We consider a fast fading AWGN multiple-access channel (MAC) with full receiver CSI and distributed CSI at the transmitters. The objective is to evaluate the ergodic sum-capacity of this decentralized model, under identical average powers and channel statistics across users. While an optimal water-filling solution can be found for centralized MACs with full CSI at all terminals, such an explicit solution is not considered feasible in distributed CSI models. Our main contribution is an upper-bound on the ergodic sum-capacity when each transmitter is aware only of its own fading coefficients. Interestingly, our techniques also suggest an appropriate lower bound. These bounds are shown to be very close to each other, suggesting the tight nature of the results.