Vector tiles changed the plumbing of web maps. Instead of serving pre-drawn image squares, servers ship compact tiles of raw geometry and attributes; the client renders them live on the GPU. The consequences cascade: instant restyling without re-processing, crisp rendering at any zoom and screen density, rotation and 3D tilt, feature-level interactivity — the fluid cartography users now take for granted.
How the format works
The de facto standard (Mapbox Vector Tiles) packs layers of clipped, generalised geometries with attributes into small protobuf tiles along the familiar z/x/y pyramid. Generalisation is baked per zoom — a coastline carries only the vertices that zoom can show — keeping tiles tiny while data stays sharp.
Styling lives entirely client-side in a style document mapping data attributes to paint rules, evaluated live by renderers such as MapLibre GL.
What it changes in practice
One tileset now serves unlimited cartographies: light, dark, thematic, branded — a style file each, no server re-render. Labels stay razor-sharp on every display and rotate upright as maps spin. Clients can query any feature under the cursor without server round-trips, and offline mapping becomes practical bundles of tiles plus styles.
Production pipelines (tippecanoe and successors, PostGIS ST_AsMVT, commercial tilers) turn geodatabases into tilesets on schedule; PMTiles even packs whole pyramids into one static file served from object storage.
Raster tiles still matter
Imagery, hillshade and scanned maps remain raster by nature; hybrid maps layer vector cartography over raster imagery. And for extremely complex geometry at low zooms, server-rendered raster can still beat client rendering. The choice is per-layer, not ideological.
The modern default, though, is clear: vector for the map’s structure, raster for its photographs.