Is global asymptotic cloning state estimation?

We pose the question whether the asymptotic equivalence between quantum cloning and quantum state estimation, valid at single-copy level, still holds when all the copies are examined jointly. For an N-to-M cloner, we consider the overall fidelity between the state of the M output systems and the state of M ideal copies, and we ask whether the optimal fidelity is attained by a measure-and- prepare protocol in the limit of large M. In order to gain intuition into the general problem, we analyze two concrete examples: i) cloning qubit states on the equator of the Bloch sphere and ii) cloning two-qubit maximally entangled states. In the first case, we show that the optimal measure-and- prepare fidelity converges to the fidelity of the optimal cloner in the limit of large M. In the second case, we restrict our attention to economical covariant cloners, and again, we exhibit a measure- and-prepare protocol that achieves asymptotically the optimal fidelity. Quite counterintuitively, in both cases the optimal states that have to be prepared in order to maximize the overall fidelity are not product states corresponding to M identical copies, but instead suitable M-partite entangled states: the simple protocol where one estimates the input state and re-prepares M identical copies of the estimated state is strictly suboptimal, even in the asymptotic limit.