+1. Recently, Yugabyte has seemed intriguing to me. (It seems like) You can launch it as a standalone instance and use it just like Postgres early in your company, and then when you need to distribute, you can do that.
I think many love the promise of a distributed database, but in practice there is a real speed and feature set compromise to make.
I found that having a NoSQL distributed database for prod that stream Change Data Captures to a read-only SQL DB to be a great compromise. The SQL data was slightly stale but that wasn't an issue for most situations (like graphing data or certain error metrics)