OpenAI *JUST* Went Mega Deep on Apple's Stock. Should you?

This is actually crazier than I first thought. When I first came across this news about OpenAI buying thousands of Mac minis and Mac Studios, I'm like, "All right, is is this really just going to be to provide some like simplistic kind of agentic tools?" And then I'm like, "Oh, wait a minute. We actually look at the data. It is quite different." Holy smokes. Let me give it to you straight. And you don't even need like a lot of technical knowledge for this because I'mma make it simple for you. Okay, here's the thing. When you compare to the 5090 Nvidia graphics card, like the standard kind of graphics card that you could buy off the shelf for a massive premium right now, uh, and play video games on it, that graphics card is pretty good. Uh, in fact, it's about 49% faster than the M5. So, what's all the hub about? If the 5090 is better, you don't even need the RTX 6000 to beat the Mac. It's just another crap Mac product, right? No. So, you can't just stop here at memory bandwidth. And for the rest of this video, you'll see exactly why. On a memory bandwidth basis, yes, the RTX5090 is faster. The way to think about bandwidth versus the total space available, what I like to do is I kind of like to think that the total gigabytes of space you have available is kind of like how many kitchen cabinets you got, how much room you got to put, you know, crap in, then you open up the cabinet, you're like, "Oh yeah, I forgot I had those coffee mugs." But bandwidth is how much room do you actually have in your kitchen? Like if you put two people in there and they're like tripping over each other because the dishwasher's open, nobody can really get by, you're gonna be really slow at getting things in and out of that cabinet. Whereas the more bandwidth you got, the faster and more efficient you are and pulling, writing, whatever, getting getting the information you need, putting it back, whatever. That's bandwidth. And in fairness, the Nvidia card, the 5090 is 50% faster, but on a 32 gigabyte model, so you don't have that many cupboards. Now, that works great for a lot of purposes. Like, we'll go through some examples of that in a moment. And there's some advantages to having just a bunch of those working on their own individualistic agentic or AI tasks. That's fine. You know, we handle a lot of that with our startup, Reinvest. But what's remarkable about the M5 is when you slide over, you can actually get up to 512 GB of unified data where you're only or or unified RAM essentially, whereas you're only getting 32 on the GPU. And so basically in the Mac both the GPU and the CPU can pull from this RAM pool whereas the 5090 is limited to the RAM pool on its chipset. So that's what limits this. Now you could go to like an RTX 6000 Blackwell chip. That's going to get you 96 GB of VRAM. That's great. It gives you a whole lot more headroom and you could run some of the models that we're going to talk about like OSS120 from GPT and OpenAI. But 512, holy moly. Now, the 512 is not available yet, but you'll see how even the 256 GB version is really, really powerful. And it explains why OpenAI is buying these. Now, later in the video, we'll go through what I think this means for Apple's price target. And, you know, is it a buy plus or minus? What valuation can you really justify here? Just know when we get down to that end, it ultimately comes down to what you think Apple's growth is going to be. And I'll give you an example of where if I was really bullish Apple, I think this could actually move Apple stock to over time because there are real constraints. You know, one of the reasons these are going to be selling like freaking hotcakes is because Apple actually has the memory allocation and that's what a lot of labs want right now. Specifically, OpenAI. See, they actually run a model that is coded to work on Mac mel metals, you know, silicons. And it's called GPTO OSS120. It's the 120, well, it's technically 117 billion parameter model. And the cool thing is it actually only uses about 61 gigabytes approximately, I'm rounding a little bit, of memory. It's about 60.8 to be more exact, which means with a 96 GB chip, you could actually run it no problem. You could actually get the 96 gigabyte M5 Ultra and run it and still have headroom, but you can also run it on the 256 or the 512. Now, why does more headroom matter? Well, because the first thing that you're going to do is you're going to actually load in some of these larger models. Now, some of these larger models are not actually going to have all of their uh sort of model uh loaded up at the same time, but let's just make an oversimplified example to make this make sense. If you're running a model that's using 60 GB, you ain't running it on the 5090. It's just not going to work. you're going to have to go to Blackwell RTX 6000, which we'll compare to that in just a moment because that's going to make you go, "Damn, Apple, that is genius." But anyway, a 96 uh let's say is right here oversimplifying is called having more headroom. What can you have in this headroom above the actual loaded model? more context, larger documents, more memory, more history about whatever it is you're blabbing about to these AI models or whatever it is you want your agent to do, your business, whatever. If you got the 256, holy smokes, you got a whole lot more headroom. And then, of course, if you got the 512, holy moly, why do you even need it? Well, because you could run a bigger model. You could then come in and say, "Wow, we got so much space. We don't actually need that much headroom. We could run a model over here at this level and still have all of this headroom up here. That's the point. These Macs are actually so freaking fast on the bandwidth point of view. These new Macs that are