Jensen Huang Says AI Costs Are Falling 30–40x Per Year

Jensen Huang says the cost of AI intelligence is collapsing at an extraordinary pace.
According to the $NVDA CEO, the cost of generating AI tokens is falling by a factor of roughly 30 to 40 times per year.
At the same time, the usefulness and quality of those tokens continue to improve.
That combination could dramatically change the economics of AI adoption.
As intelligence becomes cheaper, companies can afford to use far more tokens for reasoning, coding, agents and other complex workloads.
In other words, lower token prices do not necessarily mean lower compute demand.
If falling costs unlock much greater consumption, total AI usage could expand faster than efficiency improves.
Huang also emphasized that safety must advance alongside capabilities.
He argues that humans still control how AI systems are deployed and that infrastructure such as NVIDIA’s OpenShell should be developed aggressively to keep autonomous agents operating within defined security boundaries.
For $NVDA, the long-term thesis is that cheaper intelligence drives more usage, more applications and ultimately greater demand for the compute required to generate it.