I don’t think people are archiving LLM output. The article indicates there’s increased storage because increased security risks. I’m dubious of that. To me the size of the models themselves probably drives this. But also, I think companies are likely retaining more of their user data in the hopes that they can use it to train models on later. For example, call centers might be retaining more call recordings in the hopes that they can train a “top salesman” model to either coach other humans or directly sell itself