The Layers Get Paid Before the Winners Have Names
In 1996, you knew the internet was real, but Google did not exist and Amazon was still a bookstore. The trade was not picking the eventual logo; it was owning the layers every winner would need: fiber, routers, hosting, and the supporting infrastructure.
The AI economy is at its own 1996. Applications are everywhere, business models are unsettled, and the winners are unknowable. The layers are less mysterious. Models need interconnects, lasers, and memory. Agents consume compute, storage, and electrons. Robotics needs vision silicon, lidar, and batteries. Space needs launch, coverage, and imaging.
That is the attraction of the layer trade: every contestant pays the suppliers. It reduces picking risk without eliminating drawdown risk. The same infrastructure logic still endured the 200002 crash, and survivors took years to prove out.
The lesson is to own what future winners will need before those winners have names, while sizing positions for the risk that even a sound layer can be caught in a cycle.
