The price of AI is falling fast. That might sound like bad news for the industry. Early evidence suggests the opposite is happening.
OpenAI recently slashed the price of its GPT-5.6 Luna model by 80%. It also cut the price of its mid-range Terra model by 20%.
Then something striking happened: Customers started using dramatically more AI.
TD Cowen analysts studied usage data from OpenRouter, a service that lets developers access different AI models. They found that the effective price of using Luna fell roughly tenfold after the cuts. Meanwhile, consumption jumped about 14-fold. Terra’s effective price fell roughly threefold while usage increased about fivefold.
AI usage is commonly measured in “tokens,” the small units of information that models process. Companies often pay to access AI models based on the number of tokens they consume.
The remarkable part is that usage rose faster than prices fell. TD Cowen estimated that OpenAI’s revenue from Luna increased about 34% compared with the seven-day period before the price cut. Terra revenue rose about 45%.
The Luna situation is the most notable. When you slash a product’s price by 80%, it’s pretty unusual to actually generate more revenue.
To me, this clearly shows an idea AI executives have repeated so often that it’s become an industry cliché. I’ve barred myself from writing about this. Now, with such fascinating data points, I’m finally going to utter the phrase: Jevons Paradox.
Yuck. I feel a bit dirty now.
Anyway, the idea dates to the 19th century, when economist William Stanley Jevons observed that improvements in the efficiency of coal use didn’t reduce coal consumption. They helped increase it dramatically. Making a useful resource cheaper can encourage people to find many more uses for it.
This seems counterintuitive to many people. In Silicon Valley, it’s a common assumption because it’s happened again and again.
Take computers, for example. They used to be huge mainframe machines that cost an outrageous sum of money even to tap into for a few hours. Then came business computers that cost so much that only companies could afford them. Today, you can buy a smartphone for about $400, and that machine is more powerful than those old computers.
And guess what? Almost every single person on Earth uses these small computers and carries them around in their pockets. Did revenue plunge as prices slumped? No. Apple generated $109 billion in revenue in its latest quarter. That’s up from just over $100 million in annual revenue when the company went public in 1980 (when computers cost a hell of a lot more).
AI could be following the same pattern.
Lower prices make it economical to put AI to work on tasks that previously weren’t worth the expense. Companies can use cheaper AI models to analyze more documents, answer more customer questions, write more software, and run automated agents that perform multiple steps for every request.
Ramp found that OpenAI’s GPT-5.6 Sol model captured more business spending in July than Anthropic’s top Fable 5 model. That was likely because OpenAI’s model is cheaper, and therefore, businesses used it more than Fable.
This dynamic is particularly important because the underlying cost of producing AI tokens is falling, too. New computing systems can generate them much more efficiently, so token prices will plunge all over again.
The TD Cowen results cover only about two weeks after OpenAI’s price cuts, so it’s too early to know whether the revenue bounce will last. The analysts themselves said they want to see whether the trend holds over time.
For now, the emerging lesson is simple: The cheaper AI gets, the more of it we use.
Sign up for BI’s Tech Memo newsletter here. Reach out to me via email at abarr@businessinsider.com.
