In a multiple regression $Y = X\beta + \epsilon$, if the errors have constant variance $\sigma^2$, what is the variance-covariance matrix of the OLS estimator $\hat{\beta}$?