This Is Like Buying Internet Stocks in 1996

The artificial intelligence buildout is way bigger than any of us could have initially expected. And I'm going to show you guys the new details that just came out that prove this thesis that AI is not a bubble and that we're still early. I'm going to show you guys what's changing in the market right now, why I'm rotating my portfolio into AI infrastructure names, what to look out for next, and then the exact companies that I'm buying that you guys can also take a look at and research yourselves. But first off, let's talk about how AI is changing the market. As you can see shown off by Cotu management here that there is a large divergence between the top 20 percentile and the bottom 80th percentile the largest spread that we've seen in public history. The great AI names are going up and the software names and the companies that are not tying themselves to artificial intelligence are falling way down. And you can actually see this based on the industries that semiconductors and hardware names are up 68% over the last year. Energy names are up 18% and almost everything else is down. And this is in great part due to the fact that people are starting to realize that it's not just going to be a human-led labor market. That in fact agents, artificial intelligence that are set to do a command based on knowledge that they are recursively learning and then potentially create skills and then create new agents that can then fix those new problems and then learn new skills and then goes around and around and around. This was never possible before artificial intelligence, but now agents might be a new form of labor. This is unlocking markets larger than we've ever seen before. Essentially being able to see a total digital artificial intelligence industry roughly of around $4 trillion. Once you add on physical AI, we're talking autonomous vehicles and robotics, you could be looking at an industry that could be upwards of 6 plus trillion dollars, even larger than digital artificial intelligence. Yet, we haven't even scraped the surface of this yet. But even though we're early, we're already seeing capital raises. As you guys might know, Google just raised $80 billion to expand their artificial intelligence, infrastructure, and compute. Even Birkshshire Hathaway gave them $10 billion in order to fulfill that promise of building out more data centers. But then they took that money immediately turned around and gave it to SpaceX to buy from them approximately 110,000 Nvidia GPUs, CPUs, and memory and other related AI components. They're going to be paying them $920 million per month between October of 2026 through June of 2029. If you end up taking a look at that and breaking it down, that's about 2.2 two times higher than what they end up doing. They deal with Enthropic and SpaceX. As Jam and Ball says here, this is due to a mix shift of the types of GPUs that Google is actually doing a Colossus 2 deal, which is a different data center that is all GB200s, which are the upgraded version of the type of GPUs that Enthropic ended up getting through Colossus 1. If you end up remembering that SpaceX and anthropic deal was three times three times the market rate of what people are currently paying for GPUs, yet the Google one was 2.2 times higher than the one that was three times higher than the market. Why would they essentially shell out $11 billion or $30 billion over less than a three-year contract to get access to GPUs today? Because the prices are rising. And the prices are rising for GPUs that are set up today. Every single GPU that's out there is fully sold out. And they are finding roughly four customers for every GPU that exists. And it takes an extremely long time to set up these GPUs. So even though Google, for example, even increased that equity raise to now $85 billion, they said AI is driving an expansionary moment of Google. The company is experiencing strong demand for its AI solutions and services from enterprises and customers at levels that are exceeding the company's available supply. So, they don't even have enough chips for the amount of customers and they can't even buy the chips fast enough cuz they're not being created fast enough. So, by going to other companies that have GPUs, they're saying, "Hey, you don't use those GPUs. give them to us and we will pay a six 7x the actual amount that the market is willing to pay for similar levels of compute. But look at this. Google is already making 45 $50 billion a quarter on average of cash from operations. This is real cash that they're bringing into the business. This is not enough. They had to raise $85 billion on top of making $45 billion a quarter. They used to buy back stock every quarter about 13 billion or 15 billion dollars worth. They completely stopped buying back shares. Soon their dividend will be cut off as well because that money that is being returned to shareholders, well, they want to use it to be able to buy more compute, more chips. Look at this. They ended a 10-year spree of buying back shares. They've been buying back and lowering their overall amount of share count since 2015. Then you look at this quarter and they've completely wiped away that streak just because they just issued $85 billion. And remember, Google is not alone here. Amazon, Microsoft, Google, Meta, all expected to spend hundreds of billions of dollars this year, including Oracle or even Cororeweave. Some of the giant Neocloud companies are also spending potentially 30 billion, $50 billion a year. And now Meta just rumored the other day is considering to raise tens of billions of dollars in a stock offering as it seeks new capital to help launch more data centers to get