42 Sellers, 5x Price Spread: Open Models Enrich Big Inference Providers, Not Startups
AccBalanced · x · 2026-11-22
A counterintuitive market analysis: open models were expected to help small inference startups compete on cheapest tokens, but they actually favor big providers.
- Large providers (Fireworks, Together AI, Baseten) lock in cheap GPU via multi-year compute contracts, keeping margins even as token prices fall; startups rent short-term at high prices
- Big providers also charge more on trust factors like speed and reliability rather than price alone
- OpenRouter data: 42 companies sell DeepSeek V4.1 Flash, cheapest at $0.25/M output tokens, while Fireworks, Together, Baseten and DeepSeek itself charge the top $1.20, with Modal at $1.00
The accompanying bar chart (OpenRouter public API, Nov 19, 2026) shows a 5x spread from $0.25 to $1.20, illustrating how long contracts and head-brand premium concentrate profits among large inference providers.
More from Venture
- Applied Compute publicly thanks investors and early helpers behind its rise — ypatil125 · 2026-11-22
- Chollet: GPU spot prices keep rising as HBM sells out through 2029 — AI compute buildout isn't over — fchollet · 2026-11-22
- Cathie Wood bought NVDA at $227 and sold at $125, sparking tax-loss-harvesting questions — oilmutt · 2026-11-22
- Polymarket puts 76% odds on Tesla selling a Cybercab for $30k or less by year-end — Polymarket · 2026-11-22
- Burkov: Linux analogy explains why no one can build a business on fine-tuning open models — burkov · 2026-11-22
- A '90s-Dell-style' opportunity in mass-market sovereign AI appliances — arieljalali · 2026-11-22