Economics of Open-Weight Inference
Problem
Open-weight models face key challenges in inference cost, hardware resale, and unit-economics sustainability, potentially making them less competitive than closed-weight models in practice.
Pain points
- Lack of pricing and resale market for obsolete hardware
- AI hardware artificially inflated to stimulate new hardware sales
- Open-weight models experience downgrade pain and less sustainable unit economics compared to closed-weight models
Questions to verify
- Is there a viable market for reselling obsolete hardware?
- Do open-weight models truly have more sustainable unit economics than closed-weight models?
- Will the profit motives of closed-weight providers inevitably lead to enshittification?
Evidence
quote · @匿名
hardware as of today still has no utility value after its reported lifetime has elapsed, which prevents neolabs and smaller labs from getting older HW clusters as the banks are not willing to give out loans against them. There is no agreed upon pricing for "expired" A100 clusters or similar.
quote · @匿名
I feel like there is some artificial moat being built here to stimulate sales of new hardware, because an H100 at 1/16th the price will have comparable dollar/FLOP as Vera Rubin.
quote · @匿名
This makes the case that open weight models are not necessarily cheaper than closed weight modelsBut I'd still rather use them since it's inevitable the unsustainable margins of the closed weight providers lead to enshittification once they stop subsidizing demandI'd rather spend more today with a workflow whose underlying unit economics are sustainable and don't force me to inevitably look for other solutions once the closed weight providers start optimizing for profit and extracting value
Disclaimer: this card is distilled by AI from community discussions as “pain point evidence”. It is a starting point for need analysis and validation, not a “validated market need”. AI has analytical limits; please open the original discussion above to manually verify before key decisions.