What is Data Availability?

Data availability is the assurance that the full data behind a proposed block has been published to the network, so any participant can download it and independently verify or reconstruct the chain's state. The concept matters most for rollups, which execute transactions away from their parent chain and must publish the underlying data somewhere verifiable.

The problem DA solves is withholding. A block producer could publish a valid-looking header while hiding the transactions inside it, leaving the network unable to prove anything is wrong because the evidence is missing. Guaranteeing that the data was published, even briefly, closes that attack and lets fraud proofs and state reconstruction work.

DA is deliberately temporary. Networks only need the data held long enough for verification and dispute windows to pass, after which it can be pruned. Ethereum drops blob data after roughly 18 days, and Celestia cut its minimum pruning window to just over 7 days, keeping node storage requirements flat as throughput grows.

Publishing data is the single largest operating cost for most Layer 2 networks, so the price and capacity of DA directly set rollup fees. That has turned DA into a competitive market where Ethereum, Celestia, EigenDA, and Avail all sell blobspace to the same customers.

How Does Data Availability Work?

Modern DA systems combine erasure coding, cryptographic commitments, and random sampling to let small, cheap nodes verify that huge amounts of data exist without downloading any of it.

1. Erasure Coding

Erasure coding expands a block's data into a larger set of redundant fragments, typically using Reed-Solomon encoding, so the original can be rebuilt from only a portion of the pieces. If half the extended data survives, the whole dataset survives.

That property transforms the withholding problem. Without coding, hiding a single transaction means suppressing one tiny fragment, which random checks would almost never catch. With coding, hiding anything means suppressing so much of the extended data that a handful of samples detects the gap with near certainty.

Commitments bind the fragments to the block header so nodes can check each sample against a fingerprint. Celestia arranges its coded data in a two-dimensional square with namespaced Merkle trees, while Ethereum and Avail commit to blob data using KZG polynomial commitments, which prove a fragment belongs to the set without revealing the rest.

2. Data Availability Sampling (DAS)

Sampling is the verification technique built on top of the coding. Instead of downloading a block, a light node requests a few random fragments and checks them against the commitments. Each successful sample sharply raises the probability that all the data is out there, and a few dozen samples push confidence above 99.99%.

The security scales with participation. Every additional sampling node queries different random coordinates, so thousands of light clients collectively cover the whole dataset even though no single machine holds it. Withheld data fails samples quickly, and honest nodes reject the block before it finalizes.

Large blocks become safe because of sampling. A chain can raise its data capacity by an order of magnitude without forcing every node to buy more bandwidth, because verification cost stays roughly constant. That is the mechanism behind both Ethereum's PeerDAS and Celestia's light node network.

3. Full Nodes, Light Nodes, and Reconstruction

DA networks split verification duties across node types. Full nodes download and store complete blocks or assigned portions of them, serve samples to the rest of the network, and reconstruct missing data from erasure-coded fragments when a producer withholds part of a block.

Light nodes do the sampling. They hold only headers and commitments, run on laptops or phones, and still gain strong guarantees that the data exists. Under Ethereum's PeerDAS design, even validators no longer store every blob, since each node custodies an assigned slice of the coded data and samples the rest from peers.

Reconstruction is the safety net binding the two together. If enough honest nodes hold enough fragments between them, the full dataset can always be rebuilt and republished, so sampling networks need a minimum honest population, and larger populations harden the network rather than burden it.

4. Data Availability Committees (DACs)

Not every chain pays for full DA guarantees. A data availability committee is a fixed group of known operators who hold a rollup's data off the parent chain and sign attestations that it exists. Validiums and optimiums use this model to cut costs to near zero, accepting that users must trust the committee rather than open verification.

If a committee colludes or fails, users may be unable to prove their balances or exit the chain, a risk that sampling-based systems remove by letting anyone check the data directly. L2BEAT tracks which networks rely on committees versus verifiable DA layers, and a large share of app-specific chains still choose the committee model for its price.

Hybrid designs try to narrow the gap. Some committees post cryptographic commitments to the parent chain, add staking and slashing to punish misbehavior, or fall back to full onchain publication if attestations stop arriving.

Data Availability on Ethereum: Blobs and PeerDAS

Ethereum turned DA into a first-class product with the Dencun upgrade in March 2024, which introduced blobs through EIP-4844. Blobs are 128 KB data packages that rollups attach to blocks, priced by a separate fee market and pruned after roughly 18 days.

Capacity has climbed in steps since launch. Dencun began with a target of 3 blobs per block, the Pectra upgrade doubled that to 6, and the Fusaka upgrade then activated PeerDAS, the sampling protocol that lets nodes verify blobs without downloading them all. Fusaka also added Blob Parameter Only forks, small pre-scheduled upgrades that raise blob limits without a full hard fork.

Two BPO forks followed within weeks. The second BPO fork lifted the target to 14 blobs per block with a maximum of 21, roughly 2.7 MB of data per block, and core developers have held further increases in reserve until demand fills the new space. The roadmap runs toward full danksharding, where two-dimensional sampling supports far higher blob counts, and the upcoming Glamsterdam upgrade restructures block propagation to unlock the next step.

