Modular Blockchains Explained: Settlement, DA, Execution, Consensus, and the Celestia/EigenDA/Avail Stack (2026)

The thesis in one paragraph

A blockchain does four things: it executes transactions (computes new state), it provides data availability (publishes the data so others can verify), it settles disputes (acts as the source of truth on which fork is canonical), and it reaches consensus (orders transactions). A monolithic chain bundles all four into one. A modular chain splits them across specialized layers, on the belief that specialization wins on cost, throughput, and security per dollar of resources.

The modular thesis crystallized in the Ethereum scaling debate of 2020-2022 and went into production in late 2023. By April 2026 it is the dominant scaling paradigm for the Ethereum-aligned ecosystem, with a credible monolithic counter-thesis still active on Solana, BSC, and a handful of high-throughput L1s.

The four layers

Execution

Execution is the layer that runs transactions and computes the new state. In modular Ethereum, execution lives on rollups: Arbitrum, Optimism, Base, ZKsync, Starknet, Linea, Scroll, Polygon zkEVM, Blast, and 80+ others as of 2026. Each rollup runs its own EVM (or non-EVM, in the case of Starknet's Cairo) and can iterate on execution semantics independently. Optimistic rollups assume execution is correct and wait for fraud proofs; ZK rollups generate validity proofs upfront.

Data availability

Data availability is the guarantee that the data needed to reconstruct rollup state is published and accessible. If DA is missing, no one can verify the rollup's claims, and an attacker could publish a fraudulent state root. The DA layer accepts data, charges for storage (typically time-bounded — Ethereum blobs prune after ~18 days), and provides an availability guarantee via consensus or sampling. The major DA layers as of 2026 are Ethereum (via EIP-4844 blobs), Celestia, EigenDA, and Avail.

Settlement

Settlement is the layer that adjudicates the canonical history. For most Ethereum L2s, settlement happens on Ethereum L1 — a smart contract on L1 holds the canonical state root, accepts proofs (fraud or validity), and processes withdrawals. Settlement gives the rollup an enshrined bridge to L1 — value can flow between L2 and L1 at the trust assumption of L1. Sovereign rollups skip enshrined settlement and govern themselves at the social layer.

Consensus

Consensus is ordering — the answer to "in what sequence did transactions occur?" In modular Ethereum, consensus over rollup transactions is initially provided by a single sequencer operated by the rollup team (Optimism Foundation, Offchain Labs, Coinbase, Matter Labs). Decentralized sequencer roadmaps exist for most major rollups but full decentralization is still in progress in 2026. Some sovereign rollups push consensus to the DA layer (Celestia validators order the rollup's transactions).

Celestia mechanics

Celestia mainnet launched in October 2023 — the first production blockchain explicitly designed as a DA-only layer. It uses Tendermint consensus secured by ~100 validators, the Cosmos SDK for the chain framework, and Reed-Solomon erasure coding plus 2D Merkle commitments for data availability sampling.

The key innovation: Celestia full nodes don't validate execution. They only verify that posted data is available. This decouples DA throughput from execution complexity. A rollup posts its blobs to Celestia, the Celestia network reaches consensus on the data, light clients sample to verify availability, and the rollup's own execution layer (off-chain) processes the transactions.

TIA, Celestia's native token, has three uses: staking for validators, fee payment for posting blobs, and governance. Blob fees are paid in TIA (with paymaster-like services accepting other tokens). As of April 2026 Celestia processes 8-12 MB/s of average DA throughput, with peaks exceeding 30 MB/s during rollup batch settlement bursts.

Major Celestia DA users: Manta Pacific, Eclipse Mainnet, Caldera-deployed sovereign chains, Polygon CDK chains using the Celestia plug-in, and a long tail of app-rollups.

