Dropbox CEO: Companies should train own models with proprietary data

soumitrashukla9 · x · 2026-09-01

Drew Houston argues that with better open weights and mature post-training infra, companies with large datasets can now train their own models instead of just licensing data. While general frontier models retain an advantage in breadth, the future will likely see more specialized models leveraging proprietary data.

Original post →

More from Companies & People

Companies & People channel →