Modular Blockchains Explained: Celestia, EigenDA & Data Availability

Introduction

Celestia, EigenDA and Avail together hold about 2% of layer 2 (L2) value secured, while Ethereum's own blobs hold 86.5% (L2BEAT, 5 November 2026). The modular blockchain thesis promised the opposite: split execution, settlement, consensus and data availability, and let specialist layers win each job. Data availability (DA) is the job where that bet is being tested, and the tokens of the three challengers, TIA, EIGEN and AVAIL, trade about 97% to 99% below their highs as of 5 November 2026. This article defines the four jobs, explains what a DA layer guarantees, compares the four public layers on dated throughput and cost, and shows which rollups use which layer today, with the line L2BEAT draws between a rollup, a validium and an optimium.

What Is a Modular Blockchain, and Which Four Jobs Does It Split?

A modular blockchain splits the four jobs of a blockchain across separate layers instead of running them all on one chain. Ethereum with rollups already works this way (ethereum.org, 2026), so the open question is not whether to split the jobs but where the data availability job runs.

The four jobs: execution, settlement, consensus and data availability

A monolithic blockchain such as early Ethereum or Bitcoin performs all four jobs itself. Execution runs transactions and changes state. Settlement verifies proofs and resolves disputes between execution layers. Consensus agrees the order of transactions. Data availability (DA) confirms that transaction data was published and can be downloaded (Celestia Labs, 2026). Many layers combine two jobs, since a DA chain must also order its data. Ethereum today runs consensus, settlement and DA on its base layer, while rollups, chains that run transactions off Ethereum and post their data back to it, handle execution. A rollup such as Base executes transactions, then sends proofs and data to Ethereum.

What splitting them buys and what it costs

Specialisation is the gain. A layer built only to publish data can raise capacity without making every node execute every transaction ↗, and an execution layer can change its virtual machine without touching security. Ethereum made this its plan in October 2020 with a rollup-centric roadmap (Vitalik Buterin, 2020). The cost is trust spread across more parties. A rollup that keeps its data on Ethereum inherits Ethereum's guarantee that the data was published; a chain that posts data elsewhere inherits that layer's guarantee instead. Splitting also adds bridges, the messaging links between layers, and each is another place to fail. The real choice is therefore which layer holds the data.

Data current as of November 2026.

Of the four jobs, data availability is the one where Ethereum now faces direct competition, so it needs a precise definition.

What Does a Data Availability Layer Actually Guarantee?

Data availability is a promise that data was published and can be downloaded, not that it stays stored forever. A rollup needs that promise because users can rebuild its state, or prove a withdrawal, only if the data exists somewhere anyone can fetch it.

Why a rollup must publish its data

A rollup's operator, the sequencer, orders transactions and posts a compressed batch. If the operator posts a new state but withholds the data behind it, nobody else can check that state or prove they own funds inside it. L2BEAT, the independent tracker of layer 2 (L2) projects that renamed itself ETHBEAT on 4 November 2026, ties the word rollup to this guarantee ↗: data stored anywhere other than a permissionless layer 1 makes a project a validium or an optimium, not a rollup (L2BEAT glossary, 2026). Availability is also time-limited. Ethereum nodes keep a blob, its temporary container for rollup data, for about 18 days, long enough for anyone to check and copy it (Etherscan, 2024).

How sampling lets light nodes check availability

Data availability sampling (DAS) lets a node check a block without downloading it. The producer extends the data with redundant pieces, called erasure coding, so the whole can be rebuilt from part of it. A light node requests a few random pieces with proofs, and if every answer is valid, the odds that the whole block is available are high (Celestia docs, 2026). Ethereum's PeerDAS applies the same idea to blobs: since the Fusaka upgrade on 3 December 2025 ↗, nodes sample blob data instead of downloading all of it (Ethereum Foundation, 2025). Avail pairs sampling with KZG commitments, short cryptographic fingerprints of the data (Avail, 2025).

Sampling is the shared mechanism; the layers differ in who runs it, what secures it and what posting costs.

How Do Celestia, EigenDA, Avail and Ethereum Blobs Compare?

Four public layers compete to hold rollup data: Ethereum blobs, Celestia, EigenDA and Avail. L2BEAT's last past-day figures, from 16 October 2026, showed each carrying less than its capacity, and the tokens of the three non-Ethereum layers trade about 97% to 99% below their highs (L2BEAT, 2026; CoinPaprika API, 2026-11-05). L2BEAT has since removed its data-availability pages and says Ethereum is the only DA layer it still tracks (L2BEAT GitHub, 2026).

Celestia: a dedicated DA chain

Celestia is a layer 1 built only to order and publish data, leaving execution and settlement to the chains above it. Its Matcha upgrade went live on Mainnet Beta on 24 November 2025 (Celestia docs, 2026). Docs on 5 November 2026 list a 32 MiB maximum block capacity and an 8 MiB maximum transaction. L2BEAT's past-day average on 16 October was 0.0073 MiB/s, 0.14% of capacity, and Eclipse posted 98.66% of the data (L2BEAT, 2026). Posting a MiB costs at least about $0.015 at the 0.004 utia minimum gas price and TIA at $0.459 on 5 November 2026 (author's calculation). Celestia's Fibre protocol posted a vendor benchmark of 3.07 Tb/s on 120 test validators on 1 October 2026 (Celestia, 2026), and it went live on mainnet with the v10 upgrade on 20 October 2026, starting with smaller uploads and a slower cadence than the test (Crypto Briefing, 2026). TIA is 97.8% below its February 2024 high (CoinPaprika API, 2026-11-05).

