I completely agree! Then, capital will flow into the internet, electricity, server racks, and data centers. In fact, this is very similar to the speculative path during the dot-com bubble in 1999:

1. First, speculation in upstream infrastructure: in 1999, it was portal websites → fiber optic cables → networking equipment; now it’s GPU → HBM → optical modules → servers.
2. Then, core bottlenecks continue to spread: in 1999, insufficient bandwidth led to speculation in fiber optics; now, insufficient computing power leads to speculation in GPUs, insufficient storage capacity leads to speculation in DRAM, and inadequate connectivity leads to speculation in optical modules/CPO.
3. In later stages, capital will flow into edge assets: in 1999, data centers, telecom carriers, and infrastructure; now, data centers, electricity, cooling systems, and server racks. All of these are profits from AI hardware infrastructure, but ultimately, they must be deployed in applications.
History always repeats itself, hahaha.