further into the artificial intelligence race. And they're not even selling their compute. That's all personal for their own use. Chris Camilillo even said, "Expect Amazon to do a similar thing, right? They don't do buybacks. They don't do dividends. And they obviously need the additional capital so then they can increase their capital expenditures, buy more data centers. This all goes back to being compute constrained. We can't make enough chips. There's not enough factories in the world to create them. We're going as fast as we possibly can and there's still not enough. If you end up looking at Google here, they said we are compute constrained in the near term. As an example, our cloud revenue would have been higher if we were able to meet demand. We are working through that moment and you know we are investing. We have a robust long range planning framework. We see extraordinary opportunities ahead. Extraordinary opportunities and that's why they're paying SpaceX $920 million per month. That's why Enthropic is paying SpaceX another massive fee on top of that is because they're seeing larger demand than they can even find compute for. Look at the growth rate. Not just the actual revenue of Google Cloud, but the growth rate is skyrocketing going from essentially 22% upwards of 63% today in just the last couple of years. In the last couple of quarters, two quarters, we've gone from 33% and then on better and better comps, we're seeing higher and higher growth rates. And they're doing this at higher and higher profitability. Look at the operating margins climbing from 3% upwards of 33%. So while companies like Nvidia charge more and more for their chips, Google's happy to pay because their customers are happy to pay. So the actual bill is not being flipped to Google. It's being flipped to the end user, which could be the consumer or the enterprise that wants to use their chips. So while Google's a standout here, you can really see why they're raising a bunch of capital because they're by far the fastest growing. But that doesn't mean that Microsoft and Amazon Web Services is not also accelerating their growth rates upwards of high 20 percentages or even 40% growth rates year-over-year. And that's also expected to continue to accelerate over time. And the only thing growing faster than their revenue is the revenue backlog, which means revenue that we're expecting into the future. Look at all of these companies. Google year-over-year went from $92 billion in backlog, revenue backlog to $467 billion. It's more than 4xed. Let's not talk about essentially a 63% growth rate. How about a growth rate of close to 400 or 500%. That is insane. Now, even Coreweave, a very, very small name, has a backlog of 98 billion. That's higher than what Google was just a year ago. They're a brand new name that just came out in the field. Look at Oracle, $552 billion or Microsoft, $633 billion of backlog. Of course, they're going to continue to spend to set up new data centers. They have the customers ready to pay. This is why they're spending roughly $150 billion a quarter now. There's a lot more to come from here as well. So, who's the clear winner? The infrastructure companies. I'm talking about the NVIDIA of the world accelerating their growth rates in data centers upwards of 90 plus% growth rates or Broadcom. Look at this growth rate going from 3% up to 78% year-over-year growth rates on total amount of semiconductor solution revenue. AMD accelerating revenue up to 57%. They're skyrocketing here. Micron, the memory components of all of these data centers skyrocketing in their overall growth rate. Lummentum, we need photonix, right? The connections between data centers, that needs to be very good. Or the infrastructure names in terms of the data centers themselves, the Google's, the Microsofts, the Nebuses of the world, massive growth rates. And this is only looking to continue as long as we get new models that bring on new demand. We just saw the other day that Mythos potentially accidentally leaked that Mythos 5, which is potentially what it's called, Claude Mythos 5, was released to their Discord for just about a few seconds and then instantly removed. Potentially an accident, but this could also be close to them being very, very close to actually releasing this to the public. Or what about Chad GPT 5.6 code name Iris Alpha or Ember Alpha? This could be a massive step forward for focus on agentic workflows, which is the really big breakthrough for overall uses that is only going to lead to even more chips needed. Look at this example of the amount of code that was contributed per person for Claude as the models started to come out. So Claude 4 here, 1.5 times the average of before 2025. And then Sonic came out, then Opus 4.5, and then we saw Mythos. This is the one that's not even out yet. And we saw a massive breakthrough between 2.5x the amount of code released to now in Q2 we're at 8x and it's only increasing. If you're telling me that every single business in the world can essentially more than 2.5x or even 8x their overall amount of code that they're releasing per developer. That makes developers a lot more useful. You would be willing to pay a very hefty fee if you got this much back from each individual employee that you had on your workforce. And Claude is showing even continuous improvements for things for trivial tasks, routine tasks, substantial tasks, or open-ended problems. All of these issues are seeing higher and higher success rates by the month. We're not even making improvements by the year anymore. We're talking about month- over-month improvements in these models, which should really make you question whether or not we eventually hit an era where we