The recent literature on externalities of schooling in the U.S. is rather mixed: positive external effects of average education are hardly found at all, while often positive externalities from the share of college graduates are identified. This paper proposes a simple model to explain this fact and tests it using U.S. states data. The key idea is that advanced technologies, associated with high total factor productivity and high returns to skills, are complementary to highly educated workers, as opposed to traditional technologies, complementary to less educated. Our calibrated model predicts that workers with twelve years of schooling (high school graduates) are indifferent between traditional and advanced technologies, while more educated workers adopt the advanced technologies and benefit from the larger private and social returns associated to them. Only shifts in education above high school graduation are therefore associated with positive social returns stemming from more efficient technologies. The empirical analysis, using compulsory attendance laws, immigration of highly educated workers and the location of land-grant colleges as instruments confirm that an increase in the share of college graduates, but not an increase in the share of high school graduates, had large positive production externalities in U.S. States.