2 months ago
- AI inference is profitable, with frontier providers reporting 70-80% gross margins.
- Cost estimates show inference costs around $1 per million tokens, while pricing is $4.50-$27 per million tokens, supporting high margins.
- Open-weight models like DeepSeek-R1 charge less than half of OpenAI/Anthropic, yet still maintain over 80% profit margins.
- DeepSeek-V4-Pro inference costs on the market are about $0.87 per million tokens, close to the actual service cost.
- AI labs like OpenAI and Anthropic use high inference margins to subsidize training costs, keeping prices high for profit.
- Inference-only providers don't need to fund training, so inference remains profitable even if AI labs fail.