announced and they have unified memory which again is allowing the GPU to access all of that RAM. That is really nice. So all of a sudden now you know we can do a lot of identic work with this. I'll give you some examples. All of a sudden, now we got a price compare. So, this Mac Mini or Mac Studio that I configured is the 36 core CPU, 80 core GPU, 32 core neural engine. It's the 256 version. This version works out to just over $12,000. Now, that sounds ridiculous, right? Why am I going to spend $12,000 on a computer? But that's the whole computer. everything you need other than like the monitor, right, is in that compared to a desktop computer where hm how much is that going to cost if I wanted to run let's say a model that needed at least 96 GB? Well, here I've got 256. Well, what would have 96 GB of VRAM? Well, the RTX 6000. We own some of them. They've like doubled in value. They're great. Well, how much is that going to cost me? Because Kevin, you're telling me that Mac is going to cost me 12 grand. How much is an RTX 6000 going to cost me? And that's only going to have 96 GB of RAM. The Mac would have 512. The 96 GB RTX 6000 will cost you 15 grand. Massive MSRP premium right now. Uh Nvidia, I believe, sells these for about $11,000. So $1490 divided by $11,000 is about a 36% premium just to get your hands on it right now. Like at that point you could have the whole Apple computer because now if you bought this chip you still have to put it into a machine. You still have to put it into either a server rack or a workstation and you got to buy all the other components for it. The Mac comes loaded with it all for uh for for that um you know $12,000. Now, I I'll pull up the actual configuration so we see it and we'll go through some of the other configurations like the minis and all that in just a moment, but it actually shows you the power of what Apple has just delivered here and it makes sense why OpenAI, you know, even though there we'll speculate about some of the Apple related purposes that they could be looking for here. I think a lot of this has to do with their desire to get into agentic coding APIs because they want the eye, right? That's where the dollar is, the income. I'm just kidding. That's not We know that's not what APS stands for. But the point is, here it is. The 256 model. If I go with the M5 Ultra and we throw in the 4TB version, that cost me $12,300. I could drop down to the 96GB version, holy smokes, it's only $8,300. It's still a lot, right? But it's only $8,300. Whereas the equivalent Nvidia chip is literally going to cost me compared to $8,300. I mean, it's 15 grand compared to $8,300 rounding 10 bucks on each. That's 80% more for the Nvidia chip. What now? Yes, the Nvidia chips are faster. So, on speed alone, the Nvidia chips are very, very good. But think about some of the things that you could do with these chips. And we'll get into some of the analysis on numbers and values. So, first of all, you could run a, you know, we've got right here the 5090. You can't run OSS120. You got the RTX 6000 right here. You could run it with a little bit of headroom, right? You could do a coding agent. Uh, that's probably with a whole repo going to push you past pretty much both of those Nvidia cards, the RTX 6000 and 5090. So, you really want like a medium-sized code base. You're going to need more headroom. You can now get that if you jump into that 256 or 512 model of that M5. That's where you're getting that extra headroom. And that's the value here. So like think about just some examples for this. Let's say you're doing financial research and you got to put a whole lot of financial SEC reports in or you got to put I don't know, here's another example, 50,000 filings into a single batch. You're going to want that extra headroom. That's where this extra benefit comes from is your agent isn't going to get context limited. You could go do a lot more and you could even do things with larger models. For example, if I compare the RTX5090, RTX 6000 and the Max Studio, we really can't run any of these full-sized models. I can't run a DeepSeek R1. I can't run the Quen 3 235 billion parameter version. It's just not going to work because that's going to take 118 gigabytes of of uh you know weight. I'm going to need 336 for Deep Seek R1. I can actually run that on the M5 Ultra, right? I mean, there are endless things. A Frontier Scale Agent C, whatever, right? There are endless things that you could do here. Now, you can stack cards in fairness. If you put two RTX Pro 6000s together, you'll be at 192 GB, but that's going to cost you, you know, right here it says $26,500. That's because it was taking last week's pricing. That's going to cost you like 30 grand if you go through Amazon right now. That's the power of what Apple has packaged together right now. And it's incredible. Now, if you're running smaller scale things, you can run it faster and more efficiently and more cheaply on a 5090. So if you're running smaller models, 8 billion, 14 billion, 27 billion, maybe 32 billion parameter models, you could run these on all of those cards. So this is really task dependent. Do you want to run really big agentic workloads for coding databases or massive filings analysis? You might want that M5. Depends how much headroom you needs and what kind of model models you want to run, right? you want to run just quen 3 on a 32 billion parameter model, you only need 16 GBs of RAM. You're good on the 5090. You don't even need a 5090. You could probably run the 4090. Uh as soon as you want to get into these larger, thicker models, you're cooked. And then obviously get if you get into like Kimmy K2 or K3, these really massive models, none of these are going to work. That's when you're going to get into those larger uh chipsets. But, you know, something to think