A source of confusion is that in both cases, the data can appear in one chunk that is replicated among the components/nodes, or split ~equally among the components/nodes (think DHT). E.g. in what you call "decentralized", blockchain is fully replicated on each node, which can be a problem as the database gets larger over time. It's possible to split the data like CockroachDB does, but CRBD is limited in how many nodes it can handle in a cluster (<100), therefore it's more of a "distributed" system. I don't know if there is a truly decentralized DB that splits the data among millions of nodes..