Sharding is a scaling method that splits a blockchain's workload across groups of nodes, so no single node has to store or process everything. Each group, called a shard, handles part of the data or part of the transactions. The term comes from databases, where sharding means spreading the rows of one large table across several machines. The newest example is Ethereum. It dropped its original plan for separate "shard chains" and now shards rollup data instead. The step that brought this into production, PeerDAS, went live with the Fusaka upgrade on 3 December 2025.
How sharding works
In a non-sharded blockchain, every full node checks every transaction and keeps a full copy of the state. That is secure, but the whole network can only go as fast as one node. Sharding breaks that link in one of three ways.
- Data sharding. Nodes store and serve only part of the data, and the network relies on sampling to confirm that all of it is available. Ethereum's current approach is this kind.
- Transaction or network sharding. Nodes are divided into committees, and each committee processes a different set of transactions in parallel.
- State or execution sharding. Each shard keeps its own part of the account state and runs its own transactions. This is the hardest kind, because a transaction that touches two shards needs messages between them.
The weak point in every design is coordination between shards. Polkadot's documentation notes that the limit on composability is "common to all sharded blockchain designs, including Polkadot, Ethereum, and others." Moving value or calling a contract across shards adds latency and complexity compared with doing it inside one chain.
Sharding vs partitioning
Both split a large dataset into smaller pieces. The difference is where the pieces live.
| Partitioning | Sharding | |
|---|---|---|
| Meaning | General term: dividing logical entities into separate physical ones | A type of horizontal partitioning |
| Location of data | Can stay on one database instance | Spread across multiple database instances |
| How rows are assigned | Depends on the scheme | Schema is replicated; rows are divided by a shard key |
SingleStore's engineering blog puts it this way: "Sharding is actually a type of database partitioning, more specifically, Horizontal Partitioning," and "sharding implies the data is spread across multiple database instances." A shard key, which can contain any number of columns, decides which shard holds a given row. Blockchains borrowed the idea and applied it to nodes rather than database servers.
Ethereum: from shard chains to data sharding
What changed
Ethereum's early roadmap planned many shard chains. That plan is gone. ethereum.org now says "'shard chains' are no longer needed and have been dropped from the roadmap." The current approach is called danksharding. Rollups (layer-2 networks) do the execution, and Ethereum's job is to provide cheap, verifiable space for rollup data.
Timeline
- 13 March 2024: Dencun upgrade. EIP-4844, also called proto-danksharding, added a new transaction type that carries "blobs" of data with their own fee market. The EVM cannot read blobs, and they are pruned after about 18 days (4,096 epochs). At launch the target was 3 blobs per block, with a maximum of 6.
- Pectra upgrade. The blob target rose to 6.
- 3 December 2025: Fusaka upgrade. PeerDAS (EIP-7594) introduced data availability sampling. Each node no longer downloads every blob: a full node keeps about one-eighth of the data, and any 50% of the data is enough to rebuild the whole. ethereum.org describes this as a theoretical scale-up of up to 8x.
- 9 December 2025 and 7 January 2026: blob-parameter-only (BPO) forks. BPO1 raised the target and maximum from 6 and 9 blobs per block to 10 and 15. BPO2 raised them to 14 and 21.
- Next. ethereum.org lists the Glamsterdam upgrade as in development, targeted for Q4 2026. Its listed features are enshrined proposer-builder separation and block-level access lists, not more sharding. Full danksharding, which ethereum.org says aims at 64 blobs per block, is described as "several years away."
What it means for users
Execution is not sharded on Ethereum. Fees on rollups go down when blob space is plentiful. Fees on the main chain depend on mainnet demand and are not directly affected.
How Ethereum moved to data sharding
Step through each upgrade to see how Ethereum went from dropping shard chains to sharding rollup data, and what comes next.
- 13 March 2024Dencun upgrade: proto-dankshardingdone
EIP-4844 added a transaction type that carries blobs of data with their own fee market. The EVM cannot read blobs, and they are pruned after about 18 days (4,096 epochs). The target was 3 blobs per block, with a maximum of 6.
- Before FusakaPectra upgradedone
The blob target rose to 6 per block.
- 3 December 2025Fusaka upgrade: PeerDASdone
PeerDAS (EIP-7594) introduced data availability sampling. A full node keeps about one-eighth of the data instead of every blob, and any 50% of the data is enough to rebuild the whole. ethereum.org describes this as a theoretical scale-up of up to 8x.
- 9 December 2025BPO1 forkdone
The blob target and maximum rose from 6 and 9 per block to 10 and 15.
- 7 January 2026BPO2 forkdone
The blob target and maximum rose to 14 and 21 per block.
- Q4 2026 (target)Glamsterdam upgradeexpected
ethereum.org lists it as in development. Its listed features are enshrined proposer-builder separation and block-level access lists, not more sharding.
- Several years awayFull dankshardinguncertain
ethereum.org says full danksharding aims at 64 blobs per block and describes it as several years away.
This covers only Ethereum's data-sharding steps: it does not show rollup or mainnet fees, execution sharding, or the NEAR, Polkadot and Zilliqa designs.
NEAR: sharding with stateless validation
NEAR's design, called Nightshade, shards both state and execution. On 22 August 2024, Nightshade 2.0 went live on mainnet and introduced stateless validation. Validators no longer have to keep a shard's full state locally. Instead, they "retrieve all the information they need to validate state changes, or 'state witness,' from the network." NEAR said this lowers hardware requirements, because validators "no longer need to track all shards."
At that launch NEAR had six shards and said it aimed to reach ten by the end of 2024.
Polkadot: parachains on a shared relay chain
Polkadot splits work across application-specific chains called parachains. The relay chain at the centre "links all those states into one state." It also provides shared security, so each parachain does not have to recruit its own validator set. Parachains talk to each other using the XCM message format, and the documentation notes that these messages add latency. Chains get access to the relay chain by renting coretime, described as computation time on a relay chain core, paid in DOT.
Because each parachain has its own rules and its own purpose, this model is often called heterogeneous sharding.
Zilliqa: early network sharding, then a redesign
Zilliqa was one of the first public chains built around network sharding: its whitepaper set out a scaling approach "through sharding." Its history shows how hard sharding is to run:
- 3 January 2024. After a network disruption on 18 December 2023, Zilliqa deployed version 9.3.0. Its team said the upgrade cut the risk of further consensus disruptions "through desharding the network."
- Zilliqa 2.0. The new network moves from proof-of-work to proof-of-stake and uses Pipelined Fast-HotStuff consensus, which "replaces the pBFT algorithm used in the previous version." It keeps shard committees and adds "x-shards," customisable chains that connect to the main network. Zilliqa lists the upgrade as live in 2025.
Trade-offs to weigh
- Throughput vs simplicity. Sharding adds capacity, but cross-shard messages make applications harder to build and slower to compose.
- Security per shard. A shard secured by a small group of validators could be easier to attack than the whole network. Shared security (Polkadot) and sampling with random assignment (Ethereum's PeerDAS) are two ways designs try to avoid this.
- Operational risk. Zilliqa deliberately reduced sharding in January 2024 to stabilise consensus. Ethereum dropped shard chains in favour of rollups.
