This is not the first time I've heard of TSMC reserving a very large portion of their production capacity for a single customer, especially in the early days of

This is not the first time I've heard of TSMC reserving a very large portion of their production capacity for a single customer, especially in the early days of a new process node. I guess there's an inherent risk in being the launch customer for a new node (unexpected delays, low yield, bugs, etc.) that can only be offset by exceptionally large volume. Maybe there's a gentleman's agreement among Apple, Samsung, AMD and nvidia that somebody's gotta buy those early wafers, but nobody can do that every single time?