In the rank-one Gaussian spiked covariance model with $d/n \to \gamma \in (0, \infty)$, the leading sample eigenvalue separates from the Marchenko-Pastur upper edge $(1+\sqrt{\gamma})^2$ precisely in the supercritical regime $\lambda > \sqrt{\gamma}$. In that regime it converges to \begin{align*} (1+\lambda)\left(1 + \frac{\gamma}{\lambda}\right), \end{align*} and the squared overlap between the leading sample eigenvector $\hat v$ and $v$ has a positive limit. If $0 < \lambda \le \sqrt{\gamma}$, the leading sample eigenvalue converges to $(1+\sqrt{\gamma})^2$ and $|\hat v \cdot v|^2 \to 0$ in the standard proportional asymptotic regime.