The moat or the commons

The U.S. AI industry raised ~$1 trillion in capex on the assumption that frontier models would behave like regulated monopolies supporting high margins. Open-weight models from Chinese labs and Western open-source infrastructure (vLLM, LangChain, Ollama) have closed the capability gap to 6–12 months while cutting inference costs 10–30x, destroying lock-in at repricing. Capital is now reaching for scarcity through regulatory enclosure, vertical integration, and bundled distribution rather than accepting commodity margins.