Data availability is the guarantee that the data behind a block, or behind a batch of transactions processed off the main chain, has actually been published so that anyone can download it, check it and rebuild the resulting state. A block header only commits to its contents through a hash; if the producer withholds the underlying transactions, other participants cannot tell whether the block is valid, cannot compute balances and, for a rollup, cannot force a withdrawal or take over block production. Data availability is therefore a security property in its own right, separate from the question of whether the transactions were executed correctly.
The idea matters most for layer 2 networks, which run transactions away from Ethereum but publish enough data to it that users can reconstruct their state without trusting the operator. How and where that data is published is now one of the main cost and design choices in blockchain scaling, and it ranges from Ethereum itself, through dedicated data availability networks, to small committees that promise to keep a copy.
The data availability problem
A full node solves the problem by downloading every block in full: if any data is missing it rejects the block. This works, but it ties the network's capacity to what an ordinary node can download and store. A light node, which follows only block headers, has no such protection, because a header can be correct while the data it commits to has been hidden.
The difficulty is that withholding data is not an attributable fault. If one party says the data is missing and the producer later publishes it, outsiders cannot tell who was honest. Scaling designs therefore aim to let many nodes each check a small piece of the data and still gain high confidence that all of it was published. The tools for this are erasure coding, which expands the data so that any half of it is enough to recover the whole, and data availability sampling (DAS), in which nodes request random small pieces and accept the block only if every request is answered.
How rollups publish data on Ethereum
A rollup's sequencer orders transactions, executes them and posts compressed transaction data to Ethereum together with a commitment to the new state. At first this data was written into ordinary transaction calldata, which is stored by every Ethereum node forever and priced like any other execution data, so publishing data was usually the largest part of a rollup's costs.
The Dencun upgrade of 13 March 2024 introduced proto-danksharding (EIP-4844), which added a separate kind of data called a blob. Each blob holds about 128 kilobytes, has its own fee market separate from gas, and is kept by consensus nodes for roughly 18 days before being pruned. That window is long enough for anyone who needs the data to download it, while Ethereum commits to each blob permanently through a short cryptographic commitment. Blob fees fell well below calldata costs, and the number of blobs allowed per block was raised in the Pectra upgrade of May 2025.
Validity Proof
A validity proof is a short cryptographic proof that a batch of off-chain transactions was executed correctly, taking the chain from one agreed state to the next. It is usually produced as a zk-SNARK or a zk-STARK, and a smart contract on the base chain checks it in a fraction of the time it would take to re-run the transactions. Once the proof is accepted, the new state is final as far as the contract is concerned.
Validity proofs are the basis of zero-knowledge rollups. They contrast with the approach of optimistic rollups, which assume batches are correct and allow a challenge period, often about a week, during which anyone can submit a fraud proof showing an error. A validity proof removes that waiting period but costs significant computation to generate. It proves only that execution was correct; it does not prove that the data was published, which is why data availability remains a separate requirement.
Validium
A validium is a layer 2 design that uses validity proofs in the same way as a zero-knowledge rollup but keeps the transaction data off Ethereum. Instead, the data is held by a data availability committee, a small group of named parties who sign statements that they have stored it, or by an external data availability network. Ethereum verifies the proof, so the operator cannot steal funds through an invalid state change.
The trade-off is cost against safety. Because no transaction data is posted to Ethereum, a validium is much cheaper per transaction than a rollup. But if the committee withholds the data, users may be unable to prove their balances and withdraw, so their funds can be frozen even though they cannot be taken. StarkEx, the scaling engine developed by StarkWare, offers a validium mode and a hybrid called volition, in which each user or transaction chooses whether data goes on-chain or to the committee.
Danksharding
Danksharding is the planned long-term design for expanding Ethereum's data capacity, named after the researcher Dankrad Feist. It is a form of data sharding rather than execution sharding: Ethereum does not split into many chains that each run transactions, as earlier sharding plans proposed, but instead greatly increases the number of blobs each block can carry. Validators would no longer download every blob; they would use data availability sampling to check that the data was published, with a single block builder assembling the block and its blobs.
Proto-danksharding was the first step. It introduced the blob format, the blob fee market and the commitments that full danksharding will use, without sampling and with a small blob limit. The next step is PeerDAS (EIP-7594), in which nodes download and serve only parts of the erasure-coded blob data and sample the rest from peers; it was included in the Fusaka upgrade of December 2025. Full danksharding would extend sampling across two dimensions to support a much higher blob count.
Dedicated data availability layers
Several networks specialise in data availability alone, ordering and storing data without executing it. Celestia, whose mainnet launched in October 2023, uses data availability sampling so that light nodes can check blocks, and its native token pays for blob space. EigenDA is built on restaked ether through EigenLayer, and Avail is another sampling-based network. Rollups that post data to one of these layers while settling on Ethereum are sometimes called optimiums or validiums, depending on their proof system, because their data guarantees rest on the external network rather than on Ethereum.
External links
Ref 1070001389