A modular blockchain splits a blockchain's core jobs (execution, settlement, consensus and data availability) across specialized layers instead of doing them all on one chain. Data availability means confidence that a block's full data was really published, so anyone can check it. It does not mean permanent storage. Ethereum blobs, Celestia and EigenDA all offer this guarantee, at different costs and with different people to trust.
What are the four jobs of a blockchain?
Every blockchain has to do four things, as Celestia's guide lays them out.
| Job | What it does | Example of a specialist |
|---|---|---|
| Execution | Processes transactions and updates balances | A rollup such as Base |
| Settlement | Resolves disputes and hosts the bridge | Ethereum, for its rollups |
| Consensus | Agrees on the order of transactions | Ethereum, Celestia |
| Data availability | Ensures the data behind a block was published | Ethereum blobs, Celestia, EigenDA |
Monolithic vs modular: what actually changes
A monolithic chain does all four jobs itself. Solana is the usual example. Its big advantage is simplicity: apps share one state, so they can interact directly, with no bridge in between.
A modular design hands jobs to different layers. A rollup executes transactions, then relies on another chain for the rest. The goal is to scale each job separately, and to let a new chain launch without recruiting its own validators.
The labels are less tidy in practice. Ethereum still runs its own execution while also selling data space to rollups, so it is both. Modular designs also bring real costs: more moving parts, more bridges between layers, and liquidity split across chains. Treat "modular" as a set of trade-offs, not as an upgrade.
What does data availability mean?
Data availability is the confidence that the data needed to verify a block is really available. Here is why it matters.
Imagine a rollup operator posts a new state root, a compact summary of everyone's balances, and keeps the underlying transactions secret. The summary might say the operator now owns everything. Nobody can build a fraud proof against it, because nobody has the data. Honest users also can't work out their own balances to exit. A zero-knowledge proof would stop the false state, since validity proofs prove correct execution, but users could still be frozen out if they can't see the data.
This is called a data withholding attack. The block looks fine from its header alone.
Availability is not storage
This is the most common confusion in this topic. Celestia's FAQ draws the line clearly. Availability proves the data was published. Storage means keeping history retrievable, and it only needs one honest keeper.
Ethereum shows the difference. Blobs are only guaranteed for about 18 days, because that is long enough for anyone who cares to download them. Older data falls to archives and indexers.
How erasure coding and sampling work
Downloading every block would defeat the point, since it limits capacity to what small nodes can handle. Data availability sampling (DAS) lets a node check that data is there by downloading only small random pieces. It rests on a trick called erasure coding.
- Extend the data. The block's data is expanded with redundant parts, usually doubling its size. The original can then be rebuilt from any half of the extended version.
- Force the attacker's hand. Hiding one transaction is no longer enough, since others could rebuild it. To make the data unrecoverable, a block producer must hide more than half of the extended version.
- Sample at random. Each random piece a node requests now has at least a 50% chance of hitting a gap if the block is being withheld.
- Grow confidence. Every successful sample shrinks the odds that data is missing.
The numbers fall quickly. In this simplified model, if data is being withheld, the chance of 7 samples all succeeding is 0.5 to the 7th power, about 0.8%. With 20 samples it is roughly one in a million. Real systems use different layouts, so their exact numbers vary, but the logic holds. A node learns a lot from very little data.
One more problem remains: the producer might encode the data incorrectly. Ethereum and EigenDA use KZG commitments, a cryptographic fingerprint that lets nodes check each piece against the block. Celestia uses a two-dimensional grid of Merkle roots, and light nodes rely on at least one honest bridge node, a type of full node, to send a fraud proof when a block is badly encoded.
How do Ethereum, Celestia and EigenDA compare?
| Ethereum blobs | Celestia | EigenDA | |
|---|---|---|---|
| Who provides it | Ethereum validators | Celestia's own proof-of-stake validators | Operators with restaked assets |
| How data is checked | PeerDAS sampling and custody by nodes | Light nodes sample the data | Operators hold pieces and sign a certificate |
| How Ethereum contracts see it | Natively | Through a bridge carrying validator-signed data roots | Through a certificate of operator signatures |
| Stated capacity | Target 14, max 21 blobs per block | Blocks of up to 128 MB | 100 MB/s, as reported by Eigen Labs |
| 30-day average cost | $0.102 per MB | $0.0323 per MB | $0.0362 per MB |
| L2s using it (of 25 tracked) | 15 | 3 | 4 |
Costs and L2 counts are from growthepie and are dated November 1, 2026. Prices move a lot, so single days can look very different.