EigenDA economics

EigenDA launched on Ethereum mainnet in April 2024 as Eigen Labs' own DA service built on EigenLayer restaking. The economic model is fundamentally different from Celestia. Instead of running its own consensus, EigenDA borrows security from restaked ETH delegated to operators on EigenLayer. Operators are slashable for failing to attest to data availability.

As of April 2026, EigenDA is secured by approximately $8-15 billion in restaked ETH allocated to its operator set, making it the highest-economic-security DA layer in absolute dollar terms. Throughput is 15-25 MB/s on average. The trust assumption is the EigenLayer slashing model plus operator-set diversity.

Major EigenDA users: Mantle, Celo (post-migration to L2), Arbitrum AnyTrust DA backups, EigenLayer-aligned rollups in the Eigen Labs ecosystem.

The key tension for EigenDA is whether "borrowed security via restaking" is genuinely as strong as Celestia's native consensus security. The answer depends on how robust EigenLayer's slashing actually proves to be when stress-tested.

Avail comparison

Avail is the original DA layer research from Polygon, spun out as an independent project in 2024 with its own AVAIL token and validator set. It uses KZG commitments (rather than Celestia's 2D Reed-Solomon) for DA proofs, which produces smaller proofs but tighter cryptographic assumptions.

Avail's positioning is the technical middle ground between Celestia (sovereign chain) and EigenDA (restaked). It has its own consensus, its own slashing, but with KZG-based proofs that are friendlier to ZK rollup verification. Avail throughput as of April 2026 is 4-7 MB/s on average — lowest of the three but with adequate headroom for committed users.

EIP-4844 blobs and the Dencun fork

The most important DA event of 2024 was not Celestia or EigenDA but Ethereum's own EIP-4844, activated in March 2024 as part of the Dencun hard fork. EIP-4844 added blob-carrying transactions to Ethereum L1 — large data blocks (128KB each, 6 per block) committed via KZG and pruned after ~18 days.

The economic effect was immediate. Pre-Dencun, an L2 swap cost $0.10-0.50 in fees, dominated by L1 calldata posting cost. Post-Dencun, the same swap costs $0.005-0.05. L2 transaction count surged from ~5M/day pre-Dencun to ~25-40M/day by Q4 2024.

For the modular ecosystem, Dencun was decisive: it made Ethereum's own DA economically competitive with alt-DA layers for most use cases. Celestia, EigenDA, and Avail still have lower per-byte costs at the margin, but Ethereum blobs come with the strongest possible DA guarantee (full Ethereum consensus security) — a real moat.

The roadmap from here: PeerDAS (sampling-based DA on Ethereum L1) targeted for the Fusaka or Glamsterdam fork in late 2026 or 2027, scaling blob count from ~6 per block to 32-64; FullDAS later, scaling further. Vitalik's published target is 1-2 MB/s of DA per block on Ethereum L1 by the end of the decade.

Rollup framework wars

The execution layer has consolidated around four major frameworks:

OP Stack — built by Optimism, underlies Optimism, Base, Worldchain, Soneium, Unichain, and 60+ chains in the Superchain. Optimistic rollup, fault-proof system live since 2024, shared sequencer roadmap, AggLayer-equivalent (Superchain) interop. Largest framework by chain count.

Arbitrum Orbit — built by Offchain Labs, underlies Arbitrum One, Arbitrum Nova, and dozens of Orbit chains (ApeChain, XAI, Reality, etc.). Optimistic rollup using Nitro/Stylus, fraud proofs via interactive challenges, more sequencer customization than OP Stack. Largest framework by per-chain TVL.

Polygon CDK — built by Polygon Labs, ZK-based, underpins Polygon zkEVM and CDK chains. Unified under the AggLayer cross-chain settlement layer in 2024-2025. Strong on ZK proof technology and DeFi-grade security; smaller chain count than OP Stack and Orbit.

ZK Stack — built by Matter Labs, underlies ZKsync Era, Sophon, Cronos zkEVM, Lens Network, and others. Boojum prover, hyperchain interop, native account abstraction. ZK-first design.

