The AI Toll Takers: 5 Stocks I'm BUYING NOW

Last week I told you I was done covering semiconductor stocks and I got a lot of comments on that video. Some people thought I was selling my semiconductor stock portfolio and that's not what I said. Some people called me a loser piece Also fair, but also not accurate. What I said was the AI hardware question has been completely settled. 2027 is written in stone and the chips have already been sold. Both Nvidia and Marvell confirmed this when they reported their earnings last week and it's the big reason why both stocks are down from that report despite very good results and a great forecast. There's simply no more money sitting here figuring out whose chips are a little bit better than the other one. The bottleneck trade that every financial blogger and YouTuber has talked about over the past two or three years, it's now a full-on roadblock and that's the key distinction that every investor watching a video like this needs to understand. The AI bottleneck trade is over. It's done. AI chips now face a roadblock preventing them from producing more. And if you want to sit on the side of the road and watch the semiconductors traffic jam clear in slow motion, go ahead. Be my guest. You're wasting your time. Smart investors are going to shift their focus to what is coming next. And here's the thing. The two biggest acquisitions in AI over the last month have already told you what is coming next. Nvidia bought Hugging Face and Stripe bought Open Router. Neither one of those is a chip hardware company and neither one of them is part of this quote AI bottleneck and more importantly, neither of them is stuck behind this supply chain roadblock that is going to last for many more years. Nobody is waiting on a fab in Taiwan or a memory company here in the United States to ship a model on Hugging Face. So, why did both of these companies in the same month go spend billions of dollars on companies that don't make AI hardware or don't make semiconductors? It's because they both figured out where the toll booths are going to be in the next era of AI. And both companies were smart to get in before anybody else figured this out and priced it in. That's what today's video is going to be out. What are the toll booths? Why did Nvidia and Stripe just pay for them, and which stocks can actually benefit, and which ones you can buy today to benefit from this shift. Now, let's begin with the one line on the Nvidia conference call that I actually don't think enough people are talking about. Jensen Huang said that he believes the majority of traffic coming through AI data centers now are agents rather than humans. Here's Jensen Huang saying it on the earnings conference call. >> Today, the vast majority of AI is prompted by people. I believe that this last month has crossed. Most AI are now agentic. But in the future, sure, every company will have a whole bunch of agents. You know, we have we have 40,000 employees roughly. In the future, we'll have 400,000 agents, 4 million agents. And those agents are running continuously. They're running in the background. >> Now, think about what this means. The era of humans sitting there and typing into a computer or typing into a chatbot, that era is quickly becoming extinct. That's now the minority traffic flowing into data centers. Most of it now is agents, software talking to software. Agents can run around the clock. Humans simply don't and probably shouldn't. And that's showing you where the AI market is heading. But under all of this is a technology shift that is about to accelerate about 5 months from now. Here's the reason why I'm doing this video today instead of waiting till December or January or next year when everybody else will be talking about this. Things are about to change in AI and when it happens, everybody's going to react to it. The headlines, the analysts, the dudes on CNBC, the kids here on YouTube, they're finally going to talk about it after it happens. Now, remember that DeepSeek moment? What if you could have predicted that was going to happen? Here's the crazy thing about AI. The major advancements keep showing up at the same point on the calendar every single year. In early 2025, you had DeepSeek, you had Anthropic launch Claude Code in January of this year. A year later, almost an exact 1 year later after DeepSeek and Claude Code, you had Open Claude go viral and every major lab and AI company has copied it since then. You had three different companies, three completely different products related to AI and they landed in just a couple of weeks 1 year apart. wasn't an accident at all. They showed up a few months after brand new AI hardware went into the data centers. The new hardware goes in and then the software that can finally take advantage of the new systems, it shows up a few months later. Right now, Nvidia's new Rubin systems are being installed in Microsoft and CoreWeave data centers and really data centers across the United States. And in an AMD event that I went to last month, they said that their next generation Helios data center systems are going to arrive later this year as well. Which means the next wave of AI software is about to land 5 to 6 months from today. Smart