The article argues that generative AI resembles shipping containerization—a late-wave ICT innovation that created immense societal value but captured little for its inventors—rather than the microprocessor-driven PC revolution that minted thousands of fortunes. Containerization (SeaLand, 1956) slashed freight costs and enabled globalization, yet only founder Malcom McLean and investor Daniel Ludwig (40x return, 1965–1969) profited; competition and massive capex ($10B in ships by 1972, ~$80B today) drove margins to shippers like IKEA and Walmart, not carriers. Similarly, AI model builders face oligopolistic consolidation under Big Tech (Microsoft/OpenAI, Google/DeepMind, Anthropic/Amazon), with training costs and compute barriers preventing distributed, permissionless experimentation akin to the 1970s PC hobbyists (MOS 6502 at $25 enabled Wozniak’s Apple). Economists project AI adding 1–7% to global GDP ($1–7T), but value will accrue to customers and incumbents, not model vendors. Domain-specific apps (Cursor, Harvey, Perplexity) face cheap fine-tuning and economies of scope, making them acquisition targets. The author contends AI is in an evolutionary, not irruption, phase: no surprise, no protected profits, immediate competitive free-for-all. Investors should avoid model-layer bets unless already positioned; returns will flow to companies leveraging cheap AI infrastructure, not building it.
Read original
hackernews