The Next AI Bottleneck

First it was memory. Now...

Everyone has been following the bottleneck trades over the last few months.

Memory is the most notable.

Stocks in this group have already seen massive gains as the market began to recognize how critical they are to the AI story. Whether it’s Micron, SanDisk, or Western Digital, they’ve all gone on to have monster runs.

But today, there’s a new bottleneck that isn’t being discussed.

For the past two years, the AI trade has been very simple.

If you wanted AI exposure you followed GPUs and the infrastructure that surrounded the GPUs. AI training and inference workloads were dominated by compute, and GPUs were the clear winner in that environment.

This led to a massive concentration of capital in the GPU space.

That environment is starting to shift a bit because AI itself is evolving. AI systems are becoming more complex and more dependent on what happens outside of the model or outside of a chat bot like Claude or ChatGPT. As that happens the bottleneck is moving in a direction many people aren’t paying attention to.

Yet.

In the remainder of this article, I’m going to break down what I believe is the next major AI bottleneck, the core thesis behind it, how I’m currently positioned within this group, the names most directly exposed to this bottleneck, which names I’m looking to add, and more.