We present wide-band radio observations spanning from 1.4 GHz to 350 GHz of the afterglow of GRB 991216, taken from 1 to 80 days after the burst. The optical and X-ray afterglow of this burst were fairly typical and are explained by a jet fireball. In contrast, the radio light curve is unusual in two respects: (a) the radio light curve does not show the usual rise to maximum flux on timescales of weeks and instead appears to be declining already on day 1 and (b) the power law indices show significant steepening from the radio through the X-ray bands. We show that the standard fireball model, in which the afterglow is from a forward shock, is unable to account for (b) and we conclude that the bulk of the radio emission must arise from a different source. We consider two models, neither of which can be ruled out with the existing data. In the first (conventional) model, the early radio emission is attributed to emission from the reverse shock as in the case of GRB 990123. We predict that the prompt optical emission would have been as bright (or brighter) than 8th magnitude. In the second (exotic) model, the radio emission originates from the forward shock of an isotropically energetic fireball (10^54 erg) expanding into a tenuous medium (10^-4 cm^-3). The resulting fireball would remain relativistic for months and is potentially resolvable with VLBI techniques. Finally, we note that the near-IR bump of the afterglow is similar to that seen in GRB 971214 and no fireball model can explain this bump.