A shard key might be the customer ID, so all of one customer's records sit together on one server. Sharding lets write volume and data size grow past what a single machine can handle, and very large services such as Meta and Notion have published how they use it.
It is also expensive to live with. Queries across shards get harder, transactions across shards get harder still, and moving data when a shard fills up is a project in itself. Most companies of 20 to 200 people never need it. Measure first, then try better indexes, caching, read replicas and table partitioning, because each of those is far cheaper to undo.