start to get into recursive self-improvement, where the models are essentially identifying their own flaws, teaching them themselves, and then making sure that they can reliably make the solutions faster and faster. Listen to what Enthropic said. The length of tasks that they can reliably complete on their own has doubled roughly every four months up from an earlier trend of doubling every seven months. So, not only are the models continuously getting better and better and better, they're getting better faster. This is an unbelievable unlock and it's only going to put the power back into the hands of the infrastructure companies that are selling the essential life force, the blood of these essential new systems. Look at what Coreweave is expecting as an infrastructure play. Currently, they're doing about $2 billion, but Wall Street has them expected to grow and accelerate that growth rate up to 190% year-over-year to bring growth from 2 billion up to 5 billion just year-over-year and then 5.76 billion the next quarter. Unbelievable growth rates and we are still very early. So everyone sees Core Wee, they look at their debt and they see, "Oh my gosh, you know, this company, if anything goes wrong, they're going to go belly up, but they see the opportunity ahead of them. They're following the trends of the actual code improvements and how much this is making employees better at Claude and Open AAI, and they're saying, "Hey, if we have an opportunity now to get as much money as we possibly can and put it into this next generation, I mean, generational wealth opportunity, they want to put everything into being as early as possible. That way they benefit the most. Even Oracle now has $153 billion in debt and this scale is only looking to increase over time. I'll show you this. Look at the AI buildout scale. Now Jensen Wong just said that a single gigawatt data center is going to increase in price not decrease in price from 50 to 60 billion up to 80 to hundred billion dollars per gawatt. OpenAI is a perfect example because they showed what their road map was to get to 2030. They originally said in January of 2025 they wanted to get to 10 gawatts of compute. They're now saying they want to get to 30 GW depending on whether you look at this between 50 billion or 100 billion. You're essentially looking at a $ 1.5 to3 trillion buildout for one company. Open AAI is not even the leader anymore. Enthropic is actually doing more business than them. So what is their gigawatts that they're expecting by 2030? What is Google expecting? What is Meta for Meta Super Intelligence Labs or Microsoft or Amazon or Mistral or Midjourney or Higsfield or Runway or Coher or 11 Labs? The list goes on and on and on of all of these AI labs that are seeing a ton of growth. Perplexity, another one. There's so much demand for these companies and yet just one of them is expecting to pay $3 trillion between now and 2030. It's not even that far away. If this is the scale that we're headed towards for which all of these chips need to either come through TSMC and then eventually going to Nvidia or Broadcom or AMD, these companies are going to be multiples higher than where they are today. If just the biggest data center companies in the world, we're looking at Amazon, Google, and Microsoft, are looking to spend to try to fund this AI growth, they can essentially help pay, including the amount of debt that they're going to be taking on, roughly $12 trillion worth of growth between now and 2031, according to CSU management. And now Trump and the governments are trying to get involved as well. Trump just said that the administration might buy equity stakes in US AI companies and he'll host a meeting with AI executives as soon as next week. That was said on June 6th. But now let's talk about what I'm buying. And to understand what I'm buying, we have to go through the entire five layer cake of artificial intelligence. Or that's how Jensen Wong puts it. First at the very bottom, we have energy. There's no chip power. If we don't have energy, that's going to be a massive component. But then we have the chips. This is the actual brain of the artificial intelligence. But then we need to store that brain somewhere and that becomes the infrastructure. The models are the actual working thinking process and then the applications are what comes out of that thinking. So in order to understand that we have to go down the list. Starting with number one we have energy. The ones that I'm looking at mostly are GE Veranova, Verdive Holdings, Vistra Energy and T1 Energy. all massive components of the artificial intelligence buildout for which I actually hold none of these right now, but it's not the bottleneck that I think is going to hit first. Energy infrastructure has happened for a long time. So there is things like the grid that are going to enable being able to set up the data centers and they have the capacity to do that right now, but the chips that are coming out today are required specifically for artificial intelligence. So I actually think that the bigger bottleneck for me right now is the chip layer. And this is where my portfolio starts and actually a massive massive percentage of my portfolio. Nvidia is over 50% of my portfolio just in that one name. I also have Broadcom as well because there's going to be these specific chips for specific use cases. And I think that custom AS6 or XPUs as they're called, Broadcom is going to dominate in this area as OpenAI, Enthropic, Google, all of these very specific chips come directly out of Broadcom. But this is what sits at the heart of the second layer which is chips. The third one we have to look at Nebus, Cororeweave, Irene as being the infrastructure