How are rollups more scalable than base layer?

As far as I know, the cost of processing transactions on a rollup is the same as processing them on the base layer, and you also need to process all incoming rollups to be sure you aren't building off of an invalid previous "block". What am I missing here?