If the math would work out that way a cluster of 25 or so computers should be able to support a full blown AI. But clusters of 10's of thousands of computers are still simply executing relatively simplistic algorithms. So I would estimate that the number of transistors required for AI would be either much higher than the number of neurons (which are not easily modeled in the digital domain) or that our programming bag of tricks would need serious overhaul before we could consider solving the problem of hard AI.