Well then you're only really talking about < 0.1% of projects, since a single server and replication and caching will get you as far as you need, even for most social networks.
And if you are needing to massively shard because you're Facebook or Twitter, then it's not much of a hot take at all. But it's also massively oversimplified advice. Because the tradeoffs between joins and denormalization depend entirely on each specific scenario, and have to be analyzed use case by use case. In many cases, joins still win out -- e.g. retrieving the profile name and photo ID for the author of each post being displayed on your screen.
I'm just worried that people without experience will see your advice and think it's a good rule of thumb for their single-server database, because they think joins are scary and you've provided some kind of confirmation.
