Every stage of the modern AI stack preprocesses data independently. Here's why that's costing you — and what a unified encoding layer changes.