The point is exactly that. If you go for Kafka you will need another database to handle the writes. That's how it was born in Linkedin, to forward postgres changes to consumers. It is used as the integration platform for _different_ services.

Event sourcing is an implementation pattern that lives _inside_ a service.

Event sourcing is about saving the application state as a serie of events. Being the application a given service/bounded-context. An event sourcing implementation does not span the boundary of a service. We are talking about a different pattern when looking to Kafka usages.