Token Prices Falling Even as AI Usage Explodes

JPMorgan tracked AI token activity in June and found volume grew 70% month over month and 20 times year over year. Spending grew similarly. However, actual prices per token dropped 5% year over year. US-based AI companies (like OpenAI and Anthropic) fell from most of the volume to just 35%, though they still capture over 85% of the money spent. Separately, an industry index that tracks what companies pay to use AI dropped 20% from its May peak. That index measures token bill for AI across the industry — list prices, model sizes, and actual usage combined.

Both data points are true and both matter. AI usage is genuinely exploding. But the money is flowing in a specific pattern: US companies (OpenAI, Anthropic) still get most of the premium dollars while cheaper Chinese open-source models take most of the volume. When customers can get 90% of the quality at 10% of the price, that is what they do for anything that is not mission-critical. This is the specific pattern the framework has been predicting for six months. Companies are spending massive amounts on AI infrastructure assuming that customers will keep paying premium prices. But customers are shifting to cheaper alternatives for everything except their most valuable use cases. The math on trillion-dollar AI buildouts requires premium pricing to hold, and premium pricing is collapsing while volume expands. The AI mania has been sustained by the story that everyone will pay whatever it costs to access frontier AI. The data now shows that story breaking down in real time.