companies like Nvidia and Stripe, they're already preparing for it. They know what is about to happen. And forward-looking investors are going to start to pay attention to the stuff while retail investors are still going to be focused on the semiconductor roadblock that isn't going to clear for several years, no matter what the companies decide to do. Now, if you're new to the channel, I've actually spent 25 years of my life in technology, in web design, and investing, front end, back end, database, all this type of stuff. I use it heavily in my daily work from an investing standpoint, and even hobbies. Last night, my sons and I were doing CAD design using AI. So, what I'm going to tell you is coming from things that I'm using, but also hearing from other builders within this community. And there's a massive shift that is happening, and I think it's going to be obvious what companies are going to benefit. Right now, most people and most businesses interact with a single AI model. They bring up ChatGPT, or they have Gemini on their phone, and that one model handles everything. You pick a single one, and that's it. But, that's not what the next era of AI is going to use. What's being built by groups of agents and people working on the same project at the same time are a reliance on different models depending on what the job is. So, you might use a cheap and fast open-source model for the boring parts. You might use the more expensive frontier model for the very hard parts or the creative parts, you might use something very specialized for coding, something entirely for reading a contract, and there's legal AI. And you might use something else for image or video generation, or your marketing team. And here's the part that really clicked for me this week. Tobi Lütke, he's the CEO over at Shopify. He put out a post recently saying that they're fine-tuning small AI models that outperform the big ones on very specific tasks. Now, Tobi's not your normal CEO. He's actually very highly technical, and when he's sharing things that the team at Shopify is finding, the rest of the industry is going to copy them. And think about what this means. Shopify is a multi-billion-dollar company that has all the money in the world to pay for OpenAI, or to pay for Anthropic, and they can just pay for it, and that can be it. But instead, they're taking the small models and training them on specific in their own problems, because the small models that they specifically trained are actually doing a better job than the frontier models. That's not just going to be a Shopify thing, that's going to be what the whole industry does as we move into next year. The idea that a single model, like Claude or OpenAI, is going to handle everything, everything a business does, or everything a strongly ambitious person or startup does, it's just simply not going to be the case, and it's not good business to rely on a single point of failure. What's going to happen instead is you're going to rely on a handful, or maybe even dozens of models, some big, some small, and some that you've trained yourself. That's where the real work is going to be routing these models to the ones that you need. Sending each piece of the job to the best and the cheapest model that you need to use. And here's the key thing to understand from an investment standpoint. Every single time one of those jobs gets routed, somebody is going to be sitting in the middle providing some value and they're going to be able to take a cut. That is the toll booth. Now, let's go back to the acquisitions. Stripe is a payments company. They bought OpenRouter and OpenRouter is essentially like a switchboard or it operates actually very similar to Stripe's businesses of payment processing. Developers like myself want to communicate with multiple AI models from the frontier ones like ChatGPT, but also open-source ones like Nvidia. That's what OpenRouter allows you to do. It works the same way that a credit card reader at a restaurant or a retail shop works. The terminal doesn't care who the issuing bank is or the payment processor on the card is, it just routes the charge, it takes a cut, and it moves on to the next payment. Stripe built their entire company on being the terminal and they just went out and bought the terminal for artificial intelligence. And notice what they didn't buy. They didn't go invest in some model company, they didn't even build one internally, and more importantly, they didn't invest or make some custom silicon for all this. Instead, they just bought the model routing layer. Now, the other acquisition was Hugging Face. Hugging Face is easily the dumbest name for a company that I have ever seen, but Nvidia bought it for I think $13 billion. Now, Hugging Face is basically you You think of it like a shelf or a closet for where all these AI models happen to live. There's thousands of them. They're open, many of them are free. You can download them and you can run them in your organization just like Shopify app. Now, for the non-technical people out there, I would say Hugging Face is kind of like the Netflix of AI models. You go there as an a developer and you download and use the models that you need. So, Nvidia, the company that sells the picks