about. This is really impressive. Now, let's get into some more analysis. AI training. you're taking the largest amount of information possible and trying to get this to be incrementally better. And you need the GB300's from Nvidia, these massive training uh uh chips, essentially $50,000 chips instead of, you know, a $5,000 consumer grade chip that you end up putting in a $4 million rack. You need that for frontier training. But for aentic use, people are buying Mac Studios as a call option on what artificial intelligence uh chores, agentic chores might end up coming to an iPhone near you soon, which is kind of incredible. So, what if that iPhone fold potentially gets announced with a new operating system and by October we're able to run localized Aentic AI on our phones. The thesis is maybe if you start building out server sets on Mac studios like OpenAI, you might be the go-to Mac infrastructure supporting Mac users or iOS users. Now, a lot of that is speculation, okay? We've heard Apple come out years ago and go, "Ah, introducing AI and and they've done like nothing." Okay, maybe I shouldn't say they've done nothing. Text message summaries are cool. Notification summaries are cool. This is like scratching the surface of AI, right? Email notifications are cool. Email search has gotten better. Uh, you can now search your photos with text and which is also kind of creepy, but they basically like somehow uh categorize all your photos and then you search like burrito, all your photos of you with burritos come up and every time I do it, I just get me wearing a sombrero talking about taco Tuesday and I'm like that's that's Taco Tuesday. It's not burrito Tuesday, but I guess it's close enough. So like that's the kind of AI or like making little mmojis that Apple has been doing. And there's a bet now that Apple could actually get into agentic AI and that potentially is leading per the information via tech republic open AI to purchase tens of thousands of Mac minis and Mac studios. Now the Mac minis are still available right now. They're all pre-order uh as of uh now starting to be available uh September 22nd. And what I noticed is that the M5 Pro chip for the Mac Mini with 64 GB of memory is available on release day. Uh and it looks like that's available for all variations of memory. If I go to the M6 chip, uh let's see what we have here. We go to the M6. We can only go up to 32 GB of RAM. Oh, wait. No. Choose from more options. Okay. They've got a little bundle here. 64. Oh, no. I've got to update to something else. All right. Whatever. If I accept the change. Anyway, with the mini, I could still get this device on the 22nd. Oh, it it dumps me back down to the M5. So, you can't get the M6 with 64GB of RAM. That's what it's doing. So, those are available. But look at the Mac Studio. This is the chip people are freaking out about, the M5 Ultra chip. A lot of devs are waiting for the future 512 option because you're going to be able to run really big models on this. If they actually have the 256 GB of RAM in one device, people might be buying these M5 Ultras solely because they actually have the memory. And so when this story came out about 25 minutes before the market closed, Apple didn't move that much. You know what moved? Micron moved and Sand uh SKH Heinix moved. Not SanDisk. SK Heinix and Micron both moved on this news. I think it's because the sign that OpenAI is coming in going we will buy memory wherever it is to run agentic tasks is a way of saying a the circular financing that we're seeing of the labs blowing every freaking dollar they have is nowhere near over. It is just getting started now. That makes me bullish. This is why I say I don't think a recession is close. This is why I also say wherever I have it here, I got to I got to figure out how to organize that a little better. But this is why I also say on the bull bear scale or bare bull scale I like to call it. You I'm a 7.1. That's that's bullish because I don't think a we're close to recession and b these labs are drunk spenders. And until they IPO, and not just until they IPO, until they IPO and they blow all their money and their stock has performed so poorly that somehow they can't raise any more money. Only until that point will those companies spend money. So in other words, they'll spend all the money they can up until the point that they can't finance anymore. Then they'll stop blowing money and then we have to worry about the bubble collapsing. But until then, they are spending like drunk sailors to the point where Apple announces Max Studios and what happens? Oh yeah, you know, we're going to buy uh tens of thousands for AI agentic uh work. This is really really impressive. I mean, like maybe Teachable should buy one of these Mac Studios and make sure their software doesn't go down when we have a freaking coupon expiration because on Friday night their software went down and there were a bunch of people who emailed us. They're like, "Hey, Kevin, what the hell hockey sticks?" And then of course, Teachable 13 hours later is like, "Hey, we're sorry that went down and we're sorry all your customers are like, "Dang, bro. I wanted to sign up, but I couldn't." Uh, and so we're extending the coupon code till Friday. I think that's that's reasonable to give folks uh the opportunity to actually buy what they wanted to. But anyway, maybe they could just buy a Mac Mini and not have this problem. But looking at the rest of the article here, we could see the information has been talking about how Apple stumbled into AI hardware success with the Mac. This is an older piece where they just talked about how the machines are well suited to running agents, AI software that handles multi-step tasks from editing and testing code to automating email box uh inbox organization or summarizing documents. Basically, one