(By Sylow I, there is a subgroup of order $9$. By Sylow III, it is unique.)
Without Sylow. The other answer nicely shows the uniqueness of a subgroup of order $9$, but not its existence. As for this latter, a Sylow-free argument is the following.
- If $G$ is abelian, then it has a subgroup of order any divisor of $|G|$ (see e.g. here).
- If $|Z(G)|=9$, then we are done.
- If $|Z(G)|=6$, then $G/Z(G)\cong C_3$, and hence $G$ is abelian: contradiction. So, there's no group of order $18$ with center of order $6$.
- If $|Z(G)|=3$, then necessarily $\operatorname{Inn}(G)\cong$ $G/Z(G)\cong$ $S_3$, which has (class-preserving) elements of order $2$, and hence there are conjugacy classes of size $2$$^\dagger$, and finally centralizers of order $9$$^{\dagger\dagger}$.
- If $|Z(G)|=2$, then noncentral elements' centralizers can have order $6$, only. Therefore, the class equation yields:
$$18=2+3l$$
contradiction, because $3\nmid 16$. So, there's no group of order $18$ with center of order $2$.
- If $|Z(G)|=1$, then elements' centralizers can have order any divisor of $18$, and hence the class equation yields:
$$18=1+9k+6l+3m+2n \tag 1$$
If $n\ne 0$ we are done, because then there are centralizers of order $9$. If $n=0$, then $(1)$ leads to a contradiction, as $3\nmid 17$ (so, there are no centerless groups of order $18$ without any conjugacy class of size $2$).
$^\dagger$Otherwise, $G\setminus Z(G)$ would be split into $5$ classes of size $3$; but then an inner automorphism of order $2$ would fix one element in each
class, and hence there would be some $a\in G$ such that $|C_G(a)|=8$, contradiction because $8\nmid 18$.
$^{\dagger\dagger}$Actually such a $G$ does exist: it is $C_3\times S_3$.