Remember when I talked about unmetered and free tokens on the podcast being a major driver in open-source and local hardware trends? Here is the proof point -- cheap tokens = massive usage When you have local models running unmetered or on-prem servers, your consumption changes completely