Summary
- Data availability (DA) ensures that anyone can retrieve the necessary information (transactions) for verifying a blockchain or rollup.
- Posting all rollup data on Ethereum is secure, but very expensive and limited by Ethereum’s throughput.
- Celestia, Avail, and EigenDA each tackle off-chain data availability differently, with varying trade-offs in trust, throughput, finality, and cost.
- Eclipse selected Celestia for its "GigaCompute" performance vision, as Celestia's high-throughput, and low-cost DA remove data bottlenecks, enabling Eclipse's focus on optimized execution.
- By understanding how each solution works (and how it anchors proofs to Ethereum), you can select the best DA layer for your rollups or L2 projects.
Why Data Availability Matters
Whether a rollup uses fraud proofs (optimistic) or validity proofs (zero-knowledge), it must post enough information on L1 so that anyone can reconstruct its state. In optimistic rollups, this typically means publishing transaction data, while validity rollups often post state diffs. Historically, all of that data or state diffs got put directly on Ethereum as calldata—a highly decentralized approach, but also one that is very expensive, given Ethereum’s throughput limits.
Data Availability (DA) layers solve this cost challenge by offering cheaper ways to publish rollup data off-chain (or on a dedicated chain) while preserving the essential property that all needed data remains accessible. Different DA layers—Celestia, Avail, and EigenDA—approach this problem with unique trade-offs in terms of:
- Cost (how cheap is it per byte to store data?)
- Throughput (how much data can be handled per unit of time?)
- Trust assumptions (do we trust a new chain, a committee, or Ethereum stakers?)
- Finality (how long until the data availability is guaranteed and cannot be undone?)
This balance of the cost, scale, security, and finality shapes how each solution fits different rollup designs and developer needs.
Key Takeaways
Each solution, Celestia, Avail, & EigenDA, targets somewhat different segments of the data availability spectrum:
- Celestia:
- Fully decentralized DA checks via sampling, at the cost of a longer finality wait (~10 min) and reliance on a standalone chain.
- Great for rollups seeking low fees, high throughput or sovereign rollups (rollups that don’t rely on Ethereum for settlement or dispute resolution, but instead secure themselves using their own logic plus Celestia DA) and modular chain design.
- Avail:
- Similar in design to Celestia, optimized for fast data finality (~40s) using KZG commitments, integrates DAS.
- Lower cost per MB than Celestia, but lower economic security as well.
- EigenDA:
- Ties its economic security to Ethereum’s by letting validators restake ETH for data availability, with no slashing for misbehavior or fraud.
- Uses ETH and native tokens for payment, aligning incentives and penalties directly with Ethereum’s economy.
- Achieves high throughput via a off-chain DAC (data available committee), but no publicly verifiable data availability guarantees.
- Great for large, Ethereum-centric L2s wanting high data capacity that can tolerate Ethereum’s finality (~12-15 min).
Why Eclipse Chose Celestia
Eclipse’s approach to “GigaCompute” we plan to ship in the GSVM client requires a scalable, low-cost way to publish transaction data off-chain. An optimistic rollup model allows Eclipse to achieve hardware-limited transaction throughput, bypassing the ZK proof generation overhead that would impede our specific concurrency and hardware optimizations.. This, however, means data availability may become a critical bottleneck—if data can’t be efficiently published, the rollup’s entire performance advantage is lost.
Celestia offers exactly the kind of flexible, high-throughput DA that Eclipse requires:
- Modular Design
- Celestia focuses exclusively on ensuring data availability, leaving execution to the rollup and settlement to Ethereum. This “modular ethos” aligns with Eclipse’s philosophy of providing strong security guarantees without making compromises on performance. It lets Eclipse push aggressive concurrency, caching, and scheduling optimizations, without being constrained by a monolithic design with long upgrade cycles.
