원문 보기
Well, don't sue me, bro, but in this video, you're going to learn some really big and key differences between two very massive investment opportunities. We're talking about $3 trillion investment opportunities, Apple versus Nvidia. And so, in this video, I'm going to give you my thoughts in terms of what I'm personally doing with my portfolio when it comes to Nvidia, and maybe some things that you want to consider between the two companies, Nvidia and Apple.
We'll also talk valuation. First, we'll start a little broad with what's going on between the AI spaces. But, I want to give you a quick disclaimer. Don't Don't sue me, bro. This is just my opinion. This is not personalized financial advice, and frankly, it could be wrong. But, I do also want to start by saying I've been very bullish Nvidia.
I loaded up on Nvidia originally back in November of 2022 as we sort of bottomed out. I added a bunch in the 400s. Obviously, I wish I uh had more exposure leading up to that run. Oh, well, I took some gains, then I got back in a little higher, but I'm glad I did because obviously the company's done extremely well.
The question now though is what is a better opportunity going forward? Hindsight analysis only does uh so much. And in this video, I'll also go through what Apple just transformed in the AI space. So, this is really important. Okay. Now, something that we have to also cover before we start is the Yes, I did make some money on an Apple trade today.
I had a sweet sweet set of call options on Apple today after that event yesterday, and yes, we made 100K. , make sure you click that link down below. Join me in the Stocks and Insight Group. You get my course member live stream analysis, all my deep dive analyses that I'm doing uh before we post about them here, if we even post them here.
We do a lot of analysis there that doesn't make it to the main channel. And I sell I send all my buy sell alerts, whether I make money or I lose money, you get it all. Last week I had a rough week. This week, let's just say it's starting out really good. I actually tweeted about the options that I had and you could see I had about $80,000 of exposure to to Apple today.
So, the options did really really well. So, and that's after the gain. I mean, this one was up 103,000 103%. That one was up 76%. I go in and out of them. Uh but anyway, as always, remember and it's unclear that if you join, you'll make money. I just want to be very clear about that. The goal is teaching and education.
So, what's going on with Nvidia? Well, first, we need to know what this significant paper is that just came out regarding Nvidia because this is a risk factor for Nvidia. And all of what we had presented by Apple is a risk factor as well. That is going to compound a risk factor for Nvidia. Let's just be very clear though.
Nvidia is the leader of the pack. Nobody stands in the way of Nvidia's moat. Nvidia is a company that I want long-term exposure to. I want Nvidia exposure not just because of the innovation of the GPU cycle that is we're going from the H100s to the next levels to the Black Wells to the next levels thereafter, which they're already announcing, getting on a sort of this one-year product refresh cycle.
We know we're going from your sort of typical dumb compute to AI compute service. We already know all of that. We know we're on a new innovative S curve. The question now is, at what point does that S curve peter out and does it still make sense to buy Nvidia now post split? Especially since the post split performance of stocks historically is good, but in recent history has actually not been that good.
Consider over the last fourish years after Google stock split uh on July 18th, 2022, the stock was up 3% over the next 3 days, but then fell 10%. After Amazon stock split in 2022, it fell After Apple stock split in August of 2020, it was up 10% at first, and then plummeted 20%. And after Tesla stock split, well, that that one has just pretty much been straight down.
But anyway, at the 6 days thereafter, it was down 12%. So, stock splits don't necessarily mean in the short term the stock is definitely going to go up. Of course, great companies split their stock, their earnings grow, and they keep going up in the longer term. So, do I really want to bet against Nvidia for the very long term?
No. But would I be willing to make a bet or a trade between Nvidia and Apple now? Absolutely. And the question that comes down to everything for me is how much actual GP GPU compute do we really need? Picture this for a moment. I want to draw this for you before we get into this study because I think it's going to be enlightening, let's just say.
