PostgreSQL 1,289 keeps data on the disks of the server that queries it, so more data means a bigger server. All five cloud warehouses split the two: tables live as compressed columnar files in the provider's object storage, and query engines are stateless clusters that read them, cache hot files on local SSD, and can be started, resized or stopped independently. A metadata or "cloud services" layer holds the catalog, transactions and optimizer.
The consequences shape every bill below. Storage is cheap and charged by volume. Compute is charged by time or by work, and idle compute costs nothing if it suspends, so workloads get separate clusters instead of competing for one. Under the hood each engine is a vectorized column store (Column Stores Under the Hood) with zone maps on its files (Zone Maps and Pruning); clustering or sort keys decide how well those zone maps prune.