The competitive frontier through 2026 is interop. Each framework has its own bridge mesh: Superchain (OP Stack), AggLayer (CDK), Hyperbridge / ZK Stack mesh, Arbitrum's cross-Orbit fast finality. Cross-stack interop (a Base user transacting with a ZKsync user without going through L1) is largely missing in 2026 and is the bottleneck for further L2 fragmentation.

Sovereign rollups

A sovereign rollup uses an external DA layer for data availability and consensus over ordering, but does not use any external chain for settlement. The sovereign chain's own community decides which fork is canonical. There is no enshrined bridge to L1.

The trade-off: sovereign rollups gain governance flexibility (a sovereign chain can hard-fork without an L1 contract upgrade) but lose enshrined-bridge security (assets cannot trustlessly flow to L1; bridges are external smart contracts with their own risk).

In 2026, sovereign rollups remain a niche pattern. The clearest examples are several Celestia-DA sovereign chains, Eclipse Mainnet (Solana VM on Ethereum settlement, Celestia DA — partially sovereign in some configurations), and a handful of Sovereign Labs SDK chains. The category is technically interesting but commercially much smaller than enshrined-bridge L2s.

The Solana monolithic counter-thesis

The modular thesis is contested. The monolithic counter-argument, most loudly articulated by Anatoly Yakovenko and the Solana ecosystem, is that:

Hardware costs are minimal — a high-end Solana validator costs $10-30k/year.
Parallel execution (Sealevel) plus in-memory state plus Firedancer optimizations can deliver 50-200k effective TPS on a single chain.
Fragmenting users across a dozen rollups produces worse UX (bridges, gas tokens, latency) than running everything in one place.
Composability — atomic, latency-free interaction between protocols — is degraded by cross-rollup boundaries.

The empirical scoreboard through 2026 is genuinely contested. Solana has captured retail trading, memecoins, consumer apps, and a meaningful slice of stablecoin payments — areas where modular Ethereum has not won. Modular Ethereum has captured DeFi TVL ($90-120B vs Solana's $10-15B), institutional flows, and the L1 narrative.

The likely endgame is multi-paradigm: monolithic chains for latency-sensitive consumer flows (Solana, BSC, Sonic, Sui), modular Ethereum for value storage, settlement, and DeFi composability. The two paradigms are converging in some ways — Solana is exploring SVM rollups (Eclipse, Sonic), Ethereum is exploring more monolithic-feeling L2s (single-slot finality, native rollups).

Bottom line

The modular split is the dominant 2026 scaling paradigm for ETH-aligned chains, validated by the success of EIP-4844 blobs, the Celestia/EigenDA/Avail DA market, and the maturation of the four major rollup frameworks. The split has produced 90-99 percent fee reductions, 5-10x throughput gains, and a Cambrian explosion of execution environments.

The remaining frontiers are interop (cross-stack composability), full danksharding (PeerDAS/FullDAS to scale Ethereum's own DA), and the maturation of decentralized sequencers. The monolithic counter-thesis remains alive on Solana and a handful of high-throughput L1s, capturing real workloads that the modular stack has not yet served well.

For capital allocators and infrastructure builders: the safe bet is on continued modular dominance for DeFi and value storage, with monolithic chains carving durable niches in consumer apps and high-frequency trading.

Sources and further reading

Celestia documentation — https://docs.celestia.org
EigenDA documentation — https://docs.eigenda.xyz
Avail technical documentation — https://docs.availproject.org
EIP-4844 specification — https://eips.ethereum.org/EIPS/eip-4844
Vitalik's modular roadmap posts — https://vitalik.eth.limo
OP Stack, Arbitrum Orbit, Polygon CDK, ZK Stack official docs
Sovereign Labs technical writing — https://www.sovereign.xyz

About the author

DeFi Intel Research is the in-house research team at DeFi Intel, focused on on-chain capital markets, MEV, ZK infrastructure, and verifiable AI.

