to happen with medicine, and any sort of form of, I guess, research and development. And that’s what’s really exciting for me as a former biologist, I remember spending hours and hours for weeks and weeks at a time.Trying to figure out a very specific thing, I was trying to locate a gene that would turn a certain protein on in a particular mouse or murine model, and at that time, I remember thinking, wow, it would be really cool if I had some kind of idea what genes had worked in the past, for example, right? Now, if I had an AI model, I’d be running that through all the latest literature and it would be reading all the science papers for me and bringing me back exactly what I need. And the difference there is it’s not replacing.The whole idea here is that you smart, knowledgeable people who have been in the industry for a while, know which right questions to ask. So I’m getting a bit too sidetracked here, but the idea here is there’s so much untapped potential in so many other industries that I feel like the market is just overdrawn and not even thinking about it. Now, before we finish, I do want to round this episode out with some takes that show where the bearish outcome could actually happen on today’s episode brought to you by Huokong.Number one, layoffs seem to be increasing, although they’re not being called layoffs for some reason, they’re being called buyouts. Now, buyouts is a very nice way for a company to basically say, hey, we kind of want to do layoffs, but we don’t want to do them. So what we’re offering is a buyout program, which is, if you want to volunteer into this program, we’ll pay you something along the lines of 3 to 6 months’ worth of wages plus medical benefits that you can continue for like the next 9 months or whatever that might be.As long as you agree to quit. And so we’ll give you this kind of like retirement package, right?And so the likes of Microsoft and Google have been rumored to be doing that. Microsoft more explicitly is apparently offering up to 60,000 of these buyout options, and the craziest part about this is on the list of companies or teams rather, that they’re looking to lay off or buy out, it includes the AI team, but Google’s also doing the same thing. So the idea here is the capex spend, the amount that these companies are spending to acquire GPUs, stand them up and service all these new AI models.Including the ones from anthropic and OpenAI have become too costly to the point where these companies are on the brink of or have gone into negative cash flow. And so if this continues, if we see this continual trend of spending or overspending so much, then you could potentially see that as bearish as well as, you know, putting people out of jobs, you know, where are they going to go. So I think that’s a very fair and sensible take. What I will say is with the elections that have happened so far, and if data center rollouts do slow down over the next year.They’re not gonna have any of this money to spend on, so hopefully, optimistically in a weird way, it slows some of this down. And then the final thing is, let’s put my rumor hat on. Uh, apparently, anthropic investors are calling for the company to lower their valuation to the likes of around 1.25 to $1.5 trillion. Now, my personal take on this is that is probably sensible, not just because markets are crashing all around you, but you kind of.Don’t wanna overestimate and then underperform massively, especially on your opening day, especially for the first most major AI lab out there. SpaceX AII don’t really count because they’re primarily a space company, they’re doing a bunch of the neo cloud stuff. Anthropic is very much a pure AI lab and it’s gonna be the first of its kind to go public. So if we see that actually come into fruition, I can imagine that market participants will think, you know, more bearish AI, but I actually see it.As an optimistic thing, which is like, cool, they’re pricing it fairly. It gives investors an easier way to kind of like get involved with this stuff and yeah, maybe it ends up working out over the long term. But on the point of long term, and my final point, I promise, I think that markets are simply just too overdrawn right now and I think we’ve forgotten a bit of what got us excited about AI in the first place, and I encourage everyone to kind of refine that, and I say that for myself as well.Well, because certainly over the last two weeks, I think the, the AI stories, news trends that I’ve seen have leaned more bearish and as an optimist towards how all this technology is going to roll out, I want to remind myself that, you know, maybe the markets are overdrawn. So anyway, that’s my take, and hopefully you enjoyed this week’s worth of episodes. We had some banger episodes go out. Definitely go check them out. So yeah, thank you so much for listening and I’ll see you on the next episode brought to you by Qualcomm.