Quantum discord cannot be shared.

Bohr proposed that the outcome of a measurement becomes objective and real, and, hence, classical, when its results can be communicated by classical means. In this work we revisit Bohr's postulate using modern tools from quantum information theory. We find a full confirmation of Bohr's idea: if a measurement device is in a nonclassical state, the measurement results cannot be communicated perfectly by classical means. In this case some part of the information in the measurement apparatus is lost in the process of communication: the amount of this lost information turns out to be the quantum discord. The information loss occurs even when the apparatus is not entangled with the system of interest. The tools presented in this work allow us to generalize Bohr's postulate: we show that for pure system-apparatus states quantum communication does not provide any advantage when measurement results are communicated to more than one recipient. We further demonstrate the superiority of quantum communication to two recipients on a mixed system-apparatus state and show that this effect is fundamentally different from quantum state cloning.