If in March of 2023 to probably the 3 months thereafter, you have this bucket. I'll call this the March and to let's call it July 2023 AI, the class of AI, okay? That's a good I like it. Sort of like Navy SEALs, but it's the class of AI. Okay, and let's say you have 1,000 people in the class of AI. The question that I have for you now is how many people or how many companies rather do we think is in the class of AI, not necessarily today, but in the 2025 class?
Well, my guess is we're probably going to go down to somewhere between 50 to 100 in the class of AI. Which means if we're building out server compute for the class of a thousand, assuming that they're all going to maximize all their compute because you and myself and businesses, instead of just trying the one AI that we need, we try five different AIs.
So, we're actually creating five times the demand that we actually ultimately need for AI. Well, then what happens is everybody buys AI chips, everybody over pays for the AI chips, and Nvidia goes to the moon. And companies that have nominal AI exposure like Dell, which is just a rip-off giving away server stacks, doubles its stock valuation.
AMD, Vertiv, other companies like Super Micro Computer, all the adjacents also end up doing really well. But, is it actually based on a fugazi set of demand? Now, of course, not everybody got into AI day one. So, of course, that class theoretically should grow to let's say 1,200 as more people slowly trickle into the class.
They show up late, basically. But, does that just push us to maybe a 60 to 120 class, right? Maybe 20% more. I don't know. This is speculation, obviously. But, the point is that let's say the compute we need for this basket over here is X, okay? So, X compute. Well, the compute you need over here in a more winner-take-most environment, you could even say that AI will just boil down to five to six different AI companies in the future.
There's still be a set amount of AI demand. But, the point is you won't have the duplicative training and the duplicative inferences people test out different AIs. You're probably looking at AI demand that's something like X divided by 10. Uh and and and then maybe you grow that by like 10% a year, right?
So, you multiply it by 1.1. Now, that's not supposed to look like and remind you of your calculus class or your algebra class. It's just simply to say that demand is probably substantially smaller than what we saw in 2023 for most AI demands right now. Now, a lot of companies are still trying to build out LLMs, but even LLM demand might not use as much GPU compute as we think to train new LLMs.
This is going to get really important in just a moment. But, the point is, as you go from the many to the few, you're going to reduce the amount of demand that you have, and then you'll grow that reduced uh, set because you have people consolidating the products they're using. They're probably the best products of the bunch, which are the most efficient products.
And the most efficient products won't need, frankly, as much compute power. That's why they're the best. That's why they're the most efficient products. Okay. So, so now we've reduced our overall demand for artificial intelligence. Is there any evidence to back up that that has happened? Well, yes, there actually is.
Tesla delayed its delivery of H100s because they say they don't have enough room for the server stacks. But, if they really needed them, they would have been ready for the shipment. Reality is, these chips at Nvidia are so good that Tesla, as Elon himself says, is no longer compute constrained, and we don't need as many chips.
Especially since the Blackwalls are supposed to be 4x as fast as the H100s, you need 4x fewer chips. So, if they introduced the H100 back in 2023 with the same strength as the Blackwell today, you would need 4x fewer of them, which means you'd have potentially 1/4 of the AI chip revenue for Nvidia. See where I'm going with this?
As the competitors in the AI space shrink, demand falls by virtue of a lack of duplicative work. Yes, it's still going to grow. AI demand is going to grow. Don't worry about that. That's what I'm saying, times 10%. You're still going to get that growth. But this is something that is made worse by three different things.
One of these is matrix removal. One of them is caching. One of them is Apple. Matrix removal. Matrix matrices are used in large language models. They are used in conjunction with bias weights, and they're used inside of neural nets to compute basically likely outcome for something, whether in training or inference or whatever.
There is a paper that just came out from the University of California, Santa Cruz in conjunction with Davis. Davis is where my daughter Summer was, by the way. She's back home. Really grateful for that. She's beautiful. She was just out there smiling at me. She's so cute. I love I love walking out there, giving her a big smooch.
