Clouded Judgement: In Defense of Model Lab Profitability

The article defends the thesis that large AI model labs (OpenAI, Anthropic, Google) will become highly profitable despite current losses, drawing parallels to infrastructure software companies' evolution. Ball argues that gross margins will expand, training payback periods are sustainable relative to revenue growth, and platform switching costs will create durable retention—making the oligopolistic market structure similar to hyperscalers.