Basically, this is Jevons paradox, which is why more efficient chips and models will not reduce the demand for compute.
The cheaper and better AI becomes, the more it will be used and the more use cases it will create. Demand can grow almost overnight, but the power and data
