Manual calculation of the covariance matrix is different from that of numpy method

Question

Given there is a matrix of n samples with m features, I am calculating the covariance matrix of features by hand so it should be m by