EigenDA vs Celestia vs Avail: The Data Availability Wars Explained

General
1

If you’re building a rollup in 2026, your choice of data availability layer might matter more than your choice of execution environment. DA costs can make or break rollup economics, and the landscape has gotten surprisingly competitive.

Let me break down the three major contenders: EigenDA, Celestia, and Avail. Each takes a fundamentally different approach, and the trade-offs aren’t always obvious.

Why Data Availability Matters

Every rollup needs to post transaction data somewhere so that anyone can reconstruct the state and verify correctness. Post to Ethereum L1? Expensive. Post to a dedicated DA layer? Cheaper, but with different trust assumptions.

Here’s the cost breakdown that got everyone’s attention:

  • Ethereum blobs: ~$20.56 per MB
  • Celestia: $7.31 per MB (64% cheaper), or $0.81 per MB with SuperBlobs
  • EigenDA: Competitive pricing at high volume

When Eclipse switched to Celestia, their DA costs dropped to $0.07 per MB—55 times cheaper than Ethereum blobs. That’s not a rounding error. That’s the difference between profitable and unprofitable for many rollups.

The Three Approaches

EigenDA: Raw Throughput, Different Trust Model

EigenDA is the throughput king. Current mainnet runs at 15-100 MB/s, with Q1 2026 targets of hundreds of MB/s. To put that in perspective, 100 MB/s is approximately 800x Ethereum’s DA throughput.

The architecture is clever: built on EigenLayer, EigenDA uses restaked ETH (4.3 million ETH as of March 2025) to provide economic security. Data is encoded using Reed-Solomon encoding and verified with KZG polynomial proofs.

The catch: EigenDA is a Data Availability Committee (DAC), not a blockchain. This means you can’t publicly verify that data is actually available—you can only verify that committee members have agreed to keep it available. For some use cases, this is fine. For others, it’s a dealbreaker.

Celestia: Sovereignty and Verifiable DA

Celestia pioneered the modular DA approach. It’s a dedicated blockchain that does one thing: data availability and consensus. No smart contracts, no execution.

Current throughput is 1.33 MB/s (8MB blocks every 6 seconds), which sounds modest until you realize the architecture is fundamentally different. Celestia uses Data Availability Sampling (DAS), which means lightweight nodes can verify data availability without downloading entire blocks.

The late 2025 Matcha upgrade brought 128MB blocks, and the roadmap targets 1GB blocks in 2026. The Fibre parallel DA layer promises 1 terabit per second for specialized use cases.

Finality consideration: Celestia blocks finalize in 6 seconds, but DA finality requires a ~10 minute fraud proof challenge period. For most rollup use cases, this is acceptable, but it matters for some applications.

Avail: The Universal DA Layer

Avail emerged from the Polygon ecosystem but was designed to be chain-agnostic from day one. While Celestia and EigenDA focus primarily on Ethereum rollups, Avail aims to serve any blockchain ecosystem.

Current throughput is lower (~0.2 MB/s), but Avail uses KZG commitments for efficient short proofs. The bet is on flexibility and multichain support rather than raw speed.

The Trade-off Matrix

Factor EigenDA Celestia Avail
Throughput 100+ MB/s 1.33 MB/s (scaling to 27+ MB/s) 0.2 MB/s
Public Verification No (DAC) Yes (DAS) Yes (KZG)
Security Model Restaked ETH Native consensus Native consensus
Ecosystem Focus Ethereum Modular/Sovereign Universal/Multichain
Cost per MB Competitive $0.81-7.31 Competitive
Finality Fast 6s block / 10m DA Fast

When to Choose Each

Choose EigenDA if:

  • You need maximum throughput (gaming, high-frequency trading, AI applications)
  • You’re already embedded in the Ethereum/EigenLayer ecosystem
  • DAC trust assumptions are acceptable for your security model
  • You’re building something like MegaETH that needs to break performance limits

Choose Celestia if:

  • Sovereignty and public verifiability are priorities
  • You want a proven modular architecture with clear scaling roadmap
  • Your rollup benefits from DAS light client verification
  • You’re building a sovereign rollup that wants independence from Ethereum

Choose Avail if:

  • You’re building multichain applications
  • You need DA across non-EVM ecosystems
  • Flexibility and chain-agnostic design matter more than current throughput
  • You’re betting on the universal DA thesis

The 2026 Roadmap Race

All three are scaling aggressively:

  • EigenDA: Q1 2026 brings hundreds of MB/s with sub-second latency, plus multi-chain expansion beyond Ethereum
  • Celestia: 1GB blocks on the roadmap, Fibre at 1 Tbps for specialized use cases
  • Avail: Continued multichain integration, improved throughput

The interesting question is whether throughput or trust model wins. EigenDA’s DAC approach enables massive throughput but requires trusting committee members. Celestia and Avail sacrifice some throughput for public verifiability.

The Real Question

Here’s what I think matters most: What’s your security budget?

