Network models of opinion formation highlight homophily and interpersonal influences as profound micro-mechanisms for opinion polarization. Yet, in these models, individuals are usually presumed to behave in a single dimension where opinions determine interactions and vice versa. Consequently, these models have not explicitly considered how the population’s correlated structure configures mixing opportunities among different groups and, in turn, are incapable of making sense of the alignment-driven polarization processes that have been widely documented in empirical research (e.g., ideological identities and sociodemographic groups). Drawing on ecological approaches of network formation, this study examines how social consolidation—the correlation between positions across multiple social dimensions—interacts with micro-mechanisms to produce opinion polarization. I first construct an agent-based model to show that depending on the level of consolidation, the same set of micro-mechanisms contributing to either consensus or polarization can produce different polarization outcomes due to spatial clustering and opinion alignment along multiple dimensions that this model introduces. Then I empirically test a primary theoretical mechanism that links consolidation to polarization: spatial clustering. Using new spatial estimates drawn from 180M geo-coded voter profiles in the United States (Brown and Enos 2021), I find novel evidence that counties and census tracts of high sociodemographic consolidation (measured by a race-education correlation) tend to present greater partisan isolation (a higher proportion of co-partisans in their nearest neighbors). These findings suggest social consolidation constitutes a strong foundation for opinion polarization.