Idiots Sold Semis Last Week ... Here's Why

Any investor who thinks open-source AI models are bearish for compute are absolute morons. They're idiots and they should hire a financial advisor immediately. Anyone who sold semiconductor stocks last week because American AI companies like OpenAI and Anthropic are starting to feel the heat from Chinese competition should also join them. Seriously, talk to Chuck, Edward Jones, Fidelity. I don't care who it is, give them a call, turn off the video. You need some help. For the rest of you though, that want to maintain your self-directed individual investor account, today I'm going to clear up some misconceptions you might have and how Wall Street is looking at this AI trade completely backward and how, maybe more importantly, you can start to benefit. Let's start with the backdrop. Now, I've published several videos over the past few weeks of various CEOs saying that token costs from these AI labs have gotten absolutely out of control. They're too high. If you need further proof of that, here's the Databricks CEO on CNBC saying the exact same thing. >> Everybody wants to do cost controls and they want to control their budgets because, as one executive of a Fortune 500 told me, his costs of AI are increasing exponentially. His revenue is not increasing exponentially. So, if this continues, he will go bankrupt. >> So, the Databricks CEO is seeing the same thing everyone using enterprise AI is starting to experience. The costs are actually way too high and the ROI is simply not there. Now, many investors are making a mistake here. They're drawing the wrong conclusions when they hear executives say these things. They assume that just because token prices of frontier labs like OpenAI and Anthropic are too high, it must mean that the AI trade is about to collapse. And honestly, I can see why they think this way. But now I want you to listen to the second part of what the Databricks CEO is seeing from customers. And after seeing this, I think this is going to make things very crystal clear from an investment perspective. >> For the really difficult questions, let's ask the frontier American models and pay a lot for it. But for most of these mundane tasks that you're doing, like hey, take this field out of Salesforce and you know, summarize it into workday, use a cheap open source model. So that's what they want to do. >> Now let's unpack what he just said there, because I actually think he made two really key points that I think every investor watching a video like this needs to understand. First is that not all intelligence is created equal. So I'll give you a very simple example. If I needed to solve 1 + 1, I don't need an expensive calculator. I don't need an expensive college degree or a tool to solve that level of intelligence. And many businesses today have AI tasks that are these elementary 1 + 1 task. And he describes this as moving data from one software provider to another. Yet Anthropic and OpenAI are essentially running those very simple and mundane tasks through the most capable and most expensive model they possibly can. This would be like driving your Ferrari to the grocery store to pick up some milk. You You certainly would look cool as hell doing it, but look, taking your Tesla or your minivan would have done the job, too. Now so the second point that he made there, I actually think is even more important to understand. Just because simple task can use a cheaper AI model, it doesn't change the demand for compute at all. Over the weekend I did some research on this Kimmy AI model that everybody is talking about. And here's the thing, it's a massive frontier-level AI model that required the most cutting-edge compute from Nvidia to train, and maybe more importantly, it requires a massive amount of GPUs to run them. People hear this term open-source, and they think you can run this on your laptop or a MacBook. That's not the case. Kimi requires the exact same data set or setup that large language model companies like Anthropic and OpenAI need. The Chinese labs just aren't charging the 80% gross margins that the American companies are doing. So, what does this mean? Large language models are going to be forced to change their pricing model. Intelligence has a price depending on what it's being used for. And right now, Anthropic and OpenAI are being very lazy, charging customers the same price no matter what the tokens are being used for. They'll need much lower pricing tiers for tokens used on these mundane tasks and higher price tiers for things like maybe like medical research, complex math equations, or even these long agentic coding tasks. But what that doesn't change is the need and demand for compute. Even the owners of Kimi admitted this weekend that their servers are absolutely cooked and maxed because everybody was using the model. You even just heard the Data Brick CEO say exactly what he was going to use his latest fundraising money for. He was going to buy more GPUs. So, any weakness in the semiconductor stocks, memory names, or even the equipment makers can be confidently bought. Open-source and lower-cost models from China require the exact same compute intensity as the American models do. They're just willing to sell it at a lower price, largely because they are state-funded by the Chinese government and, maybe more importantly, OpenAI and Anthropic are trying to pad their financials as much as they possibly can so they can IPO and dump those shares onto you, the retail investor, for the highest price possible. So, I want to wrap things up and I want you to remember three key things from this video. Number one, competition from China or any other AI lab is actually a good thing for compute. Lower pricing leads to higher utilization, which is both great for hardware makers like Nvidia and it's even better for the compute providers like Amazon. Anthropic and OpenAI, they've been getting away with murder on pricing. This is like a one-size-fits-all pricing that reminds me a little bit of the early days of the internet. I don't know if you guys remember this. You might not be as old as me, but AOL would charge you by the hour to get on the internet. AT&T would charge you 10 cents per text message. Pricing is going to evolve, but that doesn't impact hardware at all. And finally, point number three, the Kimmy K2 model is not a DeepSeek moment. People mistakenly thought DeepSeek came up with some magical way to train and run AI models using less Nvidia GPUs. That turned out to be completely bogus and in fact, Kimmy K2 uses as much compute or more than the frontier models do and the size of compute requirements for these AI models is going to get larger, not smaller. That is bullish for compute and I'm confident investors who buy the dip in semiconductor stocks going to be rewarded. That wraps it up for today's video. Hopefully, you have a good one. We'll see you again soon. Good luck with your investments.