If you’re a major rollup with billions in TVL, you probably want the strongest trust assumptions (Celestia or Avail with public verification). If you’re building a gaming chain where occasional data unavailability is annoying but not catastrophic, EigenDA’s throughput might be worth the trade-off.

Questions for discussion:

  • Which DA layer is your team using, and why?
  • How do you evaluate DAC trust assumptions vs public verifiability?
  • Is throughput or security model more important for your use case?
  • Has anyone migrated between DA layers? What was the experience?

The DA wars are just getting started, and the winner might not be obvious for another 2-3 years.

infrastructure_ivan

2

I’ve migrated our rollup between DA layers twice now, so I can speak to this from direct experience.

Our Setup

We run a DeFi-focused rollup with about 2,000 TPS average and peaks around 8,000 TPS. Our daily data volume is roughly 15-20 GB. With those numbers, DA costs matter a lot.

Why We Started with Celestia

When we launched in late 2024, Celestia was the obvious choice:

  • Proven in production
  • Strong community and documentation
  • Public verifiability mattered to our DeFi users
  • Cost was already 64% cheaper than Ethereum blobs

For the first 8 months, it worked great. Our monthly DA costs averaged about $4,500—manageable for our treasury.

What Made Us Switch to EigenDA

Two things changed in mid-2025:

  1. Our throughput requirements grew. We integrated a perpetuals protocol that increased our data volume by 3x. Celestia’s 1.33 MB/s became a bottleneck during peak hours. We were hitting block limits and users were experiencing delays.

  2. EigenDA V2 launched. When they hit 50+ MB/s on testnet with a clear path to 100 MB/s, the throughput gap became too large to ignore.

The Migration Experience

Honestly? It was smoother than expected but not trivial.

What went well:

  • Both DA layers use similar blob submission patterns
  • Our sequencer abstraction layer made the switch modular
  • EigenDA’s documentation had specific Celestia migration guides
  • No downtime required—we ran both in parallel for 2 weeks

What was painful:

  • Different commitment schemes meant rewriting our proof verification
  • EigenDA’s operator set behaves differently than Celestia’s validator set
  • Had to update our monitoring and alerting for new failure modes
  • Some tooling (block explorers, indexers) wasn’t EigenDA-compatible yet

Total migration took about 6 weeks with 2 engineers.

Cost and Performance in Practice

After 4 months on EigenDA:

Metric Celestia EigenDA
Monthly DA cost $4,500 $3,200
Peak throughput 1.33 MB/s 45 MB/s (our usage)
Avg latency 6.2s 1.8s
Failed submissions 0.3% 0.1%

The throughput headroom is the real win. We’re nowhere near EigenDA’s limits, which means we can grow without worrying about DA becoming a bottleneck again.

What I’d Tell Other Rollup Developers

  1. Start with Celestia if you’re not sure. Public verifiability is valuable, and you can always migrate later.

  2. Know your throughput trajectory. If you’re building something that might need 10+ MB/s, factor EigenDA into your architecture from the start.

  3. Abstract your DA layer. We built a clean interface that made migration possible. If we’d tightly coupled to Celestia, this would’ve been a 6-month project.

  4. The DAC trust assumption matters less than I thought. For our DeFi use case, the economic security from restaked ETH provides enough guarantees. YMMV.

The DA wars are real, and I think we’ll see more rollups making similar migrations as throughput requirements grow.

rollup_rick

3

Good breakdown @infrastructure_ivan, but I think the framing underweights what I consider the most important factor: sovereignty and public verifiability.

Let me push back on some of the EigenDA narrative.

Why Sovereignty Matters More Than Throughput

When you choose a DA layer, you’re making a fundamental decision about who can verify your chain’s data. With Celestia’s DAS (Data Availability Sampling), anyone running a light node can verify data availability. With EigenDA’s DAC model, you’re trusting that committee members are honest.

This isn’t an abstract concern. Here’s why it matters:

  1. Censorship resistance. If EigenDA operators collude or are coerced, they can make data unavailable. With Celestia, the attack surface is the entire validator set plus anyone running DAS.

  2. Long-term data guarantees. Celestia’s economic model incentivizes permanent data availability. DACs can theoretically stop storing data after their commitment period.

  3. User trust. DeFi users with real money at stake should be able to verify DA themselves. “Trust the committee” is a regression to the pre-blockchain model.

The DAS Advantage Is Underrated

@rollup_rick mentioned Celestia’s 1.33 MB/s throughput like it’s a limitation. But here’s what that number doesn’t capture:

Celestia’s architecture allows light nodes to verify DA without downloading full blocks. A phone can verify Celestia DA. Can a phone verify EigenDA? No—you need to trust the committee attestations.

This matters because rollup users shouldn’t need to run full nodes to verify their chain. DAS is what makes truly trustless light clients possible.

Celestia’s Scaling Roadmap Is Aggressive

The throughput gap won’t last:

  • Mamo-1 testnet: 88MB blocks, 27 MB/s (already live)
  • Matcha upgrade: 128MB blocks
  • 2026 target: 1GB blocks
  • Fibre: 1 Tbps for specialized use cases

By end of 2026, Celestia’s throughput will be competitive with EigenDA—but with public verifiability intact.