Anyway, scale matrix free basically language modeling. So here's basically a paper suggesting, can we get away from using matrices in language learning models? And here's what they find in their abstract, and we'll look at the conclusion. This is very, very important if you're either an Nvidia or an Apple investor, because you're going to see something that's going on here.
And I'm I'm don't sue me, bro, because I know there are a lot of people who see this stuff that are like, "Oh Kevin, you know, you're just biased against Nvidia or whatever." It's like, "No, I'm I'm long exposed to Nvidia." But yeah, I am hedging myself in the short term and I'll show you why. Take a look at this.
In this work we show that the mat mold, which is matrix multiplication models, uh basically typically use most of the GPU performance. So, let's say you have 100 watts of power being used, let's say the matrices use like 90 watts of that power. So, they're making the argument that a lot of power of as a percentage uh compute power, actual energy, that has different implications as well, goes into mat molds, matrix multiplications.
Okay, interesting. So, what experiment did they run? Well, they ran experiment that says, "Our experiments show that a mat mole free, so no matrices, model achieves performance on par with state-of-the-art transformers that require far more memory during the inference scale up." In addition to that, we also provide a GPU-efficient implementation of this model, which reduces memory usage by up to 61% over an unoptimized baseline during training.
In other words, we can create large language models using field program programmable gateway arrays, which are quick for low power operations, especially mobile devices. Basically, we can use FPGAs instead of GPUs and get quick results with less energy consumption, which I want you to think about this for a moment.
When somebody uses a computer, when you use GPT or Siri, do you want a damn pinwheel every time you're getting an answer. Of course not. It's like when you shoot a gun when you're playing Rust and and you're raiding your enemies, do you want a big lag latency time? Of course not. You want fast. You want as little latency as possible to maximize your ability to make money or to succeed at whatever it is you're doing.
Sometimes that could just be having a video load while you're all alone. I don't know. Point is, regarding this as a competitor or or something to Nvidia, is this a way of saying, "Hey, if people can find more efficient ways of running large language models?" And the great big innovation that's really come out of the latest AI revolution has been, frankly, glorified chatbots that are really good.
Don't get me wrong, they're great. Great sales assistants and otherwise. But outside of generative AI, these large language models, if you could run them off GPUs with less power, less latency, and less cost, then yeah, people are going to do that, especially if the results are similar. In fact, you could jump over to the conclusion, and you can see they say, "By prioritizing the development and the deployment of matmul-free architectures, as in architectures that don't require as much of a GPU lift, the future of large language models will only become more accessible, efficient, and sustainable." In other words, as soon as we get GPUs out of our life, oh, we can actually have more LLMs and more competition, which more competition in LLMs puts even less or I should say puts companies like OpenAI at risk for more competition.
And let's just say, I'm making this extreme argument because obviously OpenAI could use these matmul-free models as well. But let's just for giggles say that GPT is going to be 100% GPU-based, and a competitor comes along and it's 0% GPU, and we're just going to say comp over here. And let's say this one is faster and just as good as GPU uh GPT, but you don't have any pinwheels.
Well, then GPT will go bankrupt and the competitor will succeed. The competitor won't be using or relying on those GPUs, which is bad for Nvidia. Interesting. This is something maybe we haven't thought of before. Now, obviously, it's likely that GPT, you know, OpenAI would adapt and use these matmul free models as well.
It's also possible this paper is wrong, but you may not have thought yet that oh, damn. I thought AI meant GPU, which meant Nvidia, man. That That's as far as I got. That That may have been your thesis, but now we actually have to go, "Oh, wait a minute. We can have GP U-free AI." Mess is the thesis up a little, Kevin.
What are you doing? I was just waiting until I get to the Apple part. We're just getting started. Okay. What about caching risk? That's risk number two, the caching risk. Caching is basically way of saying, "Hey, for of the image generation or video generation that or or just quite frankly text generation that people need, a lot of that we don't actually need to run to a GPU server a GPU-based server every time to operate a result.
