The latest ‘crack in the thesis’ for the trillion-dollar AI boom: Tokens are getting cheaper

The AI economy is built on bets about the future, with companies like Nvidia, Anthropic, and OpenAI financing data centers before they have tenants. The premise is that demand for AI compute is virtually limitless, so companies will either pay more for smarter models or use them so much that it doesn't matter. Jensen Huang, Nvidia's CEO, refers to this as the "two exponentials" driving the price of AI compute: increasingly complex models and growing adoption.

However, there's a new development that casts doubt on this thesis: the price of AI tokens is falling rapidly. According to data from Ramp, the effective price per million tokens paid by American businesses has dropped by 41% since March, from $1.15 to 68 cents. Frontier models, which are the most advanced, now account for only 45% of usage, down from 53% in early August.

Additionally, the top 1% of spenders, responsible for 80% of the enterprise revenue of OpenAI and Anthropic, reduced their per-employee spend by nearly 10% in August. This trend suggests that tokens are becoming more like a commodity—interchangeable and less valued than previously thought. Morgan Stanley has flagged vulnerability for up to $300 billion in bonds financing neocloud buildouts if token prices don't continue to rise.

This shift in the AI landscape poses a challenge to those expecting unbridled growth without any constraints.