According to the documentation of Matlab regarding the function cov(): "if A is a vector of observations, then cov(A) = C is the scalar-valued variance". But according to Wikipedia the covariance matrix of a vector of random variables is the matrix whose entries are the covariance between the $i$ and $j$ columns of the vector.
What exactly am I missing?
What you're missing is that in the MATLAB documentation of the case of A being a vector of observations, the random variable vector has length (dimension) = 1, and there are several observations of this one-dimensional random vector. In this case, per the Wikipedia definition, the vector has only one column, so the covariance matrix consists entirely of the covariance between the 1 and 1 column, i.e., the variance.
Think in general of an m by n matrix, in which there are m data points, each consisting of an n dimensional random variable, In your called out MATLAB case, m = 1.