1. must hit a ceiling and the rate of increase will significantly slow,
2. or Bitcoin mining will break SHA-256 and render it insecure.
I find 1 to be a the most likely outcome however the most plausible path for 2 would be a novel cryptanalytic attack on SHA-256 which could be exploited to increase mining power. It seems very unlikely that 2 would occur because mining power approaches the brute force limits of SHA-256.
1. exploiting cryptanalytic weaknesses in SHA-256,
2. successes in new computing paradigms such as a large scale quantum computer using Grover's algorithm.
3. or turning the local galactic cluster into computronium for SHA-256 miners. Although at that point we have bigger problems. =)
If you think SHA-256 will not be broken anytime soon then that suggests that you believe that mining difficulty increases will slow down.
Can you elaborate on this a bit more? I'm not following what you mean by the "mining power approaches the brute force limits of SHA-256."