The future of AI pricing splits along one axis: whether users will keep paying large margins for closed frontier models. Coding agents have answered that question for at least one market. Past capability thresholds like Anthropic's Claude Opus 4.5 and OpenAI's Codex 5.2, users are not switching because of habit. They are switching because measurable output is higher. The author states he would pay $2,000 per month for current tools. That number is the thesis.
Closed labs, currently Anthropic and OpenAI with Google expected to close the gap, will continue to extract performance gains across every dimension: speed, intelligence, watts, and seconds. Benchmark saturation does not end the race, it redirects it. Meanwhile, open models will not chase Claude and GPT on the Artificial Analysis index indefinitely. Economic pressure and demand for low-cost, niche deployment will force a fork. Enterprises will pin to a model that clears a performance threshold for a specific task and stop there. Open model value will exceed OpenAI and Anthropic combined, but that value fragments across a wide commodity stack with thin margins at each layer.
The argument worth reading in full is not just the endpoint valuation of $2 to 10 trillion for each frontier lab by 2030 to 2035. It is the mechanical explanation of why closed labs have compounding integration advantages that open models structurally cannot replicate, and why API businesses at those same labs will quietly atrophy as labs protect their best weights and chase higher-margin use cases. The two ecosystems are not converging. They are diverging on separate exponentials.
[READ ORIGINAL →]