A shard is explained in context under sharding, which covers how splitting a network works, its benefits and its security challenges. This page explains the single word, which often appears in technical roadmaps and exchange glossaries.
A shard is one slice of a larger system. In databases, a big table may be split into shards held on different servers, each storing only some rows, so that the system can grow beyond the limits of one machine. In blockchains, a shard is a subset of the network that processes its own group of transactions and keeps its own part of the state, while a coordinating chain or protocol ties the shards together. Validators are assigned to particular shards, so each does less work than if it had to check everything.
The appeal is capacity: several shards processing in parallel can handle more transactions per second than one chain. The difficulty lies in security and communication. Each shard has fewer validators watching it, so designs must stop attackers from concentrating on one shard, and transactions that touch several shards need extra steps.
Networks that use sharding include NEAR, whose shards produce pieces of blocks called chunks, and MultiversX. Ethereum once planned dozens of shard chains but moved to a rollup-focused design. Related ideas include adaptive state sharding.