When do you think sharding a database for scaling is just too early to worry about?

Hey everyone, I've been thinking about how we scale databases. Sharding is super strong for horizontal scaling, especially with huge datasets or tons of write activity. But it makes things way more complicated. Distributed transactions are tougher, rebalancing can be a total nightmare, and your app has to know about the sharding key. Sometimes I think we jump to sharding too fast, instead of really looking at vertical scaling, read replicas, or just better indexing and optimizing our queries. So, when do you all think sharding is actually needed? And when does it just feel like we're optimizing too soon, causing more trouble than it's worth?