Physical AI may have a data problem hiding behind the model hype. A robot can have a strong model and plenty of compute, but it still needs fresh information about the physical world: what a warehouse actually looks like, how a space changes, where objects are positioned, and what the environment looks like from ground level. That is where spatial data becomes infrastructure. @vangrid_io is approaching this from a different direction: use smartphones as capture nodes, turn real-world captures into structured spatial data, then let specific data requests create demand through bounties. The important part is not “3B+ smartphones.” It is whether a network can turn those devices into a reliable supply of on-demand ground truth. That is a much harder problem to solve, and probably the more interesting one to watch.