Sharding

Sharding is a scaling technique that splits a blockchain's workload across multiple parallel segments — shards — so the network can process more than any single node could handle alone. Borrowed from database design, the idea is that instead of every node verifying every transaction, different subsets of the network handle different shards, multiplying total capacity while keeping individual hardware requirements modest.

The hard part is doing this without weakening security: if each shard is validated by only a slice of the network, an attacker needs to overpower only that slice, and transactions that span shards require careful cross-shard communication. These difficulties reshaped Ethereum's roadmap. Its original plan for 64 execution shards — parallel chains all running transactions — was abandoned in favor of rollups doing execution, with the base chain sharding only data. The first step, "proto-danksharding" (EIP-4844), shipped in March 2024, adding cheap temporary "blob" space that cut rollup costs dramatically; full danksharding, which would let nodes verify blob data by sampling small pieces rather than downloading everything, remains ahead. Other networks took different routes: NEAR runs sharded execution in production, and TON is designed to split shards dynamically under load.

A common misconception is that Ethereum "has sharding" in the original sense of parallel transaction-processing chains; what it is building is data sharding in service of rollups. The broader lesson is that sharding is not one technique but a family of designs, each trading complexity for capacity in different places.