Computing approximate stationary points of min-max optimization over the hypercube is PPAD-hard for quadratic polynomials. This result holds even for multilinear polynomials where each variable appears in at most three monomials, with inverse polynomial approximation factors. As a consequence, two-team zero-sum polymatrix games are proven to be PPAD-hard.