— Dusty hyperluminous galaxies in the early universe provide unique environments for studying the role of massive stars in the formation and destruction of dust. At redshifts above 6, when the universe was less than 1 Gyr old, dust could have only condensed in the explosive ejecta of Type II supernovae (SNe), since most of the progenitors of the AGB stars, the major alternative source of interstellar dust, did not have time to evolve off the main sequence. We present analytical models for the evolution of the gas, dust, and metals in high redshift galaxies, with a special application to SDSS J1148+5251, a hyperluminous quasar at z = 6.4 hereafter referred to as J1148+5251. We show that an average SN must condense at least 1 M⊙ of dust to account for the mass of dust in this object, when grain destruction by supernova remnants (SNRs) is taken into account. This required yield is in excess of ∼ 0.05 M⊙, the largest mass of dust inferred from infrared observations of Cas A. If the yield of Cas A is typical, then other processes, such as accretion onto preexisting grains in molecular clouds are needed to produce the mass of dust in J1148+5251. For such process to be effective, SNR must significantly increase, presumably by non-evaporative grain-grain collisions during the late stages of their evolution, the number of nucleation centers onto which refractory elements can condense in molecular clouds.