I think the implicit assumption was that if multiple instances of the same concert series are independent shards, multiple concert series are also automatically independent shards.

You can tune your Kafka etc. to handle the traffic they can, and when it stops scaling, just have multiple instances of the service.

I totally appreciate that I don't understand a lot about actually tuning these systems in production. My mental model is that of a systems academic (that should be doable within 1000 cycles, so 3M/s/core) :).