Instead, if a million people every single year ask why George Washington had wooden teeth, let's just run that calculation one time, and then the other 999,999 times somebody asked about George Washington's teeth, we're just going to feed them the same answer from basically downloaded memory, like a hard drive.
Like imagine downloading on a future iPhone 50 GB of a GPT encyclopedia that has a cached answers text-based answers for everything, even potentially canned uh image generation or video generations. And then, the only portion you have to send to the GPU-based cloud is just what's different in the question.
So, then you can provide an answer that's partly encyclopedia-based in storage or memory, and the other that is based on actual cloud compute. Okay. That's another risk. Caching. That's not like the cash in your pocket. That is c a c h. E would be cash. Often people think weapons cash. Um Obviously, cash is what I'm trying to teach everybody how to get more of.
That's why I teach how to build wealth in the courses of building your wealth link down below. We've got an expiring coupon code tomorrow. Uh tomorrow is June 12th. It is CPI day and Fed day. Big expiring coupon code. Really hope that uh I can keep positive performance going. Uh we are up year-to-date quite a bit.
This is the P&L today. You saw the trade sizes on these two. Uh they weren't uh you know, they were less than this. So, so we did really well. Uh Coinbase was a little larger, but that was also not a derivative play. That was a um uh a share position. But uh but anyway, check it out. Link down below.
Coupon expires tomorrow. Again, cannot guarantee uh or imply that you're definitely going to make money, but the goal is to learn how to make money as much as possible. See, we got a little disclaimer banner now. Okay. So, if you have questions, by the way, email us at [email protected]. So, now the question becomes, all right, we have risks to Nvidia.
We have matrix removal, which is a GPU risk. We have caching risk, which is an Nvidia risk. We have Nvidia that trades at about a 3.5 PEG right now, which is a price-to-earnings growth level, which basically suggests that their growth is going to slow to somewhere around 13 to 15% over the next 4 years on average per year.
That you'll go from basically this doubling to to a slowing to potentially even negative. So, you get this average 13 to 14% growth. And there's a real risk that those growth expectations even in those years are too high. That those growth expectations might actually go or the actual growth numbers might actually go negative way earlier, which would make Nvidia's valuation even higher, which is somewhat scary because it's already kind of rich.
Now, it is a cash-generating behemoth. It makes a lot of money. But, what about Apple? And what is Apple doing that could potentially hurt a lot Nvidia? Introducing on-device artificial intelligence. Now, you might think to yourself, okay, on iPhone, who cares? But, wait a second. You don't like latency.
We already talked about latency. And you're not going to like latency on your phone. I hate when GPT pinwheels or Siri has to think. It should be fast. With the fastest, highest quality product is going to be a product that can answer my question with an on-device neural net or language large language model or whatever.
Basically, AI on the device. If Apple can protect if perfect, which we don't know that they will be able to, but if they can perfect on-device artificial intelligence, forget about upgrade cycle for a moment for Apple and what that means for introducing new iPads, laptops, desktops, and phones. But, what I want you to think of for a moment is if Apple does on-device, guess who's going to do on-device next?
Android will. Because, do you want to buy an Android that always sends your data to the cloud? That's not good for privacy. That's not good for operational security, opsec, and it's certainly not good for efficiency because you're waiting for the pinwheel. So, if Apple does on device, then Android will probably do on device.
Which means probably all hardware in the future will do on device. Which means in the future, iPads will do all the on device work. This is the iPad, by the way, I got because I had a I I made a bet on X on Twitter. Follow me there if you don't already, Real Me Kevin. But, I made a bet if I made trading profits on a certain trade, I'd buy a new iPad even though I don't need it.
Then I bought it, and then I realized that classic Apple move, the Apple Pencil doesn't actually work with with the new one, the Apple Pencil Pro. You have to get the new Apple Pencil Pro. And I'm like, "Oh, of course. Of course." But, anyway. So, the iPhone, the Android transitions, the iPad transitions, what's going to transition next?