EigenDA: restaked operators and reserved bandwidth

EigenDA is a DA service run by operators who restake ETH through EigenLayer. On 17 June 2025 Eigen Labs launched EigenCloud, a platform that bundles the EigenLayer services, EigenDA among them; EigenLayer stays the restaking protocol and EIGEN its token (Eigen Labs, 2025). How EigenLayer restaking works ↗ explains the stake behind its operator set. EigenDA's own claims are 50 MB/s for V2 and 100 MB/s in its docs (Eigen Labs, 2025). L2BEAT's past-day average on 16 October was 0.05 MiB/s and its record 15.4 MiB/s on 3 July 2024, with RISE posting 82.08% (L2BEAT, 2026). L2BEAT listed no slashing and a permissioned committee. The 2024 list price was 0.015 ETH per GB, about $37 at ETH $2,450 (Kairos Research, 2024). EIGEN is 96.7% below its December 2024 high (CoinPaprika API, 2026-11-05).

Avail: the smaller sampling chain

Avail is a data chain that combines erasure coding, KZG commitments and sampling. It reports 4 MB blocks on mainnet and a goal of 10 GB blocks, both vendor figures (Avail, 2025). L2BEAT's past-day average on 16 October was under 0.00001 MiB/s against 0.2 MiB/s of capacity, and Lens posted 96.16% of the 353 KiB (L2BEAT, 2026). Chainscore's live tracker put posting at about $0.0003 per MB on 16 October 2026 (Chainscore, 2026). Sophon and Lens use Avail as validiums. Avail's Nexus mainnet, launched 9 December 2025, moved the pitch toward cross-chain coordination (Avail, 2025). AVAIL is 98.8% below its July 2024 high (CoinPaprika API, 2026-11-05).

Ethereum blobs and PeerDAS: the in-protocol rival

Ethereum is the in-protocol competitor. Fusaka added PeerDAS on 3 December 2025, and a second blob-parameter fork raised the blob target and maximum per block to 14 and 21 on 7 January 2026; live eth_config on 5 November 2026 shows no later fork scheduled (Ethereum Foundation, 2025; Ethereum RPC, 2026). Target capacity is 0.146 MiB/s, and L2BEAT's past-day average on 16 October was 0.038 MiB/s, 25.86% of it, with Base posting 33.59% (L2BEAT, 2026). Counting full blobs, the 7,200 blocks to 10:58 UTC on 5 November 2026 carried 6.8 blobs each on average, 0.071 MiB/s or 48.6% of target (Ethereum RPC, 2026). A MiB of blob data cost about $0.010 at the 10:53 UTC blob price that day and at the day's median, at ETH $2,417 (author's calculation from Ethereum RPC).

Data current as of November 2026.

The throughput figures show how much data each layer carries; the next question is whose rollups put it there.

Which Rollups Use Which Data Availability Layer Today?

Ethereum blobs or calldata carry 86.5% of layer 2 value secured ↗, across 42 of 83 projects, while EigenDA, Celestia and Avail together carry about 2%. Value secured is L2BEAT's measure of assets held on a chain, not total value locked (L2BEAT, 2026).

Where the value secured sits

L2BEAT tracked 83 layer 2 projects, layer 3s excluded, with $38.8B of value secured at 09:00 UTC on 5 November 2026 (L2BEAT, 2026). Base Chain ($14.8B) and Arbitrum One ($9.6B) post to Ethereum. Custom setups hold 11.4%, mostly Polygon PoS at $3.88B, which L2BEAT lists with a custom DA solution. EigenDA serves seven projects with $594M, led by Celo and Ronin. Celestia serves six with $107M, led by Manta Pacific at $89M. Avail serves Sophon and Lens with $6M. Ten projects with $122M rely on data availability committees, groups of named signers who attest that data exists, among them the real-world-asset chain Plume Network ↗. One project lists no DA solution. The measure counts canonically bridged, externally bridged and natively minted assets, so a chain's own token adds to its total (L2BEAT, 2026).

Who moved and who stayed

Mantle announced on 22 January 2026 that it would move its data from EigenDA to Ethereum blobs after Fusaka raised blob capacity, and L2BEAT now lists it on blobs with $1.3B of value secured (Mantle, 2026; L2BEAT, 2026). Mantle keeps EigenCloud for other services, and the move followed its 16 September 2025 upgrade to OP Succinct validity proofs ↗ (Coinspeaker, 2026). The move shows that a DA choice can be reversed and that the chain pays for the change. One customer dominates each external layer's daily volume, as the shares above show, so a single rollup's activity sets that layer's headline throughput.

