A Voronoi cell is the region of space consisting of all points closer to one particular centre — a centroid — than to any other. Together, a set of centroids divides the whole space into non-overlapping Voronoi cells, each owned by its nearest centroid, like territories drawn around a set of capitals.
In vector databases, Voronoi cells are how cluster-based IVF indexes partition the vector space. Running k-means clustering produces a set of centroids, and each stored vector is assigned to the cell of its nearest centroid, grouping the data into regions. At query time, the database finds the centroids nearest the query and searches only their cells, skipping the rest and dramatically reducing the work.
The quality of this partitioning directly affects search quality. Well-placed centroids create cells that group genuinely similar vectors together, so searching a few cells captures the true nearest neighbours. Poorly trained centroids scatter similar vectors across many cells, hurting recall unless more cells are searched. Understanding Voronoi cells explains why IVF indexes balance the number of clusters created against the number searched per query.