Key insights
- Stanford's AI Supercycle course highlights that despite rapid AI app growth, semiconductor companies like NVIDIA capture the majority of AI revenue. The key question is when, or if, the application layer will dominate like cloud software. This suggests continued strength in semiconductor stocks in the near term, but potential long-term shifts in value capture within the AI ecosystem.
Apoorv Agrawal is teaching the Economy of AI Supercycle class at Stanford, and the first lecture is based on the AI revenue triangle.
https://youtu.be/-ubIUNA-zRA?si=dBx65akX8rIaf58K
AI Revenue Triangle Has Barely Moved Despite 10x App Growth.
Agrawal's central thesis: despite the AI application layer growing more than 10x over two years, the shape of the AI revenue triangle has barely changed. Of approximately $350B in new AI revenue added over that period, roughly 75% went directly to semiconductors - predominantly NVIDIA. The app layer, even with enormous user growth, has not made a meaningful dent in the distribution. The inference layer (infrastructure) remains the most competitive and structurally unstable segment.
This is the analytical foundation of the entire course. Agrawal uses it to frame the key investment question: how many years - 5, 10, 15, or never — before the AI stack reaches cloud-software-style proportions, where the application layer dominates?
Will history repeat and the application layer will capture the most value in future? If that’s true then we are in for a long ride.