names. And I know there's many more and we'll talk about them here in a second, but these are the pure play infrastructure names. Whenever we're taking a look at these companies, the interesting part is that they're ranked. There's a performance in terms of the actual difference in how these companies operate. And Cororeweave is number one. For number two in the sort of gold tier, we have names like Oracle, Nebius, Microsoft Azure, Crusoe. In the third place, we have names like Google Cloud, AWS, Lambda, Together AI. Amazing companies. And all of these are worth a significant premium, but there's only one in platinum tier. Just like the reason that Google ends up paying SpaceX so much for data centers, it's because it takes forever to actually get access to these GPUs. Roughly a 1 gigawatt data center can take multiple years, four, five, six years in order to set up a gigawatt data center. But Cororeweave is essentially being able to build data centers extremely quickly and has a very strong relationship with Nvidia to be the first company to get access to their newest chips, which obviously they can charge a lot more for because they are much higher performance than the previous generations. For number four, we have the models and these are the open AIs and the enthropics of the world. There are more and we'll talk about that in a second. And yes, I own them slightly but not directly. Indirectly, I own both of these names. In the model names, we have other names like Google and other ones that we'll talk about here in a second. But whenever we take a look at who owns these companies, Microsoft has a major investment in OpenAI. Nvidia has a 3.5% ownership in OpenAI. Google and Amazon own massive stakes in Enthropic. Google owns 14% of Enthropic that just recently got valued at almost a trillion dollars. It's a massive benefit to their bottom line. But then we have the application layer. We have Microsoft, Google, Amazon, and Meta. The only other one that I would actually put on here and feel comfortable with doing so. There's obviously a lot of names, but the real standout here is also Palanteer. Yes, there's Inno Data, and there's Sound Hound, and there's a bunch of other names, but the other one that I would add on to here is Palanteer. I don't currently hold that one, but I am looking at it very closely. Whenever you're looking at these application layers, the most exciting part about these sort of big tech names is that they're across all of these parts. Yes, Microsoft, Google, Amazon, they make their own chips. They own their own data centers. They are building their own models. And of course, they are the most dominant application layers for artificial intelligence than any other names in the world. Now, for energy, this is also a little bit speculative. They don't really have an energy business. That's for external customers, but they are buying their own nuclear plants. They are getting really, really heavy control of that energy stack and that's very important, but it's not really for external customers. Whereas, whenever we're looking at the chip business, both Amazon and Google last quarter talked about starting essentially an external chip business, which could make them add on another business layer that could add tens of billions of dollars to their bottom line. But of course, the real standout name here for me is Google. Google has a massive share in SpaceX, enthropic. They're across the application layer, chips, models, energy. They are dominating this entire space and they will be at the forefront of artificial intelligence. That SpaceX investment alone, which is not only getting involved with Terra Fab and getting interested in creating their own chips and dominating the space industry and creating their own Neocloud, they also, and by the way, they're paying SpaceX to get access to their compute. They also own roughly $100 billion in this company. So, it benefits them both ways. But I don't own 50% of my portfolio in Google. That still belongs to Nvidia. And that's because they are the most dominant straightforward pick for the AI infrastructure names. Not only are they dominating this space, they're the fastest growing, they have the best margins, and they're cheap as ever, but they also are investing in finding heavy returns on other AI infrastructure names. They're doing exactly what Google did 20 years ago, creating Google Ventures in finding names like SpaceX that they could make a hundred billion dollars off of. Now, Nvidia is getting in the chance of investing in very, very young companies that could blossom into major businesses, just like they did with Coref, investing $100 million back in April of 2023, back before anyone was paying attention to this company. That's made them over 20x on their original investment. artificial intelligence is here and it's here to stay and there's going to be a lot more money to be made and people get tied up with the current market caps of companies or what the overall rate of growth has actually been and can that be sustained? But what is sustaining if we're going from a world where we used to be run by humans to a world where we could be run by agents that are running 24/7 that don't take days off, don't have holidays, and are extremely cheap to run yet do better work at higher levels than what humans are doing. What rate of growth does that lead to? And who stands to benefit other than the names in the chip era, in the energy era? I think that this cortile that I originally showed you of the best AI companies outperforming the lowest rung is only going to get wider from here. Ladies and gentlemen, let me know what you guys think in the comments down below. But until next time, bye for