The message of the talk was very much that you can scale to massive throughput without having to shard and having only a single master.

Of course they considered it, but the tradeoffs didn't match what they wanted to do - plus they found you could scale to this level without sharding.

The talk seems to be mostly about all the limitations and workarounds they've had to deal with, because they choose not to shard. Apparently, they have a policy of adding no new functionality to the database, which presumably means additional separate database services being setup for each new feature. That sounds a lot like accumulating tech debt very rapidly, just because sharding is not on the table, for whatever reason.