(I was asking a similar question before but maybe this is the best way to put it: )

I am wondering whether there exists constant $C$ such that for any $n$, there exists a square matrix $A$ of dimension $n\times n$ such that the spectral radius $\rho(A)$ of $A$ is bounded away from 1 (say, $\rho(A) < 1-1/n$ or $\rho(A) < .99$), and in the meantime, there exists a $k\ge 0$ such that the operator norm of $A^k$ is very big, say $\|A^k\|\ge n^{C}$. 

Thanks to the comments such bad situation could happen. Although my point here is not really to solve a brain-teaser but to try to build some constructive theory here. For example, my plan for rescuing the situation is as follows: 

Could one assume additionally  some matrix norm bound on $A$ so that this doesn't happen. That is, whether there exists a matrix norm so that for any matrix with some matrix norm $\le poly(n)$ (with a fixed poly) and $\rho(A) < 1-\delta$, it is true that $\|A^k\|\le poly(n)$ for any $k$. (Or slightly stronger, I would like to have $\|A^k\|\le poly(n) (1-\delta)^k$. 

Thanks a lot! Any pointers/references would be most appreciated!