Pricing matters as much as capacity. Blob fees float on their own auction, so rollups pay very little when blobspace is slack and spike together when it saturates, a cycle that played out when demand caught up with Pectra's limits before Fusaka reset the headroom.

Ethereum Blobs vs Celestia vs EigenDA vs Avail

Rollups now choose between Ethereum's native blobspace and dedicated DA networks, and the decision shapes their cost base, throughput ceiling, and trust assumptions. Ethereum offers the deepest security and keeps data verification inside the same validator set that settles the rollup, while the alternatives sell far more raw bandwidth for less.

Celestia is the largest purpose-built DA network, a proof-of-stake chain that does nothing except order and publish data blobs verified by light node sampling. Its Matcha upgrade raised maximum blocks from 8 MB to 128 MB and halved inflation, part of a roadmap targeting 1 GB per second of throughput. Sovereign rollups and high-volume chains such as Eclipse publish there for the price and headroom.

EigenDA takes a different path, running as a service secured by ETH restaked through EigenLayer instead of operating its own chain. Its second version reached 100 MB per second on mainnet, the highest live throughput of any DA system, which is why performance-focused chains like MegaETH build on it. The trade-off is a dispersal model closer to a decentralized committee than an openly sampled chain.

Avail sits between the two, a sovereign DA chain using the same KZG commitments and sampling design as Ethereum's danksharding roadmap. It runs 4 MB blocks today with a stated path toward multi-gigabyte capacity, and pairs the DA layer with Nexus, a cross-chain coordination system now live on mainnet.

Here is how the main options compare:

Data Availability vs Data Storage

DA and storage answer different questions. Availability proves that data was published and checkable at the moment a block was produced, which is what consensus and fraud proofs require. Storage keeps data retrievable long afterward, which is what explorers, indexers, and users replaying history require.

Blockchains only enforce the first. Once Ethereum's roughly 18-day blob window or Celestia's pruning window passes, the protocol makes no promise the data still exists, and the incentive to keep it shifts to parties who need history, including rollup teams, archive nodes, explorers, and indexers.

Permanent storage networks fill the second role. Systems like Arweave and Filecoin pay nodes to hold data indefinitely, and some rollups archive their pruned blob history there. Conflating the two leads to a common misconception that pruning makes rollups unsafe, when the security-relevant window has already closed by the time data expires.

Why Data Availability Matters for Rollups

Every rollup's security reduces to a claim that outsiders can check its state, and that claim fails without published data. Optimistic rollups need the data so challengers can spot an invalid state root and submit a fraud proof within the dispute window. If the sequencer could hide transactions, no one could construct the proof, and theft would finalize unchallenged.

ZK rollups need it for a different reason. A validity proof already guarantees the state transition was computed correctly, but users still need the underlying data to know their own balances and to exit the rollup if its operators vanish. Proofs without data leave a chain correct but unusable, which is why both rollup families treat DA as non-negotiable while validiums accept weaker committee guarantees to cut costs.

DA capacity also caps rollup performance. A chain can only process transactions as fast as it can publish the data behind them, so blob limits and DA bandwidth set throughput ceilings and fee levels across the Layer 2 ecosystem. That link is why every major scaling upgrade of the past two years, from Dencun through Fusaka to Celestia's Matcha, has been a DA upgrade first.

Challenges of Ensuring Data Availability

Data withholding remains the core attack. Sampling defeats it statistically, but the guarantees depend on assumptions that must hold in practice, including enough honest sampling nodes, unpredictable sample requests, and working reconstruction when fragments go missing. Celestia's block reconstruction protocol for minimal light node populations is still under active development, and PeerDAS shipped on Ethereum mainnet barely half a year ago.

Economics create a second set of pressures. Dedicated DA layers price blobspace so cheaply that fee revenue stays thin, raising questions about how their validator sets are funded long term, while Ethereum faces the opposite tension because pushing data to external layers moves fee income away from its own validators. Committee-based systems inherit familiar centralization risks, since a small operator set can censor, collude, or simply go offline.

Scale adds engineering risk of its own. Propagating 128 MB blocks or 100 MB per second of dispersed data pushes networking limits, and each capacity jump has to preserve the property that a modest home connection can still verify the chain. The sector's roadmap targets, gigabyte blocks and beyond, all depend on sampling continuing to hold up as datasets grow by orders of magnitude.

Final Thoughts

Data availability started as an obscure corner of rollup design and became the axis the entire scaling debate turns on. Whoever publishes the data defines the security of the chain built on top, and whoever prices it defines the fees users pay.

The field has consolidated around sampling. Ethereum's PeerDAS, Celestia's light nodes, and Avail's KZG design all converge on the same insight, that random checks over erasure-coded data let small machines verify enormous blocks, while EigenDA shows how much raw bandwidth a restaked operator set can deliver today.

We expect demand to decide the next phase. Blob capacity outruns usage on Ethereum, Celestia holds 16x headroom over its old limits, and the open question is which applications, from high-frequency trading chains to onchain games, grow into the space that has been built for them.