Wickham Says Kimi K3 Token Prices Roughly Match Serving Costs
Summary
Hadley Wickham argues Kimi K3 token prices roughly match the cost of serving them, citing a self-hosted 8-GPU Supermicro server at $550,000 and about $23,000 a month, or $0.25 per million tokens at 75% utilization. Subsidies sit in subscriptions instead: SemiAnalysis estimates a $200 monthly plan delivers $2,000 to $10,000 in tokens.
Key Points
- Moonshot AI's published vLLM recipe for Kimi K3 recommends at least eight GB300 GPUs, Wickham notes.
- Wickham calculates 124 billion tokens a month from SemiAnalysis's estimate of 6,000 tokens per second per GPU.
- Wickham says the subscription subsidies apply only to individuals and small businesses, not enterprises billed per token.