The Cost of AI Tokens in Free Fall: Should We Be Worried?

AI Token Costs Plummet Below One Dollar

The cost of using artificial intelligence models continues to drop rapidly. Silicon Data's LLM Token Expenditure Index has fallen below the symbolic threshold of one dollar per million tokens, now at $0.97. The index, which measures the effective price weighted by the usage of different models, has lost over 50% since its peak last summer.

This decline is explained by increasing competition among providers, the advancement of open-source or open-weight models, and the emergence of cheaper solutions for simpler tasks. Companies are also learning to better distribute their requests among different models to reduce their expenses.

OpenAI and Anthropic Not Exposed in the Same Way

For the AI giants, this deflation does not have the same impact depending on their business model. At OpenAI, a significant portion of revenue comes from fixed-price subscriptions, such as ChatGPT Plus, Pro, or enterprise offerings. In this case, a decrease in the public price of tokens does not directly reduce the amount paid by users. However, revenue from APIs is more exposed, as clients pay based on their consumption.

Anthropic appears to be more dependent on professional usage and the consumption of its models via API. Therefore, the drop in token prices may exert more direct pressure on its revenue at constant volume. But this reasoning overlooks one essential element: consumption is increasing very rapidly.

The Example of Uber Facing Price Declines

The case of Uber illustrates this phenomenon well. The company has deployed Claude Code to several thousand engineers and had already consumed its planned AI budget for the entire year of 2026 by spring.

Uber then revised its strategy by using more cost-effective models and optimizing request routing. As a result, the number of weekly requests addressed to its AI agents multiplied nearly tenfold in a few months, while expenses were largely stabilized. In other words, a decrease in unit price can be offset by an explosion in volumes.

Margins Under Pressure

This is where the real challenge lies for OpenAI, Anthropic, and their competitors. Their revenue does not depend solely on the price of the token but also on the number of tokens consumed. If prices decrease by half but usage increases five or tenfold, revenues can continue to grow.

The risk lies more on the side of margins. Labs have invested tens of billions of dollars in data centers, chips, and cloud contracts. These costs are often fixed over several years, while model prices can be quickly reduced under competitive pressure.

Thus, the drop in token prices is not necessarily a sign of a crisis for the AI industry. On the contrary, it may accelerate adoption by making applications and agents cheaper to use.

The real question is whether OpenAI, Anthropic, and other labs will be able to sufficiently grow volumes while maintaining margins capable of funding their massive investments.