The Falling Cost of Intelligence
9/28/2025 • 2 min read
It is widely claimed by AI CEOs and pundits that the cost of intelligence is “falling to zero”, or “dropping exponentially”. It’s about time someone actually bought the data to the table.
Costs Are Indeed Falling Dramatically
The graph below shows the falling cost of tokens (bits of text) generated by the AIs that power products like ChatGPT.
If you want to upgrade your model making sure that you both increase intelligence and reduce cost, follow the connected lines from your current model toward the right.
We can see that the cost is falling dramatically, perhaps exponentially, over time.
The cost per token from an equally intelligent AI has dropped 150x in the past year.
However, this is a per-token cost; Today's most advanced models all make use of reasoning. Reasoning is where AI systems generate an internal monologue of tokens to "think through" the problem, in addition to the tokens in the final result given to the user. Reasoning increases the total number of tokens they generate, increasing cost.
Does the cost of extra reasoning tokens negate the falling token cost?
Good news. We can see that even whilst generating more tokens, the overall price continues to fall over time. The cost of extra tokens is far less significant than the price slashes.
The total cost of getting responses from an equally intelligent AI has dropped 54.8x in the past year.
Why Are Costs Falling?
For the same amount of artificial intelligence, prices are falling thanks to:
- Algorithmic progress -- Today's state of the art AI algorithms are more efficient, requiring less power to train and use. Compute required to train an equal or better model is dropping 5x per year.
- Scaling compute -- As supply of available compute increases, cost for the same amount of compute falls.
- Hardware efficiencies -- Today's computers (GPUs) use less power to do the same amount of work (to train models and use them after).
Conclusion
The cost of generating intelligent answers from AI is less than 2% of what it was this time last year. The cost of AI is falling exponentially, driven by improvements in algorithms, computing hardware innovations, and players rushing into the market to increase supply for the rising demand of AI.
Keep in touch to see how the trend progresses, how it will affect the margin of your new or existing AI products, how it might affect what you already pay for AI solutions, or what you should really be paying for them.