and shovels of AI, they just bought the shelf where all of the AI is going to live. And Stripe, which is a company that is used to taking a cut out of every transaction, they just bought the switchboard of AI. Two of the biggest companies on Earth just placed the exact same bet where AI is going and they both bet on not a single model winning. In fact, dozens, if not hundreds of them, winning. Now, thumbs up on this video if you get any value from it. Don't forget to hit the subscribe button. We're about to get to the payoff. This is who wins here because this is the part that matters for you, the investor. And the test that I'm using is actually very simple. Does the company that you're look at get paid more when the number of models it uses goes from just a single model to, let's say, dozens or maybe even 50. If the answer is yes, they're sitting on a toll booth and that's the place to be. If the answer is no, they're just one of these many companies that are going to exist that are simply paying a toll. Think about it like this. There's going to be three groups. Group one is the toll booth themselves. These are the companies that sit between the businesses like maybe Shopify and the AI models like OpenAI. And they take a cut no matter which model wins. Stripe and Nvidia just bought their way into this, but Amazon was already there. Bedrock is their AWS AI service, and it's has a menu of AI models that everybody can use on the platforms. And that was the knock on a company like Amazon for a while that they didn't have their own frontier model. Now, the market is seen that is the entire point, and it's a great investment. They were never betting on a single model. So, now they don't have a model to protect, one that they have to invest a bunch of money in. Whichever one developers want to use, they're going to sit in the middle and take a cut. Now, Microsoft's doing a similar type of thing. They've had a long road to get there, but they now have a large model catalog, and they even have a model routing product as well inside of Copilot that picks the right model for you and for your business. And one company that's a sleeper here is Cloudflare, because they also just built their own open router system that actually sits right on their network in a similar way of Amazon. And a lot of these are small models that do the exact type of things that you might not be able to do or want to do inside of a big data center. You might want to do it on the edge, which is what Cloudflare is hoping for. Now, group two are companies that own the data and that have the engineers to train and use these small models as they become more available. Shopify, obviously. company could be Intuit. Now, Intuit's stock has been decimated, but they have been working on their own models for tax and accounting data for many, many years. If they get their act together, they could have a compelling product, but I think the crown jewel that exemplifies all of this is probably Palantir. They get on their conference call every quarter and say, "We don't care which model that you use. We're just going to protect your data and create the best product for you." And then you have companies like Snowflake. Again, more of a pick and shovel type company because if you're going to fine-tune models with your own data, it means that data has to sit somewhere kind of in an organized fashion first, and then you get to route that to the model that might do something useful. And now finally, the group that I'm staying away from. These are the AI application companies that are really just a wrapper on a frontier model. They don't have any underlying data or very little proprietary underlying data underneath it, and they're just going to route their application through a model. In a world where there's lots of AI models out there, these guys are the toll payers and not the toll takers, and I would rather invest in somebody that is taking the money rather than having to pay it. So, here's the bottom line. For the last 2 years, the whole AI trade was about who made the chip. Where was the secret little sneaky bottleneck in there? That question has been answered. It's over. You can't make any more of these chips. That discussion is sold out into 2027 and really out into 2028, maybe even as far as 2030 at this point. The next question that investors have to decide is who takes a cut every time a piece of work is routed in an AI application. And Nvidia and Stripe just answered it with the big acquisitions that they made recently. Now, this is just the first video that I'm going to make on what comes next with AI cuz I don't see enough people talking about this. There's still people talking about TPUs and GPUs and CPUs and XPUs. It doesn't matter. You could make the worst GPU ever made and it would still get sold. So, there's nothing left to discuss there. Instead, I've got three predictions coming about the hardware cycle. Some of them how memory is going to change, some of it is how these chips are actually going to have to be manufactured in the future. And so, make sure you subscribe to the channel if you don't want to miss it. Thanks for tuning in today. I'll see you again later this week. Good luck with your investments.