Model commoditization
Also known as: AI model commodification, commoditized AI, race to the bottom models
In 2023 it felt like whoever had the best model had the business. By 2025, that logic had broken down. Open-weight models closed the quality gap with proprietary ones. Inference prices dropped 78% or more for some providers. Newer, cheaper models often matched or exceeded older flagship models. The model stopped being the moat.
This changes the business calculus for builders. If the underlying model is a commodity, the value lies elsewhere: in proprietary data, in workflow integrations that are hard to replicate, in distribution, in the quality of the product experience, and in the depth of domain knowledge embedded in the application. 'Value moves up the stack' became a common framing as this reality set in.
Model commoditization also creates pricing pressure for companies that charge per token or per API call tied to model costs. If the model gets cheaper, customers expect prices to fall accordingly. Vendors who have pegged their margins to a model-cost metric face renegotiation every time a new release drops. The smarter strategy is to price on value delivered, not compute consumed.