Well, of course, the laptops are going to transition next to on device artificial intelligence. Which, if the laptops at Apple transition, then the competitors will transition, which means the PCs may also transition. Now, PCs won't use Apple's M1 chips where you have sort of the built-in graphics into, you know, the M1 chip and the neural net built into there.
PCs might still use AMD, you know, 4090s or or whatever. I mean, these are these are fantastic chips. Let's be real, okay? I'm not going to play video games on a a Mac. It's just not going to happen. I'm going to play video games on a delicious PC because, well, after all, they're really good. In fact, I love these things so much, you could actually see right on screen here that I have my 4090 on this computer that I use when I show a PC.
So, I've got a lot of these. I think I've got three or four computers that I own that have the 4090s. I love GeForce Experience. I love Nvidia. I think they're a great company. They have great products, and they're just good. They're best of class. But, that's for video gaming. We're talking AI here.
And what is Apple potentially going to be able to pull off? Well, first of all, if we don't need GPUs to create large language models because of the matrixy risk, if that ends up working out. And then Apple uses on-device artificial intelligence to do something which, honestly, I was blown away by. I almost had tears in my eyes during the presser of the Apple event, which was, "Oh, we're going to let you take themes from your memories or your pictures or video or whatever that you took on a vacation, and then ask the phone to basically create a custom video out of it.
We'll throw music on for it. We'll edit it together for you, and you could show off this movie that was custom-made for you using your own pictures and video on device, so it's not leaving your device, using the compute power on your phone. That's insane. Like, I wish I had the patience to go through and make collages of all my vacations because I would have many of them.
That would be really fun, and I'd be very grateful for that. I love children, and I love vacations. But, if the phone can do that on device, then that's a huge competitive advantage over me creating a generic piece of imagery or video that I don't really care about. It's not personal. Now, obviously, the on-device aspects that we learned about from Apple, some of them were great as well, such as artificial intelligence to, like, you know, which of the uh notifications you get are pressing right now while Kevin is filming a video versus not pressing while Kevin is busy filming a video.
You know, little things like TLDRs or summary or summaries with the calculator app or the notes and being able to make modifications. These are great. These are really functional useful applications. Making Siri better finally. These are great applications of AI. But we have to go way beyond this idea of a product refresh cycle for a moment and how useful some of these services are going to be as disappointed I was in the event up in the first half.
The second half was actually pretty good. We actually have to look and say, "Wait a minute. Is there a chance that the next AI play is not Nvidia? That the next AI play is actually Apple?" Now, Apple also has a PEG of about 3 and 1/2. So, their valuation is ironically similar. But if Nvidia's valuation is based on growth rates that are too high, then their PEG ratio is going to go up as their growth rates come down.
And if Apple is being thought of as a dead company that doesn't innovate, then their growth rates are way too low, which means their PEG ratio is probably going to come down. So, in my estimates, Nvidia probably has a PEG ratio now of five. And Apple probably has a PEG ratio of closer to 1 and 1/2 to 2.
In other words, people expect this for Nvidia and that for Apple. Well, that changed yesterday thanks to WWDC 2024. But it didn't just change Apple's trajectory. It also creates real concerns for Nvidia. So, what's the trade? Here's my trade. Short term, I'm short Nvidia. Now, that's in a trading portfolio.
I've also reduced my long-term exposure to Nvidia. I've brought I've I've roughly I think a little bit more than halved my Nvidia exposure right around 120 bucks a share. That's $1,200 post split or pre-split rather. So, I've substantially reduced my exposure to Nvidia. Don't get me wrong, I still think they have PP pricing power.
I still think Nvidia has a lot of pricing power. Don't get me wrong. I just think their valuation has gotten them carried away, so it's worth reducing exposure to them. At the same time, it is worth, in my opinion, maintaining pretty high exposure to Apple because this could be the next sort of AI play that everybody forgot.
