If you need to perform the multiplications really a lot of times AND matrix A has an SVD decomposition A = U S V’ which can be truncated without altering the representation of A, then you might try to pay the price for the SVD decomposition and gain during the calculation of A’B A since there you multiply smaller matrices.