Whilst everyone fears AI inflation, Payden & Rygel sees exactly the opposite. The cost of using AI equivalent to GPT-4 has fallen from $60 to $0.05 in two years. The asset manager warns that this “token deflation” could become a deflationary factor in the global economy.
By Payden & Rygel
With the deluge of inflation data published this month, we continue to hear inflationary fears related to data centre construction: from higher electricity prices to more expensive hardware. However, as we always say, it’s much more complex!
The cost of using AI models that generate tokens for the end user may not be rising, but falling. How is this possible? Well, just as with measuring inflation in goods and services, it is necessary to adjust for quality improvements.
In the case of AI, this involves calculating the cost per token of a constant or adjusting the level of intelligence. And it turns out that, for all models with the same level of intelligence as ChatGPT-4 launched in November 2023, the cost of producing one million tokens has fallen by over 99.9%, from the $60 initially announced to $0.05 in January 2026, not to mention that some of these models far surpass ChatGPT-4.
Consequently, token deflation means that the use of AI is becoming cheaper, generating better results and increasing overall efficiency, which, if sustained in the long term, could provide a deflationary boost to productivity.
