We investigate the large-distance asymptotics of optimal Hardy weights on Z d , d ≥ 3, via the super solution construction. For the free discrete Laplacian, the Hardy weight asymptotic is the familiarHere we prove that the inverse-square behavior of the optimal Hardy weight is robust for general elliptic coefficients on Z d . Namely, (1) annular averages have inverse-square scaling in general, (2), for ergodic coefficients there is an almost-sure pointwise inverse-square upper bound, and (3), for weakly random i.i.d. coefficients there is a probabilistic inverse-square lower bound. The results have continuum analogs and also yield |x| −4 -scaling for Rellich weights on Z d . The technique relies on Green's function estimates rooted in homogenization theory. Along the way, we identify for the first time an asymptotic expansion of the averaged Green's function and of its derivatives up to order d+1 at small ellipticity contrast. This leverages a recent perturbative approach to homogenization theory developed by Bourgain. The result implies rather universal asymptotics of the random Green's function via concentration bounds of Marahrens-Otto.