AI BriefingHadley Wickham: AI Token Prices Reflect True Costs, Not VC Subsidies
·2026.11.04 03:03
Key point
Self-hosting Kimi K3 costs roughly $0.25 per million tokens, significantly below the $1.13 API price, indicating healthy margins rather than unsustainable subsidies.
Details
- Token prices for open models like Kimi K3 are reasonably accurate reflections of underlying costs, with competition driving prices down to near-cost levels rather than relying on heavy VC subsidies.
- Self-hosting Kimi K3 on 8x GB300 GPUs costs approximately $23,000 per month, yielding a cost of $0.25 per million tokens at 75% utilization, compared to $1.13 per million tokens for API providers like BaseTen.
- Subsidies are occurring in subscription plans rather than per-token API costs; a $200/month plan provides $2,000 to $10,000 worth of tokens, a benefit available only to individuals and small businesses.
- Model training costs, potentially several hundred million dollars, are currently funded by investors including the Chinese government, which Wickham argues should be viewed as a societal investment in open models.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
Agent Swarm Experiment: Model Choice and Prompt Order Drive Emergent Leadership