I’m guessing they’re running NLP inference on some GPU server that costs X per day and can process Y pages. I would guess that it’s the indexing that incurs the

I’m guessing they’re running NLP inference on some GPU server that costs X per day and can process Y pages. I would guess that it’s the indexing that incurs the most of the costs, not the actual queries performed by the users - though I could be wrong.