[lern_too_spel]

Training is not the issue. Each company already has a trained model that is useful for a wide range of tasks. Inference is the problem. Nobody has the infrastructure to run these models at scale. Until they do, they have to pick and choose the most valuable places to deploy them.