The modular blockchain thesis has moved from whitepaper concept to functioning infrastructure. At its core, the idea is straightforward: monolithic blockchains (where every node handles execution, consensus, settlement, and data availability) are an engineering constraint, not a design necessity. Separating these concerns allows each layer to be optimised independently. Data availability is the layer that has seen the most competitive development in 2024 and 2025, with four distinct approaches now operating at production scale.

This article covers what data availability actually is, why it is the bottleneck it has become, and how the four main open networks (Celestia, EigenDA, Avail, and 0G) differ in their approach. The goal is a technical assessment, not a token thesis.

What data availability actually means

When a rollup posts a transaction batch to a DA layer, it is making a guarantee: any participant in the network can verify that the data behind a state transition exists and is accessible. This is distinct from data storage (keeping data indefinitely) and from execution (verifying that transactions are correct). DA is specifically about proving that data was published and is available to download, for a configurable window of time after publication.

The naive approach of posting everything to Ethereum calldata works but is expensive. A rollup paying Ethereum L1 gas for each byte of batch data inherits Ethereum's cost structure directly. The DA problem is how to guarantee data availability at dramatically lower cost per byte, without sacrificing the cryptographic guarantees that make the availability proof meaningful.

DA layers use two key cryptographic techniques: erasure coding (which encodes data redundantly so that the full dataset can be reconstructed from any sufficient subset of fragments) and data availability sampling (DAS), which allows light clients to probabilistically verify full availability by downloading only random small samples rather than the entire dataset.

Celestia

Celestia is the network that defined the DA category and holds roughly 50% of the market as of early 2026, having processed over 160 gigabytes of rollup data across 56+ active rollups on mainnet. It launched mainnet in late 2023 and has been iterating on capacity ever since.

The architecture is purpose-built. Celestia validators handle only two things: ordering transaction blobs and guaranteeing their availability through a combination of erasure coding and DAS. There is no execution layer. Rollup frameworks (Arbitrum Orbit, OP Stack, Polygon CDK, and others) have all added Celestia as a DA option. The result is that posting a blob to Celestia costs orders of magnitude less than posting equivalent data to Ethereum L1.

Current mainnet throughput is intentionally conservative at around 1.33 MB/s, with capacity increases gated by bandwidth requirements for light node participation. The Matcha upgrade (v6), expected on mainnet in Q1 2026, doubles block sizes to 128MB. This is the largest single capacity increase in Celestia's history. Looking further out, the experimental Fibre Blockspace protocol targets 1 terabit per second, a number that renders the current throughput figures almost irrelevant as a long-term constraint.

For node operators

Running a Celestia validator requires a full node with sufficient bandwidth for the current blob size limits. The hardware bar is modest by validator standards. The network is designed so that increasing throughput does not price out smaller validators through compute requirements, only through bandwidth. This matters for infrastructure operators who want to participate in DA security without deploying GPU capacity.

EigenDA

EigenDA, built by EigenLayer, takes a fundamentally different approach. Rather than launching an independent chain with its own validator set and token, EigenDA is a Data Availability Committee (DAC) secured by restaked ETH. Ethereum validators who have opted into EigenLayer as restakers can be slashed for misbehaviour on EigenDA, inheriting Ethereum's existing economic security rather than bootstrapping a new one.

The V2 release achieved 100 MB/s throughput (significantly higher than Celestia's current mainnet figures) by operating with a smaller, curated committee rather than a fully decentralised validator set. This is the core trust trade-off: EigenDA's security model requires trust in both EigenLayer's slashing contracts and the operators who make up the committee, whereas Celestia and Avail rely on open participation and cryptographic light client verification. Neither model is categorically inferior. The right choice depends on how you weight throughput against trust assumptions.

EigenDA V2 also introduced a direct dispersal model where rollups post data directly to EigenDA operators rather than routing through an Ethereum transaction. This reduces latency and costs substantially. OP Stack and Arbitrum integrations are live; several large rollup deployments have adopted it specifically for the throughput headroom.

Avail

Avail launched its mainnet in 2025 with a focus on unifying DA across multiple execution environments. This includes not just Ethereum rollups, but also Cosmos-based chains and sovereign rollups. The technical approach combines KZG polynomial commitments with DAS, providing stronger cryptographic guarantees per byte than pure erasure coding alone. KZG commitments allow a DA layer to produce a succinct proof that a specific piece of data is correctly encoded, which is verifiable by light clients without downloading the full block.

Avail's positioning is as a credibly neutral DA layer. It is not affiliated with Ethereum's validator set (unlike EigenDA) and not a rollup-specific infrastructure play. The project has attracted interest from rollup teams building on non-Ethereum stacks where Celestia's dominant Ethereum ecosystem integrations are less relevant.

The network also features Avail Nexus, a cross-chain proving layer that aggregates ZK proofs from multiple rollups and chains, enabling unified state verification across heterogeneous systems. This is a meaningful differentiator for applications that need to compose state across multiple rollup environments rather than posting data to a single DA layer in isolation.

0G (Zero Gravity)

0G is the most differentiated of the four, positioning itself not as a standalone DA layer but as a full decentralised AI operating system (dAIOS). The network integrates data availability, storage, and compute into a single stack explicitly designed for AI workloads. This includes specifically the use case of AI agents and models needing decentralised, verifiable data access.

The 0G DA component claims to outperform Celestia and EigenLayer on throughput and latency benchmarks through a separation of data flow and metadata consensus: data is transmitted peer-to-peer between storage nodes while only metadata is committed to the chain, dramatically reducing the on-chain bottleneck. The storage layer is designed for persistence. This is not just the temporary availability window that DA layers provide, but long-term decentralised storage with economic incentives for durability.

For infrastructure operators, 0G is worth tracking but occupies a different category from the other three. If your interest is in DA for rollup data posting, Celestia, EigenDA, and Avail are more directly comparable. If you are building infrastructure for AI systems that need decentralised data access or verifiable training data provenance, 0G's unified compute and storage integration is more directly relevant.

How to think about DA for infrastructure operators

If you operate nodes on any of these networks, the infrastructure requirements vary meaningfully by design choice. Celestia and Avail are independently-secured chains with open validator sets. Running a validator means staking their native tokens and participating in consensus, which requires 24/7 uptime, slashing protection, and monitoring. EigenDA operators are Ethereum validators who have opted into additional responsibilities through EigenLayer restaking. If you already run Ethereum validators, the marginal infrastructure cost of becoming an EigenDA operator is lower than starting a new validator set from scratch.

For teams building rollups rather than operating DA validators, the selection criteria are simpler: EigenDA offers the highest current throughput for high-volume rollups; Celestia offers the widest ecosystem tooling and the largest existing rollup community; Avail offers the strongest path if you need cross-chain composability or are building on non-Ethereum stacks.

The DA space will look different again by the end of 2026. Celestia's Matcha upgrade and Fibre Blockspace roadmap, EigenDA's continued V2 rollout, Avail's Nexus integrations, and 0G's AI-specific positioning are all actively developing. For now, these four represent the production-viable landscape. Each has different trust assumptions, throughput profiles, and ecosystem integrations.