This work falls within the context of predicting the value of a real function f at some input locations given a limited number of observations of this function. Kriging interpolation technique (or Gaussian process regression) is often considered to tackle such problem but the method suffers from its computational burden when the number of observation points n is large. We introduce in this article nested Kriging estimators which are constructed by aggregating sub-models based on subsets of observation points. This approach is proven to have better theoretical properties than other aggregation methods that can be found in the literature. In particular, contrary to some other methods which are shown inconsistent, we prove the consistency of our proposed aggregation method. Finally, the practical interest of the proposed method is illustrated on simulated datasets and on an industrial test case with 10^4 observations in a 6-dimensional space.