Less Dumb Investing

And here’s why you should feel safe buying the dip in semiconductors. Feel safe buying Nvidia and ASML, as well as AMD.

Let’s make a few assumptions.

1. Everyone loves AI (I know it’s not true)
2. Every startup loves AI
3. Investors love AI

It’s no secret that the most common word of 2024 was AI. Or perhaps it was GPT, AGI, or any number of other acronyms. I’m not exactly sure which of those acronyms was spoken most frequently among all populations, countries, and non-English speakers, but I’m confident it was one of them.

But seriously, the widespread adoption of AI as a commonplace tool has changed the world as we know it. These systems are powered mostly by GPUs created by Nvidia, Huawei, and AMD, with other players holding fringe market shares. These chips are extremely expensive, with some costing more than a Honda Accord. The high price of these chips has made it difficult for smaller firms to acquire enough hardware to develop their own dedicated AI models. At least, that was the logic prior to today’s announcement from Deepsink.

Deepsink is a one-year-old startup that unveiled an AI model called R1. As of this writing, R1 isn’t as advanced as ChatGPT or Claude, but it’s significantly cheaper to operate. One estimate suggests it uses only about 10% of the processing resources required by ChatGPT. That translates to substantial savings in both operating costs and initial investment, making it an intriguing option for firms looking to adopt AI without breaking the bank.

That leads us into assumption number two. I’m not saying every startup is actively implementing AI, but it does seem to be a common trend among startups to explore whether automation can improve productivity.

With that in mind, a small hedge fund in China has achieved something incredible: it has found a way to make something previously insurmountable… mountable? Suddenly, it’s no longer a crazy ask for a startup to work on building their own AI models from scratch, without relying on prebuilt models from the giants like OpenAI, Anthropic, Meta, or Google. Sure, most of these startups will fail. But if they choose to move in this direction, they will all spend money.

Enter assumption 3. Investors in these venture funds will throw money at the wall, hoping to discover the next ChatGPT or Deepsink. I don’t really need to expand on this one—it’s pretty self-explanatory.

Now, all of a sudden, Nvidia has a widespread base of customers instead of just a few giant players. Their customer base becomes less concentrated while still selling the same number of chips. Their top-tier products will likely remain reserved for the biggest players, but the overall pie grows in size. ASML’s fabs become more critical, and AMD continues to grow and cater to the smaller players.

With Nvidia down 16% today, it seems like a great time to buy the dip across semiconductors. But you’re not holding in the short run, you’re waiting for the medium term if you do so.