Comment by ehsanu1

But those aren't things you can really separate for proprietary models. Keeping inference running also requires staff, not just for the R&D.