step at a time tasks that you could really do with a machine uh like a smaller, more consumer-rade machine. But this wasn't really the piece that moved everything. it was this this piece about them buying tens of thousands of these is a gamecher and that's where we kind of have to look at how could this potentially affect Apple's you know revenue first we already see their Mac sector growing percentage-wise faster than the iPhone grew revenues $10 billion but the Mac grew at 7 percentage points faster the M5 Ultra is also Apple's first quad die architecture their most powerful chip And people cannot wait to get the uh uh 512 version. Right now you can only order up to the 256 version. So uh this has been talked about before like even in the Apple's earning in the Apple earnings calls they've talked about how people are deploying clusters of these to run frontier class models locally. Open AI blowing money on Apple was not priced into this stock in my opinion. And so I think the memory stock or or the memory situation has become so ubiquitous that we could really see some actual growth to these MAC numbers. The problem is we have real constraint issues. Growth rate estimates for MAC right now are only about 8%. So even if you I don't know in my opinion you go to 25% on Mac you're constrained by revenue and you're just not moving that bottom line that much because if I go 10.3 billion and I multiply that by the current estimate of 8% growth I get that's billion right I get uh $820 million if I go with my number of screw it let's go with 50%. That's 5 billion. That's four more billion dollar. Okay. The bottom line in sales would be about 4% more sales. Call it, you know, $4 billion coming to the bottom line. So, let's go to their income statement really quick. And let's see what we've got here. So, here we have, let's see here. These are the shares outstanding. That'll be useful in a moment. If we look at their net margins, their net margins are 29.78 divided by 109 29.78 divided by 109.4. I'm at about 27.2 comes net to the bottom line times $4 billion. That brings me down about a billion to the bottom line essentially on net. Okay. If I take another billion bucks, 1 divided by 30, I'm going to increase my net by about 3.3%. So, my earnings per quarter go up like 7 cents. You know, it's not that much. If I multiply that by the year, I get to about 27 cents more of earnings. Okay, 27 cents more in earnings. Uh, you know, what does that actually do for the stock? Not much because you're memory constrained. And that's probably why we saw those memory stocks move more than Apple because 27 cents on let's see what the current forecast is for Apple and the PEG ratio is already elevated for Apple. It's you know it's it's up there. So if I pull Apple financials right now I'm going to go look at $8.85 27 into that increases their earnings. Yeah. 3% is all you're really increasing earnings per share. I think what's more interesting is you could really escalate the potential growth rate for the company. See, the growth rate for the company is only estimated at 9% right now. So earnings would only grow by 9% per year. That 9% per year works out to about 75. So you're increasing that by a third. Okay, let's get more aggressive. Let's assume memory supply really starts coming online in 27, 28, 29, and 30 and people actually pick up these Mac Studios. Pricing power stays high for Apple. Pricing power stays high. Memory costs come down. Apple takes more margin. This is the play. The play is if you believe that Apple can get to 20% EPS growth because all of a sudden their AI takes off, people are buying Mac Studios to run models locally. Screw Open AI, we'll compete them away with our own Chinese open weights, but we're going to run them on Mac minis and Mac Studios. Then here's the play. 20% average expected growth. We take their current EPS and we multiply it by about a 24 peg. probably this could be a $425 stock. So that's potentially where you could see that upside compared to the 316 where we sit today. 316 divided by 425. These boosted expectations uh could potentially lead the stock to run up about 35% just on expectations like that. That would be my initial analysis. But it really requires the memory shortage like stopping being a memory shortage because otherwise Apple's going to be too constrained to see this happen. And really what it means is OpenAI is going to get in line in front of you who sat around and waited a couple extra days to place your Mac Studio order. And now Sam Olman's getting them all. Damn it, Sam. He's back at it again. But yeah, I mean this is happening at the same time that their new CEO is stepping in. So, a lot of people are like, "That's right. This is becoming a data center provider, baby. Let's go, Apple." Uh, and so that's the big play here. And yeah, that could actually meaningfully rerate the stock, especially if they could bring better AI to consumer grade hardware that we actually use as consumers. And they could start supplying the Frontier Labs because those people spend like drunk sellers. And anybody that is selling to those is just going to see their margin and their earnings per share go up way more than analyst expectations. So, personally, I'm I can't be bearish Apple on this. Now, I don't really want to pay these levels for Apple, and I think there are a lot of really juicy plays out there, so there's a limited amount of money to go around, but I certainly would not be short Apple. >> Why not advertise these things that you told us here? I feel like nobody else knows about this. >> We'll we'll try a little advertising and see how it goes. Congratulations, man. You have done so much. People love you. People look up to you. >> Kevin Pra there, financial analyst and YouTuber. Meet Kevin. Always great to get your take.