The development of big cities (or wide megapolis) makes appearant very segmented "patchworks" of productive areas evolving separately, each one close to the others, without any clearly appearent order. This communication aims to highlight that, in spite of an appearant disorder, productive activities in such wide cities evolve according to an historical trajectory the determinants of which can be pointed out. This trajectory is progressing from very specialized clusters in which highly institutional density creates rigidities and irreversibilities unfavorable to the capacity of adapting toward weakly specialized clusters in which, reversely, very open social networks leads to a collective ability of local productive actors in adapting and in engaging themselves in innovative ways. This trajectory appears as a "territorial collective learning". By using the concepts proposed by the Economics of Proximity, we will point out the determinants of the trajectory and of the collective learning related to it. By proposing the terms of "heavy proximity" and "light proximity", we will try to produce a better understanding of the links between proximity, network, coordination and local economic development in some big cities of developed countries. We will use the empirical results of several research realized in Marseille area (south of France) since the middle 70\'s.