While the TechCrunch AI report points to data showing frontier models retaining a disproportionate share of spending, this analysis may be premature. The theory of a neat 'discovery vs. production' split presupposes a rational, frictionless migration of proven use cases to cheaper models, which ignores the immense practical inertia of enterprise deployments. Many applications are tightly coupled to a specific model's API and capabilities, making a switch costly and risky.
Furthermore, as the article notes, the data itself is thin and captures only a slice of the market; the entrance of giants like Nvidia with models like Nemotron could rapidly reconfigure the entire cost-benefit calculus. The apparent stability of this two-tier economy may be a temporary artifact of the market's current growth rate, not a durable feature.