The Business of Building God
7 hours ago
- Frontier labs argue for regulation but also face pressure from open-source models and competition.
- The main moat of frontier labs is a lead of 1-2 models, aided by talent and compute advantages.
- Data capture from users helps improve models, but this data is abundant and may erode profits.
- Labs diversify into areas like ads, robotics, and biotech to sustain revenue.
- Achieving $100B+ revenues in other industries is extremely difficult.
- Customers are beginning to question AI spending and may restrict usage to cheaper models.
- Options for labs include serving cheaper models or seeking regulatory barriers to block competitors.
- Regulatory restrictions on AI models are hard to implement and counterproductive.
- The equilibrium depends on whether frontier models keep improving globally or only in niches.
- RSI (recursive self-improvement) is an unknown unknown that could drastically change dynamics.
- AI capabilities are multifaceted; solving hard problems doesn't guarantee broad usefulness.
- The business of AI will face ordinary economic constraints despite transformative potential.
- Scientific progress may be substantial but uneven across domains.
- Investors can only plan for a competitive world, not a monopoly by labs.
- The moat could be depreciating if based only on model lead or data capture; RSI could be decisive.