Suppose there is a computer system that performs some specific event in multiple iterations and requires some amount of power and time for each iteration to be calculated. At every iteration system learns from last iteration and in next iteration doesn't perform the calculation that is already performed in last iteration. So In that sense that system performs less calculations in every next iteration compared to previous.
Time taken for single iteration reduces with each iteration, So does the power consumed by system for each iteration also get reduced. In that hypothetical System, It it can be said that power consumed by each iteration is directly proportional to the time taken by each iteration.
How this power and time relationship can be described mathematically which shows that hypothetical system consumes power in directly proportional manner with time.
When you say "Time taken for single iteration reduces with each iteration", that could mean anything. Any decreasing function will serve for the purpose. Depending on the type of the function, you can reach completely different conclusions.
For example, suppose that the first iteration takes exactly 1 second.
Suppose also that the second iteration takes 1/2 of the second, the third iteration 1/3 and so on... The total time (with infinitely many iterations) is $\sum_1^\infty\frac1n=\infty.$
If the next iteration takes one half of time spent on the previous one, the total time is $\sum_0^\infty\frac1{2^n}=2.$
So the end result could be... anything. And we even don't know if the number of iterations is finite or infinite. Your question looks more like a thought experiment rather than a real math problem.