Last updated: 2026-04-30

Entities mentioned

Celestia EigenDA Avail Ethereum Solana BNB Chain Arbitrum Optimism

Frequently asked questions

What is a modular blockchain?

A modular blockchain is a design philosophy that splits the four core functions of a blockchain — execution, settlement, data availability (DA), and consensus — across separate specialized layers, rather than bundling them into a single monolithic chain. The thesis is that specialization wins: a chain optimized purely for ordering transactions and providing data availability (Celestia) can be cheaper and more scalable than a chain that has to do everything (Solana, BSC). Execution moves to rollups that inherit security from a settlement layer (Ethereum) and post their data to a DA layer (Ethereum blobs, Celestia, EigenDA, Avail). The modular thesis emerged from the Ethereum scaling debate of 2020-2022 and went into production with the Celestia mainnet launch in October 2023 and EIP-4844 (proto-danksharding) on Ethereum in March 2024.

What is the difference between Celestia, EigenDA, and Avail?

All three are dedicated data-availability layers — they store rollup transaction data and provide a guarantee that anyone can reconstruct it — but they differ in trust model and economics. Celestia is a sovereign Cosmos-SDK chain with its own validator set and consensus, securing DA via its own proof-of-stake economic security backed by the TIA token. EigenDA is built on EigenLayer restaking — it borrows economic security from restaked ETH, making it the highest-security DA layer in absolute dollar terms but with the dependence on the EigenLayer slashing model and operator quality. Avail is the spinout of Polygon's original DA research, launched as an independent token-incentivized chain in 2024 — it uses KZG commitments and its own validator set, positioning between Celestia (sovereign chain) and EigenDA (restaked) on the trust-model spectrum. As of April 2026, Celestia handles roughly 8-12 MB/s of average DA throughput, EigenDA 15-25 MB/s, Avail 4-7 MB/s — but all three are far below their theoretical capacity.

What is EIP-4844 and proto-danksharding?

EIP-4844 (proto-danksharding) was activated on Ethereum mainnet in March 2024 as part of the Dencun hard fork. It introduced a new transaction type called blob-carrying transactions and a new data primitive called blobs — large data blocks (up to 128KB each, 6 per block) that are committed to via KZG commitments but are not part of the EVM execution payload. Blobs are pruned from full nodes after roughly 18 days, since they only need to be available long enough for fraud-proof or validity-proof verification. Each blob costs a separate fee priced via its own EIP-1559 mechanism. EIP-4844 reduced L2 data-posting costs by 90-99 percent overnight; major rollups (Arbitrum, Optimism, Base, ZKsync) saw transaction fees drop from $0.10-0.50 to $0.005-0.05 per swap. Full danksharding is the eventual upgrade, scaling to ~64 blobs per block via PeerDAS and FullDAS sampling — likely 2026-2028 deliverable.

What is data availability sampling and why does it matter?

Data availability sampling (DAS) is the cryptographic technique that lets light clients verify a large block is available without downloading the whole block. The block is encoded with Reed-Solomon erasure codes and committed via Merkle or KZG commitments; light clients sample random chunks and verify the commitment. If enough independent light clients each sample successfully, the block is statistically guaranteed to be available — even if no single client has all of it. DAS makes scalable DA possible: Celestia, Avail, EigenDA, and the planned Ethereum FullDAS all rely on it. Without DAS, a DA layer either becomes prohibitively expensive (every full node downloads everything) or insecure (blocks can be selectively withheld). The PeerDAS upgrade planned for the Fusaka or Glamsterdam Ethereum forks will bring DAS to L1 itself.

What is the difference between OP Stack, Polygon CDK, Arbitrum Orbit, and ZK Stack?

