This article identifies the most cohesive multi‐polar regions of the network of world cities, which differ from the unipolar centre–periphery model largely created by the high weight of central city connections. We use a community detection algorithm that outlines the high densities of city networks (in relative weights). Various patterns of industries and services, which are divided into two skill levels, are identified. We use a global database of the network of 1.2 million direct and indirect ownership links between the 800,000 subsidiaries of the top 3000 multinational groups in 2013, allowing us to build four comparable networks of 503 metropolitan areas. Comparing the obtained partitions with continental, regional and economic benchmarks, classes of cities partially correspond to free trade zones (FTZs) but exhibit interesting cross‐continental patterns. A few cities, changing their classes according to the activities, are discussed in the light of the multinational firms' strategies.