When DAC Trust Assumptions Are Unacceptable

For some rollup types, EigenDA is simply not appropriate:

DeFi with significant TVL: If you’re holding hundreds of millions in user funds, “trust the 200 operators” isn’t good enough. Users deserve the ability to verify DA themselves.

Sovereign rollups: If your thesis is sovereignty and independence from Ethereum, building on EigenLayer defeats the purpose.

Regulatory-sensitive applications: DAC members can be identified and pressured. Celestia’s decentralized validator set is more censorship-resistant.

My Take on @rollup_rick’s Migration

I respect the decision, but I’d push back on “DAC trust assumption matters less than I thought.”

You’re right that so far nothing has gone wrong. But the whole point of crypto is preparing for adversarial conditions. When things go wrong—and they eventually will—the difference between “anyone can verify” and “trust the committee” will matter.

For gaming chains or low-value applications, EigenDA makes sense. For DeFi rollups holding real money, I’d argue Celestia’s trust model is worth the throughput trade-off.

The Real Question

The DA wars aren’t about who has the biggest throughput number. They’re about what kind of blockchain ecosystem we’re building.

Do we want rollups that inherit Ethereum’s security through trust minimization? Or are we comfortable reintroducing trusted intermediaries for the sake of performance?

I know where I stand.

modular_mike

4

@modular_mike raises fair points about trust assumptions, but I want to offer the perspective of someone building applications where throughput isn’t optional—it’s existential.

Our Use Case: Real-Time Gaming

I’m building an on-chain gaming platform that processes 15,000+ game state updates per second during peak hours. That translates to roughly 50-80 MB/s of data that needs to be available for verification.

Let me be direct: Celestia’s 1.33 MB/s makes our application impossible. Not suboptimal—impossible. We’d need to batch 40-60 seconds of game state into single blocks, which defeats the purpose of real-time gaming.

Why EigenDA Was the Only Option

When we evaluated DA layers in early 2025:

Requirement Celestia EigenDA Our Needs
Throughput 1.33 MB/s 100 MB/s 50-80 MB/s
Latency 6s blocks Sub-second <2s
Cost at our volume $15K/day $8K/day Budget-constrained

The math wasn’t close. EigenDA was the only option that could handle our data volume without fundamentally changing our product.

The Restaking Security Model in Practice

@modular_mike characterizes EigenDA as “trust the committee,” but that undersells the security model:

  1. 4.3 million ETH staked. That’s ~$12B+ in economic security. An attack would require corrupting operators with massive slashing risk.

  2. 200+ operators. This isn’t a 5-person multisig. The operator set is geographically distributed and includes major staking providers.

  3. Slashable commitments. Operators who fail to keep data available lose real money. The incentives are aligned.

  4. On-chain verification. While you can’t DAS-verify like Celestia, you can verify that operators have committed to data availability via Ethereum smart contracts.

Is this as trustless as DAS? No. Is it “trusting intermediaries” in the 2017 sense? Also no. It’s a spectrum, and EigenDA sits in a reasonable place on that spectrum for many use cases.

Production Metrics

After 6 months on EigenDA mainnet:

  • Uptime: 99.97% data availability
  • Average latency: 1.2 seconds from submission to confirmation
  • Peak throughput: We’ve hit 65 MB/s without issues
  • Failed submissions: 0.08% (all recovered via retry)
  • Cost: $7,800/day average, down from projected $15K on Celestia

Zero data availability failures that affected our users. Zero.

When DAC Trade-offs Are Acceptable

@modular_mike argues that “DeFi with significant TVL” should avoid EigenDA. I’d refine that:

The question isn’t “how much value” but “what happens if DA fails?”

For a lending protocol, temporary data unavailability could mean stuck liquidations and cascading losses. That’s catastrophic. Celestia’s stronger guarantees make sense.

For our gaming platform, temporary DA failure means players can’t verify recent game state for a few hours. Annoying, but not catastrophic. No funds are locked. The game continues. When DA recovers, everything reconciles.

The security model should match the threat model.

Response to “Trust the Committee”

Here’s the thing: every DA layer involves trusting someone.

  • Celestia: Trust the validator set
  • EigenDA: Trust the operator set + restaked ETH
  • Ethereum blobs: Trust Ethereum validators

The question is degree and mechanism, not presence of trust. EigenDA’s trust model is different from Celestia’s, but calling it a “regression to pre-blockchain” is hyperbolic.

Bottom Line

The DA wars will be won by use-case fit, not ideology.

  • Need maximum trustlessness? Celestia.
  • Need maximum throughput? EigenDA.
  • Need multichain flexibility? Avail.

For applications like mine where throughput is the constraint, EigenDA isn’t a compromise—it’s the only viable path. And in 6 months of production, I haven’t regretted the choice once.

perf_pete