Ethereum blobs with PeerDAS
Blobs are data attachments of about 128 KB with their own fee market. PeerDAS went live with the Fusaka upgrade on December 3, 2025. It splits extended blob data into 128 columns, and a regular node subscribes to at least 8 of them, which works out to 1/8 of the original data's size. Nodes can rebuild the full data from half the columns.
Later small forks raised capacity to a target of 14 and a maximum of 21 blobs per block in January 2026. At 128 KiB per blob and 12-second blocks, that maximum is about 0.22 MiB per second, modest next to the claims below. In return, you get Ethereum's own security. L2BEAT shows about 815,000 validators and $66.7 billion in economic security as of November 2026, and full nodes reject blocks with unavailable data.
Celestia
Celestia is a blockchain dedicated to consensus and data availability. It doesn't execute rollup transactions. Its light nodes sample random pieces of each block, so availability checking is open to anyone running one. The Matcha upgrade, activated on mainnet on November 24, 2025, raised the protocol's maximum block size from 8 MB to 128 MB.
The catch appears when a rollup settles on Ethereum. An Ethereum contract can't run sampling, so it depends on a bridge. Celestia's Blobstream was designed to require signatures from validators holding at least two-thirds of Celestia's stake, and newer versions use ZK proofs to verify those signatures. If that supermajority lied, the contract would believe it.
Separately, Celestia's Fibre protocol, announced in January 2026, runs alongside its main blobs. The announcement says sampling is supported for Celestia's main blobs, and it describes Fibre as validators storing pieces and signing receipts. Its headline throughput comes from benchmarks, so check its status and its trust model separately.
EigenDA
EigenDA uses restaking: operators are backed by delegated stake from EigenLayer. Eigen Labs describes the V2 flow this way:
- A disperser takes a rollup's blob and erasure-codes it into 8,192 chunks, any 1,024 of which can rebuild it.
- Validators collect their chunks and check them against a KZG commitment.
- Validators sign, and the disperser combines the signatures into a certificate.
- The blob counts as confirmed once validators holding at least 55% of stake have signed, by default.
This works more like a staked committee than an open sampling network. The post describes no light-node sampling, so as described, outsiders can't check availability the way they can on Celestia. You rely on signers, along with the dual-quorum design of ETH and EIGEN, in which a blob stays available as long as one quorum is uncorrupted. Eigen Labs listed slashing for EigenDA as a roadmap item in September 2025, so check whether penalties are active before counting that stake as a deterrent.
The real cost and trust trade-offs
The savings look large in percentage terms and small in dollars. At the averages above, publishing 1 GB in a month would cost about $102 on Ethereum, $36 on EigenDA and $32 on Celestia. Across all three layers, the L2s growthepie tracks paid only about $18,100 in the 30 days to November 1, 2026.
At today's demand, then, data cost is rarely what separates these options. Prices could change if demand reached capacity, because blob fees rise when usage stays above target, so headroom matters for a busy chain. For most teams, the bigger question is who they are asking users to trust.
That choice also changes what a chain is. L2BEAT labels chains that keep data off Ethereum as Optimiums or Validiums, not rollups. Orderly, which it describes as "an OP stack Optimium on Ethereum using Celestia for data availability," is one example.
If an outside data layer fails or lies, the rollup's users bear the loss. They may be unable to prove fraud or withdraw, even while Ethereum runs normally. With blobs, unavailable data would mean Ethereum itself had broken its own rules.
Questions to ask about any chain
Before trusting a chain's "modular" branding, check:
- Where is the data actually posted? Look for this on L2BEAT's project page.
- Can Ethereum verify it directly, or does it rely on a bridge and signatures?
- Whose stake backs it, how much is there, and can it really be slashed?
- Can independent people check availability, through sampling or by downloading it all?
- What happens if the data layer stalls? Is there a fallback to Ethereum, or does the chain stop?
- How long is data kept, and who keeps it after that?
You can follow related topics in our scaling and storage coverage.
Key takeaways
- Modular chains split execution, settlement, consensus and data availability. Monolithic chains do all four together.
- Data availability proves data was published. It is a different promise from long-term storage.
- Erasure coding means an attacker must hide over half the extended data, so a few random samples catch withholding with high confidence.
- Ethereum blobs offer the strongest guarantee and the least capacity. Celestia and EigenDA offer more capacity with separate trust assumptions.
- As of November 2026, DA fees are small in absolute terms, so trust is usually the deciding factor.
- A chain that keeps data off Ethereum is, in L2BEAT's terms, not a rollup, and its users carry that extra risk.
Sources
Get the weekly commit
New blockchain deep dives every week.
