tokenized, time-segmented discounted inference tranches
- lets assume new inference providers want to tokenize
- lets assume they have some inference budget
- lets assume they are using some of that inference budget, but not all
- lets assume they want to reduce their opex
- they could: leverage internal analytics to allocate some portion of their inference daily dependent on realized and expected internal use
- depending on when this inference capacity is released into the tokenized wrapper, it really belongs in different tranches
- ex: one tranche for heavily discounted inference sold within short time frames
- ex_2: another tranche for reliable buckets of inference sold on longer timeframes/in longer contracts
- the yield is actually relatively simple here because it comes directly from sales, but how do you price the *flow*, the "principle" of the *source* of inference, esp. when its variable on a day/week/monthly basis.
extremely interesting design space
if any of this makes sense to you, come ship with me