Data virtualization works by creating an abstraction layer over various data sources, which can include databases, cloud services, big data repositories, and enterprise applications. This layer allows users to query data across these sources as if it were stored in a single location. The technology leverages metadata to map and transform data from different formats into a unified schema.