Abstract: Abstract: In recent years, several cities in China have been carrying out the "talent snatch war". But this “war” which is inclined to the level of quantity can not be suitable from the perspective of shifting in driving forces. Firstly, this paper uses the panel-data model to find a significant nonlinear relationship between high-tech and traditional industrial agglomeration in the national sample, Eastern, central and western regions, namely, the hypothesis of Williamson. Secondly, combined with generalized quantile regression (GQR), we find that there exists optimal quantile structure of talents in eastern, central and western regions, and the cities with the best loci are beneficial to shift between new and old kinetic energy. Finally, the optimal locus structure is used to set the locus interval of high, middle and low gear positions to evaluate the talent policies of major cities which participating in the "talent snatch war". In recent years, the agglomeration of R&D talents has gradually increased in that cities, but the structure of the agglomeration has been destroyed. The agglomeration of R&D talents deviated from the optimal structural in related cities belonged to eastern and western regions, which is not benefit to shift in driving force in regions. The findings show that the city "snatching war" is not only stressing on the quantity level of talents, but also to match the structure of regional talent agglomeration. It is essential to absorb or release high-level talents to shift in driving force according to the optimal site.