Hacker News
'Why Current LLM Costs Are Not Sustainable' Hits 107 Points / 189 Comments on HN
A widely debated essay argues today's LLM API pricing is propped up by VC and hyperscaler cross-subsidies, drawing 189 comments on HN. Corroborating analyses circulating the same week estimate providers absorb 90%+ of the true cost of every token processed, and that per-token prices have fallen ~10x annually since 2021 even as consumption rose ~100x for some workloads (the 'LLM cost paradox'). The builder takeaway: expect 12–24 months before meaningful price normalization, and weigh on-prem/hybrid inference for high-volume, sensitive workloads.
↳ Follow the thread