When the model takes away the moat (www.techradar.com)

A recent analysis highlights the fragility of AI-powered products that heavily depend on third-party foundation models. While this reliance has fueled a surge in AI innovations, it poses significant risks, as seen with Jasper, whose revenue plummeted after ChatGPT integrated similar features for free. Analysis reveals that 81% of AI startups that wound down in 2025 were “wrappers and apps,” indicating that many companies built on external APIs lack true defensibility. The ability of these wrappers to provide unique value is often shallow, undermined by competitors potentially replicating their offerings directly through the underlying model. The report emphasizes the pressing need for democratizing access to model training and ownership to nurture sustainable AI businesses. Although creating competitive models demands substantial resources, the article advocates for a shift that allows more companies to train and fine-tune their own models, reducing dependence on major providers. True democratization of model ownership may fundamentally alter the dynamics in the AI sector, enabling startups to maintain control and innovate without the constant threat of being subsumed as a mere feature by their foundational model providers.