All four are rollup frameworks — software toolkits that let teams launch a new L2 with shared sequencer, fraud or validity proofs, and bridge infrastructure — but they differ on technology and ecosystem. OP Stack (by Optimism) is the optimistic-rollup toolkit underlying Optimism, Base, Worldchain, Soneium, Unichain, and a 60+ chain Superchain ecosystem; it shares a sequencer set, an L1 bridge, and a roadmap. Arbitrum Orbit (by Offchain Labs) is the Nitro-based optimistic-rollup framework underlying Arbitrum One, Arbitrum Nova, ApeChain, XAI, and dozens of Orbit chains; it offers more sequencer customization than the OP Stack at the cost of less interop. Polygon CDK (Chain Development Kit) is the ZK-based rollup framework from Polygon Labs, originally built on the Polygon zkEVM stack and unified under the AggLayer cross-chain settlement layer in 2024-2025. ZK Stack is the Matter Labs (ZKsync) ZK-rollup framework underlying ZKsync Era, Sophon, Cronos zkEVM, Lens, and others; it offers boojum prover, hyperchains, and an interop-first mesh. The competitive picture: OP Stack leads on chain count, Orbit leads on per-chain TVL, CDK and ZK Stack lead on ZK technology.

What is a sovereign rollup?

A sovereign rollup is a rollup that uses an external chain (like Celestia) for data availability and consensus over transaction ordering, but does not use any external chain for settlement. Settlement — i.e. the canonical truth about which fork is correct — is determined by the sovereign rollup's own social and execution layer. There is no enshrined bridge. Forking the chain is a permissionless action governed by the sovereign chain's own community, more like a Cosmos zone or an L1 than a smart-contract rollup. Sovereign rollups trade enshrined-bridge security for governance flexibility — a sovereign rollup can hard-fork without any L1 contract upgrade. Examples in 2026 include several Celestia-based sovereign rollups (Eclipse Mainnet, Manta Pacific in some configurations, Caldera-deployed sovereign chains), and the Sovereign Labs SDK targets this model.

How does the Solana monolithic counter-thesis fit in?

Solana, BSC, Sui, Aptos, and to some extent Sonic represent the monolithic counter-thesis: that the modular split sacrifices too much UX, latency, and composability for marginal scalability gains. The monolithic argument is that hardware is cheap (a high-end Solana validator costs $10-30k/year), parallel execution (Sealevel) plus in-memory state (Solana's account model) plus Firedancer optimizations can deliver 50-200k TPS on a single chain, and that fragmenting the user across a dozen rollups creates worse UX than running everything in one place. The empirical scoreboard through 2026 is genuinely contested: Solana has captured retail trading, memecoins, and consumer apps in a way modular Ethereum has not, while modular Ethereum has captured DeFi TVL, institutional flows, and stablecoin payments. The likely endgame is a multi-paradigm world: monolithic chains for latency-sensitive consumer flows, modular Ethereum for value-storage, settlement, and DeFi composability.

What is the realistic 2026-2028 modular roadmap?

Three threads to track. First, on Ethereum L1: PeerDAS in Fusaka or Glamsterdam (likely late 2026 or 2027), expanding blob throughput from ~6 per block to 32-64 per block; later FullDAS extends DAS to all consensus participants. Second, on alt-DA layers: Celestia is rolling out Lotus-Lemongrass upgrades and expanding to ZK-rollup-friendly throughput targets, EigenDA is iterating on operator quality and slashable security, Avail is pushing optimistic + ZK light clients. Third, on rollup frameworks: cross-stack interop (Superchain↔AggLayer↔Orbit↔ZK Stack) is the bottleneck — every framework is shipping its own interop layer in 2025-2026, with consolidation likely as users force interoperability. The macro thesis is that by 2027-2028 the modular split is irreversibly the dominant scaling paradigm for ETH-aligned chains, while monolithic chains continue to compete strongly for consumer-app workloads.

Stay current on this topic

Get the weekly DeFi Intel brief — entity-graph intelligence delivered weekly, free to your inbox.

Subscribe free See Pro →