We analyze the effect of an increase in knowledge-intensive activities on spatial inequality in US cities. We leverage a predetermined network of patent citations to instrument for local innovation trends. Between 1990 and 2010, a one-standard-deviation increase in patent growth increases income segregation by 0.65 Gini points, corresponding to 0.31 standard deviations of the over-time change in income segregation. This effect mainly arises from the sorting of residents by income, occupation, and education. Local shocks to innovation induce a clustering of knowledge-intensive jobs and residents, amplified by the response of rents and amenities. (JEL D31, O31, O33, O34, R23, R32)