On financial analyst benchmarks, top models from Anthropic, Google, and OpenAI are now almost indistinguishable in capability. This convergence suggests the frontier is commoditizing, questioning the return on investment for massive training runs and shifting value up the application stack.

While techniques like model distillation can reduce costs for near-frontier AI capabilities, this hasn't dampened demand for the absolute best models. The market shows very little desire for the third-best model, but exceptional demand for the top-performing one for any given task, demonstrating a winner-take-all dynamic.

The market for AI models follows a power law with a very strong preference for quality. Amodei compares it to hiring employees: people will disproportionately seek out the single best "cognitively capable" model, making price and other factors secondary.

Current AI pricing models, which pass on expensive LLM costs to users, are temporary. As LLM costs inevitably collapse and become commoditized, the winning companies will be those who have already evolved their monetization to be based on the value their product delivers.

Contrary to the idea of AI for all, the most powerful models will likely be restricted to a few high-paying clients to prevent distillation and maximize revenue. This creates a future where competitive advantage is defined by exclusive AI access, potentially allowing large incumbents to crush smaller competitors.

The current oligopolistic 'Cournot' state of AI labs will eventually shift to 'Bertrand' competition, where labs compete more on price. This happens once the frontier commoditizes and models become 'good enough,' leading to a market structure similar to today's cloud providers like AWS and GCP.

The key metric for winning the AI race is shifting from pure benchmark scores to efficiency. Perplexity's CEO argues that the company providing the most "token value per watt per user"—balancing accuracy, latency, cost, and intelligence—will ultimately dominate the market, making efficient intelligence the new goal.

When multiple models can solve a task reliably ('benchmark saturation'), the strategic goal is no longer to find the most intelligent model. Instead, it becomes an optimization problem: select the smallest, cheapest, and fastest model that still meets the performance bar, creating a major competitive advantage in inference.

The moment a new, more powerful AI model is released, user demand for the previous “state-of-the-art” version collapses. This intense desire for the absolute best model means only the frontier provider has significant pricing power, while older, slightly inferior models become commoditized almost instantly.