Source: Upstartsmedia
As AI companies and data labeling shops build training datasets, they're paying significant premiums for code written in less common languages like Ruby and C++—mirroring how scarcity economics work in physical goods. Specialized technical knowledge and niche codebases are harder to source and validate than commodity data, making them disproportionately valuable for companies trying to train models on diverse programming tasks. The dead startup data marketplace is becoming a real intermediary layer in the AI supply chain, not just a novelty.