Interesting approach. Who do you see as the primary users—AI startups, researchers, or individual developers? How do you ensure GPU availability matches reservations?