AI inference is obviously profitable
The article argues AI inference is profitable, with estimated costs around $1 per million tokens and API prices far higher, supporting 70-80% gross margins. AI labs charge high margins to subsidize training, but inference itself remains a viable business even if those labs fail.
Background
- "Inference" = running an already-trained AI model to answer queries (e.g., using ChatGPT). "Training" = the vastly more expensive process of building the model. A common claim says AI companies lose money on every query and survive only on investor subsidies. This article argues that's wrong when you separate inference from training costs.
- Gross margin (70–80%) = revenue minus direct serving costs (electricity, hardware), not counting R&D or training.
- DeepSeek is a Chinese AI lab that publishes open-weight models anyone can download, creating price competition.
- Key distinction: AI labs charge high inference prices to fund training. Pure inference providers (who don't train) could profitably serve the same models cheaper — so inference would survive even if big AI labs collapse.