danksharding

Danksharding is Ethereum's long-term plan for scaling data availability so that rollups can post huge amounts of cheap data to the base layer. Rather than sharding execution (splitting the chain into many separate sub-chains that each run transactions, the old 'eth2 sharding' design), danksharding shards only data: every block carries a large, unified market of data blobs that rollups buy space in, while consensus stays on one chain. It is named after researcher Dankrad Feist, and proto-danksharding (EIP-4844) was the first live step toward it.

The mechanism combines several pieces from this field. Each block reserves space for many blobs of erasure-coded data — full danksharding targets roughly 64 blobs per block, on the order of 16 MB, far more than any single node downloads today. Validators do not each store all of it; instead they use data availability sampling to collectively guarantee the blobs were published, with KZG commitments tying samples to the committed data. Because no node needs the whole dataset, the chain can offer vastly more data bandwidth than its weakest participant could store, which is exactly what rollups need to drive fees down.

Danksharding also leans on proposer-builder separation. Building a block that packs and erasure-codes tens of megabytes of blobs is heavy work, so that role is given to specialized builders, while the validator who proposes the block only needs to verify a succinct commitment — keeping ordinary validation cheap and decentralized. As of early 2026 full danksharding is still a multi-year research-and-engineering effort; what is live is the proto-danksharding foundation, with intermediate steps (such as PeerDAS) incrementally adding the sampling machinery. The end state is a base layer that is a cheap, abundant data-availability engine for a galaxy of rollups.