Closed, capital-intensive frontier AI labs risk repeating Intel's historical fab complacency: massive fixed-cost training runs face rapid commoditization from open-weight architectures and lean inference optimization.