I think there are a lot of limits and challenges that make AI prompted games difficult, but speed/hardware should not be one of them.
Always assume speed will grow exponentially, through software techniques or hardware getting cheaper. Then ask yourselves what is possible.
Gmail was born in a time when storage was very expensive, but they bet storage costs would fall and gave users a huge amount of mailstore. It helped make them the leader.
Same is true of AI. Assume cost will fall to almost nothing, now what can AI do?
Sometimes costs don't fall in the short term, but in the long run, they've always fallen.