Are AI scaling ladders limited by money or mentality?

One post speculates about a possible flagship training problem and whether a broad architecture family works above 1T scale. Its author worries that compute-poor, talent-rich labs unable to afford a scaling ladder risk such failures. Another poster replies that, beyond a minimum budget, a good scaling ladder depends on mentality rather than money, and links to a thread about how small scaling-law tests can get without losing fit or predictability.