Data current as of November 2026.

Leaving Ethereum for data changes what a project is called, and the name carries a security claim.

Rollup, Validium or Optimium: What Does Leaving Ethereum for Data Cost?

L2BEAT reserves the word rollup for projects whose data sits on a permissionless layer 1. Moving data to Celestia, EigenDA, Avail or a committee makes a project a validium or an optimium, which L2BEAT calls a light L2 with added trust assumptions.

How L2BEAT separates rollups from validiums and optimiums

A rollup posts data to Ethereum and proves its state with validity proofs, as a zero-knowledge (ZK) rollup ↗, or with fraud proofs, as an optimistic rollup. A validium uses validity proofs but publishes data externally; an optimium uses fraud proofs and publishes data externally (L2BEAT FAQ, 2026). L2BEAT first excluded both from its definition of L2s and now tracks them as light L2s, against strong L2s such as rollups. On 5 November 2026 it categorised 95 projects, layer 3s included: 12 ZK-rollups, 10 optimistic rollups, two validiums, one optimium and 70 labelled Other (L2BEAT, 2026). Sophon and Lens are the validiums, both on Avail; Celo, on EigenDA, is the optimium.

What committees and external layers add to the trust list

An external DA layer adds its own validators to the trust list. L2BEAT's data-availability pages, retired in October 2026, labelled Celestia and Avail as secured by staked assets with sampling for fraud detection, and listed EigenDA's V2 certificate verifier with no slashing and a permissioned committee (L2BEAT, 2026). A data availability committee is weaker still. Nine of the 13 committees L2BEAT listed there had a single signer, so one party attests that the data exists, and most name Ethereum as the fallback if the signers fail (L2BEAT, 2026). The DA row on a project's L2BEAT page states which case applies. Moving data off Ethereum buys capacity and a lower data bill, and costs a weaker answer to whether anyone can rebuild the chain without its operator.

FAQ

?Is a modular blockchain the same thing as a layer 2 or a rollup?

No. Modular is a design approach in which separate layers handle execution, settlement, consensus and data availability. A rollup is one kind of execution layer within that design, and it can use Ethereum or an external layer for data. A layer 1 such as Celestia can itself be a modular component that leaves execution to the chains above it (Celestia docs, 2026).

?What is the difference between a sovereign rollup and a smart-contract rollup?

L2BEAT calls a rollup without a bridge to its base layer, or one viewed without the bridge, a sovereign rollup (L2BEAT glossary, 2026). Its own nodes enforce the rules and no settlement contract verifies proofs. A smart-contract rollup posts proofs or state roots to a contract on a settlement layer such as Ethereum, which lets users withdraw through a trust-minimised bridge.

?What happens to a rollup's users if its external data availability layer goes offline or withholds data?

Without the data, nobody except the operator can rebuild the rollup's state or construct a withdrawal proof, so funds can be frozen until the data returns. L2BEAT labels external DA as a distinct risk for this reason (L2BEAT, 2026). Some committee-based designs name Ethereum as a fallback if the committee stops signing, which limits the damage.

?Which token pays for posting data on Celestia, EigenDA and Avail?

Celestia fees are paid in TIA, set by a minimum gas price in utia (Celestia docs, 2026). EigenDA accepts payment in ETH, EIGEN or a project's own token (EigenCloud docs, 2026). Avail uses its AVAIL token to pay for posting data (Avail, 2025). Ethereum blobs are paid in ETH.

?How long does a data availability layer keep the data, and who stores it afterwards?

Celestia's minimum pruning window is seven days and one hour, after which full-history storage falls to archival nodes and data providers (Celestia, 2025; Celestia docs, 2026). Ethereum nodes drop blobs after about 18 days with no protocol promise beyond it. In both cases the rollup's own nodes, block explorers and archive services keep the copies that users rely on later.

?Where can a reader check which data availability layer a rollup uses?

L2BEAT's scaling summary shows a DA badge on each project, such as "Ethereum with blobs", "EigenDA" or "Celestia". L2BEAT retired its separate data availability section in October 2026 and now tracks only Ethereum blobs on a Blobs page, so throughput for external layers has to come from their own explorers (L2BEAT GitHub, 2026). Each project page repeats the choice in a Data Availability row under its risk summary.

?If external data availability is cheaper, do rollup users pay less?

Little, on the largest chains. In the seven days to 4 November 2026, tracked L2s paid Ethereum $40.9K in rent ↗, 3.6% of the $1.145M users paid in fees, much of it on one day of congestion, and Base paid 1.4% (growthepie, 2026). Execution costs and operator margin make up the rest, so a cheaper data layer shrinks a small line, and nothing obliges the operator to pass the saving on.

?Can a rollup change its data availability layer after launch?

Yes, with a network upgrade. Mantle announced on 22 January 2026 that it would move from EigenDA to Ethereum blobs. Mantle cited the burden of maintaining a custom integration in its OP Stack batcher and node software (Mantle, 2026). The change touches node software and the contracts that verify data commitments, so the operator, not the users, carries the engineering work, and it set no completion date in the announcement.

References / Sources