The rapid decline in AI pricing may appear concerning for the industry, yet initial findings indicate a positive trend. OpenAI has recently reduced the price of its GPT-5.6 Luna model by an impressive 80%, while also lowering the cost of its Terra model by 20%.
Following these drastic price reductions, an intriguing phenomenon occurred: there was a significant uptick in AI usage. Analysts at TD Cowen examined usage statistics from OpenRouter, a platform designed for developers to access various AI models. Notably, OpenRouter has been offering discounts on models like Luna and Terra. Their analysis revealed that the effective cost of utilizing Luna plummeted nearly tenfold post-price cut, resulting in a remarkable 14-fold increase in usage. For Terra, the effective price dropped about three times, and the uptake surged approximately fivefold.
AI usage typically gets quantified in terms of "tokens," the basic units of data that these models process. Companies usually pay for access to AI models based on token consumption.
What stands out is that the surge in usage exceeded the decline in prices. TD Cowen estimated that OpenAI's revenue from Luna increased by roughly 34% compared to the week before the price drop, while Terra's revenue shot up by around 45%.
The situation with Luna is particularly noteworthy. It is rare for a company to experience a revenue increase after slashing prices by such a dramatic margin.
This scenario reflects a concept repeated by AI leaders, now often regarded as an industry cliché—they call it Jevons Paradox. I’ve hesitated to mention it until now, but the compelling nature of the data is hard to ignore.
The term, originating from 19th-century economist William Stanley Jevons, describes how efficiency improvements in coal use did not diminish consumption; instead, they led to a substantial rise in usage. When a valuable resource becomes cheaper, it encourages exploration of new applications.
While this might seem counterintuitive, it’s a well-accepted notion in Silicon Valley, where history has often validated this outcome.
Consider the evolution of computers. Once massive, costly mainframes that were accessible only for exorbitant fees, they evolved into business computers that primarily large companies could afford. Today, a smartphone, priced around $400, offers far greater power than those initial systems, and virtually everyone worldwide now owns and operates these devices. Did this dramatic decline in prices lead to a drop in revenue? Absolutely not. For instance, Apple reported $109 billion in revenue for its latest quarter, a staggering increase from just over $100 million annually at the time of its IPO in 1980, when computing was far more expensive.
AI appears to be on a similar trajectory. Reduced prices make it feasible to deploy AI for tasks that would previously have been prohibitively expensive. Companies are now able to leverage more affordable AI models to conduct broader document analysis, handle increased customer inquiries, develop software, and manage automated agents enhancing efficiency across various requests.
Ramp's research observed that OpenAI's GPT-5.6 Sol model drew more business investment in July compared to Anthropic's leading Fable 5 model, likely due to the lower cost of OpenAI's offering prompting higher usage.
This trend is critical, especially given that the costs associated with generating AI tokens are also decreasing. New computing technologies are improving production efficiency, which will likely lead to further reductions in token prices.
However, it’s important to note that TD Cowen’s analysis only covers a brief period of two weeks following OpenAI's price adjustments, so it's prematurely definitive about the sustainability of this revenue growth. The analysts themselves acknowledge the need to monitor whether this trend persists over time.
For now, the key takeaway is clear: as AI becomes more affordable, our usage of it is bound to increase.
