In this article, we propose a critical approach to the big data phenomenon by deconstructing the methodological principle that structures its logic : the principle of aggregation. Our hypothesis is upstream of the critics who make the use of big data a new mode of government. Aggregation, as a mode of processing the heterogeneity of data, structures the thinking big data, it is its very logic. Fragmentation in order to better aggregate, to aggregate to better fragment, a dialectic based on a presumption of generalized aggregability and on the claim to make aggregation the preferred route for the production of new syntheses. We proceed in three steps to deconstruct this idea and undo the claim of aggregation to assert itself as a new way to produce knowledge, as a new synthesis of identity and finally as a new model of solidarity. Each time we show that these attempts at aggregation fail to produce their objects : no knowledge, no identity, no solidarity can result from a process of amalgamation. In all three cases, aggregation is always accompanied by a moment of fragmentation whose dissociation, dislocation and separation are different figures. The bet we are making then is to make hesitate what presents itself as a new way of thinking man and the world.