The AI Layer That Actually Makes Money

Jensen Huang's five-layer cake, and where the opportunity really is.

6 May 2026 · first published on Subwave

Everyone wants to know where the big opportunity in AI is right now. The clearest framework I’ve found comes from Nvidia’s CEO Jensen Huang, who describes AI as a five-layer cake. Work through each layer and the answer becomes obvious.

At the very bottom is electricity. No power, no AI — it’s the silent foundation everything else depends on. Above that come the chips: Nvidia’s GPUs, Google’s TPUs, and the growing field of custom silicon. These are the actual brains doing the heavy lifting. Then comes infrastructure — the data centers that pull all those chips together into a coherent, productive unit. Layer four is models: Gemini, GPT, Claude, and the rest of the research labs building the foundations that make everything else possible.

And then, at the very top, comes applications.

If you can use these layers to actually solve a problem for somebody, that’s where the big opportunity is.

This is where companies like Perplexity and Waymo live. This is where builders take everything underneath — the power, the chips, the infrastructure, the models — and turn it into something that solves a real problem for a real person. It’s the icing on the cake, and it’s the most interesting layer by far.

The application layer is where AI value gets captured, not where it gets created.

The layers below are critical, but they’re also increasingly commoditized and capital-intensive. Building a data center or training a frontier model requires billions of dollars and years of lead time. Building an application that uses those models to solve a specific, painful problem? That’s a very different game — one where insight and execution matter more than infrastructure spend.

The cherry on top, as the framework goes, is finding an opportunity within that application layer that nobody else has properly addressed yet. Nail that, and you get to eat the whole cake. 1:23​

The stack isn’t just a mental model for understanding AI — it’s a map for finding where to place your bet. Most of the noise is about layers two, three, and four. The real action is at layer five.