Any strongly consistent database is going to be limited by a single machine's throughput. That's just what you trade for strong consistency. You can shard it yourself but then the DBMS isn't giving you consistency so you'd better be very careful. You can use a tool like PgDog to aid with sharding but it's not doing magic, you still have to be aware how it works and the limitations of sharding.

However 95% of projects are going to be fine with a normal single-machine database and another 4% are going to be well served by upgrading the hell out of that machine. Only the absolute busiest projects actually need a distributed database and you can cross that bridge when you actually get to it.

They say Amazon processes 20k orders per second. That seems not unachievable for postgres with fast SSDs and careful query optimization, though they don't choose do it that way. You're not Amazon, you have at most 20 orders per second and that's nothing.