The Substitution Line
The model-layer commoditisation that broke into public view at the end of April kept compounding through this week’s earnings calls. Two CTO-level confirmations on different stages put GPT-4-equivalent inference at roughly one-thousandth its early-2023 cost. The released value did not disperse evenly. It flowed up the stack to vendors with proprietary context above the models, and crashed down on vendors whose own labour cost was sitting in the path AI was busy automating. Three Inferential daily posts and four more market events between May 1 and May 7 located the same line, drawn from the same mechanism, with vendors landing on opposite sides of it. ...