- Cost Efficiency & Scaling
- Celestia employs Data Availability Sampling (DAS), where light clients verify data availability by checking only small, random samples instead of downloading the entire block. This core design decouples block size from the resource requirements of individual verifying nodes, allowing Celestia to increase block sizes significantly and handle more data cost-effectively. Eclipse’s GigaCompute vision depends on this kind of affordable, high-throughput blockspace for large-scale workloads like AI, gaming, and DePIN.
- Data Availability Sampling (DAS)
- Celestia’s DAS approach ensures that honest nodes can verify data availability by checking severalrandom shares, rather than needing to download everything. This matches Eclipse’s push to minimize hardware overhead—nodes can stay lightweight, while still guaranteeing no part of the block data is being withheld.
- Optimistic Security Model
- Because Eclipse aims to use fraud proofs for execution correctness, it can rely on Celestia’s fraud-proof approach for DA. If a Celestia block producer withholds data, honest participants can detect it during sampling and raise a challenge. In practice, given high validator incentives, malicious withholding should be exceedingly uncommon.
Source: Fees paid per MB posted
In March 2025, Eclipse continued to realize dramatic DA cost savings by utilizing Celestia. Eclipse's data posting cost averaged approximately \$0.07 per MB, significantly lower than the estimated \$3.83 per MB for Ethereum blobs during the same period. This translates to Celestia being over 55 times more cost-effective per megabyte for data posting in March. This finding aligns with our previous analysis from January 2025, where we demonstrated Celestia's cost advantage, noting a 510X greater efficiency measured on a per-transaction level.
Overall, Celestia’s throughput, affordability, and modular design complement Eclipse’s mission of building an ultra-performant Solana-based rollup on Ethereum. By reducing DA bottlenecks and letting Eclipse harness flexible blockspace, Celestia ensures that Eclipse can focus on hardware–software co-design, concurrency, and dynamic scaling—without paying prohibitively high fees or sacrificing throughput at the data layer.
Comparison Table
Below is a brief snapshot of how Celestia, Avail, and EigenDA compare on some of the key metrics:
Summary of Strengths & Weaknesses
Celestia
- (+)
- Decentralized and anyone can verify DA via sampling; highly scalable throughput per block
- Simple design leveraging proven BFT consensus and fraud proofs
- Supports DA bridging to Ethereum (Blobstream)
- (−)
- New separate chain to trust (100 active validators, 260 total validators and growing)
- Longer wait for DA finality (fraud-proof window ~10 minutes)
- Potential stake centralization due to DPoS model.
Avail
- (+)
- Immediate DA proof at finality thanks to KZG commitments (no challenge period)
- Fast confirmation (~40s)
- Robust validator decentralization plan (NPoS, many validators)
- Supports DA bridging to Ethereum (Vector)
- (−)
- Economic security is only around 1/10th of Celestia's or EigenDA's
- Currently has the lowest throughput compared to competitors.
EigenDA
- (+)
- Ethereum-centric security: fees paid in ETH, restaked ETH and native tokens for node staking
- Extremely high throughput potential (horizontal scaling)
- Cheap for rollups (only post small blob refs on L1)
- Accepts payment in native tokens and supports native token staking, and leverages existing Ethereum validators
- (−)
- Currently, a permissioned committee with incentives – users can’t verify DA without running a full node, which is a centralization of trust
- Not easy to detect fraud, and detection on-chain is not trivial; currently has a centralized component (disperser) for liveness guarantees which is a weakness until decentralized
- New operators can be rejected by EigenDA quorums
- Reliant on the correctness of the EigenLayer smart contracts (contract risk); Ethereum dependency means you inherit Ethereum’s latency (and any issues therein)
- No slashing on EigenDA yet. EigenLayer rolled out slashing on April 17, opt-in for AVS. The status for EigenDA is unclear.
References
- Celestia Documentation & Forum
- Avail Resources
- EigenDA / EigenLayer
- General Comparisons
- KZG
- DA layers Economic Security