$s^2$ is always larger than $\sigma^2$.

$s^2 = \frac{n}{n-1}\sigma^2$

The population variance uses $n$ as the divisor.

$s^2$ is an unbiased estimator of $\sigma^2$ if divided by $n-1$.