In fact, if you recall, I made a video on Apple AI and people got mad at me. People got so mad. Meet Kevin, Apple artificial intelligence. Let's take a look at this, okay? So, we made a video called leaked Apple documents reveal AI Siri plans. Apple artificial intelligence. That's what I call it back then.
AAI. I should have put it together that they were going to call it Apple intelligence. It's kind of smart. Uh but anyway, you can see that video yourself here. And I actually really encourage you to check it out because you're going to see that on April 2nd, I was talking about how I wanted to buy more Apple stock on April April 2nd.
And go back for a moment to where April 2nd is. April 2nd is right here. Apple stock is $170 per share. From then to now, Apple stock is up about 21 to 22%. And it's only been about two months and two weeks. Here's a video that only got that got less than 50,000 views that talked about a lot of what I talked about in this video where they're going to do something incredible on device and it's going to be a game-changer.
So, I was transparent about it then. I'm transparent about it now. I think this is a big deal. Now, uh so, my take is uh longer term, less exposure exposure in Nvidia. So, let's say if I'm like 5% Nvidia, 10% Apple, I let those ride for long term. This is not personalized financial advice. I want to be really clear about that.
When I give these like allocations and stuff, this this is not for you. Disclaimer, not personalized advice. This is very clearly just theoretical and what I'm personally doing uh with with my own money, where I want my own money to be. I personally like investing via an actively managed ETF that I don't have to, you know, sort of um how should I say uh worry about too much concentration risk in because there are legal compliance issues with those, uh which is nice.
You can't get like overly heavy in one thing. So, um that's my Nvidia Apple thesis longer term, but shorter term, I'm just going straight short Nvidia through options. You could go out 3 weeks, 60 days, 90 days, whatever you decide ultimately is for you. Again, this isn't personalized advice. Um but that's what I'm thinking.
Uh and then the same thing for Apple call options shorter term, 21 days to 60 days, somewhere in that range. Because what I think you'll see is you'll see money flow out of Nvidia and into Apple. It's just a thesis I have. It It could be very, very wrong. Uh but this also, worth noting, brings up, "Wait a minute, what about energy usage?" Remember how everybody's like, "Oh, energy usage is going to skyrocket because of AI data centers." In fact, one of my team members brought me an entire research piece and they're like, "Oh, Kevin, like are investing in data centers a big play?" You know, there there are a lot of people saying energy use is going to explode because of these these chips.
And it's funny because I I I said I don't think so. And they're like, wait, what what are you doubting AI? Oh, no. Like, what what do you mean then? I go all it's going to take is a more efficient chip that does four times the work at a fraction of the energy output. And the fact of the matter is human productivity just isn't going to go up at the same rate that energy costs for artificial intelligence go down or energy demand goes down.
Like, for example, let's say creating an app through artificial intelligence is going to take me a 100,000 units of power. You know, kilowatts, let's call it. Kilowatt hours. Okay? Is is it and that's that's at today's level. Okay? Well, if I could get that down to 25,000 kilowatts of power, am I going to make four apps?
No, I only need one. Okay, maybe I'll have a second. Do you see what I'm saying? Okay, but then I'm only at 50,000 kilowatt hours of use. That's less than the prior estimate. So, I think that's overblown. That's my opinion. Again, I could be wrong. I Sorry, I like I know sometimes people like, Kevin, you should disclaimer heavy.
And I'm like, well, I just I really just want to be very clear that like I think I I try to give really good perspective. We did an amazing course member livestream tomorrow this morning, probably one of the best course member livestreams with a big trade we've got going that I think I've ever done.
It's really good. I encourage you to watch it and we'll see how the trade plays out, but we got a huge one cooking. And I'm really enthusiastic. So, check out the courses on Building Wealth link down below. Expiring coupon code hits tomorrow. O'Leary th to .
Comments 0