Trains are a high fixed-cost, low marginal-cost business. They are a natural monopoly: markets has a few firms making high profits Planes are a low fixed-cost, high marginal-cost business. They're highly competitive: free markets provide great services at near marginal cost

Model training is a fixed cost. If AI basically depends on training compute, then the outcome is a natural monopoly: there will be one or a few large AI firms with the best models, charging high markups in the absence of regulation

But model inference - thinking - is a marginal cost. If it turns out that good AI depends mostly on inference -not on how long is spent training the base model, but mostly on how long is spend thinking about query questions - then AI becomes a marginal cost business, like planes

The resultant market structure is highly competitive. Many AI firms coexist, providing similar-quality services, priced at close to the cost of compute. None of the firms make much money: most of the surplus accrues to the AI users, that is, us