Even though the provided string is infinite, the string has a very particular structure. Let's analyze the example, specifically $\text{COWWCO}$ to $\text{COWWCOOCOWWC}$. In particular, note that the index 7 maps to index 6, but indices 8 through 12 map to indices 1 through 5.
For notational convenience, let $F^k(s) = F(F^{k-1}(s))$, and define $F^0(s) = s$.
We can compute the smallest $k$ such that $F^k(s)$ contains the given index. If $k=0$, then we are looking at the original string and can therefore return the character directly. Otherwise, take $F^k(s)$ and split it into two halves. The desired index must be in the second half of $F^k(s)$. If it is the first character in the second half, then we want to return the last character in $F^{k-1}(s)$. Otherwise, due to the rotation, we decrease the index by 1 and return that character in $F^{k-1}(s)$.
Applying that approach directly gives us an $O(\log^2 N)$ algorithm, which is fast enough for our purposes. It can also be improved to $O(\log N)$.
import java.io.*; import java.util.*; public class cowcode { public static void main(String[] args) throws IOException { BufferedReader br = new BufferedReader(new FileReader("cowcode.in")); PrintWriter pw = new PrintWriter(new BufferedWriter(new FileWriter("cowcode.out"))); StringTokenizer st = new StringTokenizer(br.readLine()); String s = st.nextToken(); long index = Long.parseLong(st.nextToken()); pw.println(parse(s, index-1)); pw.close(); } public static char parse(String s, long index) { if(index < s.length()) { return s.charAt((int)index); } long length = s.length(); while(2*length <= index) { length *= 2; } if(length == index) { return parse(s, length-1); } return parse(s, index - length - 1); } }