Skip to main content

How Celestia's Data Availability Sampling Hits 1 Terabit Per Second: The Technical Deep Dive

· 13 min read
Dora Noda
Dora Noda
Software Engineer

On January 13, 2026, Celestia shattered expectations with a single benchmark: 1 terabit per second of data throughput across 498 distributed nodes. For context, that's enough bandwidth to process the entire daily transaction volume of Ethereum's largest Layer 2 rollups—in less than a second.

But the real story isn't the headline number. It's the cryptographic infrastructure that makes it possible: Data Availability Sampling (DAS), a breakthrough that allows resource-constrained light nodes to verify blockchain data availability without downloading entire blocks. As rollups race to scale beyond Ethereum's native blob storage, understanding how Celestia achieves this throughput—and why it matters for rollup economics—has never been more critical.

The Data Availability Bottleneck: Why Rollups Need a Better Solution​

Blockchain scalability has long been constrained by a fundamental trade-off: how do you verify that transaction data is actually available without requiring every node to download and store everything? This is the data availability problem, and it's the primary bottleneck for rollup scaling.

Ethereum's approach—requiring every full node to download complete blocks—creates an accessibility barrier. As block sizes grow, fewer participants can afford the bandwidth and storage to run full nodes, threatening decentralization. Rollups posting data to Ethereum L1 face prohibitive costs: at peak demand, a single batch can cost thousands of dollars in gas fees.

Enter modular data availability layers. By separating data availability from execution and consensus, protocols like Celestia, EigenDA, and Avail promise to slash rollup costs while maintaining security guarantees. Celestia's innovation? A sampling technique that inverts the verification model: instead of downloading everything to verify availability, light nodes randomly sample tiny fragments and achieve statistical confidence that the full dataset exists.

Data Availability Sampling Explained: How Light Nodes Verify Without Downloading​

At its core, DAS is a probabilistic verification mechanism. Here's how it works:

Random Sampling and Confidence Building​

Light nodes don't download entire blocks. Instead, they conduct multiple rounds of random sampling for small portions of block data. Each successful sample increases confidence that the complete block is available.

The math is elegant: if a malicious validator withholds even a small percentage of block data, honest light nodes will detect the unavailability with high probability after just a few sampling rounds. This creates a security model where even resource-limited devices can participate in data availability verification.

Specifically, every light node randomly chooses a set of unique coordinates in an extended data matrix and queries bridge nodes for the corresponding data shares plus Merkle proofs. If the light node receives valid responses for each query, statistical probability guarantees the whole block's data is available.

2D Reed-Solomon Encoding: The Mathematical Foundation​

Celestia employs a 2-dimensional Reed-Solomon encoding scheme to make sampling both efficient and fraud-resistant. Here's the technical flow:

  1. Block data is split into k × k chunks, forming a data square
  2. Reed-Solomon erasure coding extends this to a 2k × 2k matrix (adding redundancy)
  3. Merkle roots are computed for each row and column of the extended matrix
  4. The Merkle root of these roots becomes the block data commitment in the block header

This approach has a critical property: if any portion of the extended matrix is missing, the encoding breaks down, and light nodes will detect inconsistencies when verifying Merkle proofs. An attacker can't withhold data selectively without being caught.

Namespaced Merkle Trees: Rollup-Specific Data Isolation​

Here's where Celestia's architecture shines for multi-rollup environments: Namespaced Merkle Trees (NMTs).

A standard Merkle tree groups data arbitrarily. An NMT, however, tags every node with the minimum and maximum namespace identifiers of its children, and orders leaves by namespace. This enables rollups to:

  • Download only their own data from the DA layer
  • Prove completeness of their namespace's data with a Merkle proof
  • Ignore irrelevant data from other rollups entirely

For a rollup operator, this means you're not paying bandwidth costs to download data from competing chains. You fetch exactly what you need, verify it with cryptographic proofs, and move on. This is a massive efficiency gain compared to monolithic chains where all participants must process all data.

The Matcha Upgrade: Scaling to 128MB Blocks​

In 2025, Celestia activated the Matcha upgrade, a watershed moment for modular data availability. Here's what changed:

Block Size Expansion​

Matcha increases maximum block size from 8MB to 128MB—a 16x capacity boost. This translates to:

  • Data square size: 128 → 512
  • Maximum transaction size: 2MB → 8MB
  • Sustained throughput: 21.33 MB/s in testnet (April 2025)

To put this in perspective, Ethereum's target blob count is 6 per block (roughly 0.75 MB), expandable to 9 blobs. Celestia's 128MB blocks dwarf this capacity by over 100x.

High-Throughput Block Propagation​

The constraint wasn't just block size—it was block propagation speed. Matcha introduces a new propagation mechanism (CIP-38) that safely disseminates 128MB blocks across the network without causing validator desynchronization.

In testnet, the network sustained 6-second block times with 128MB blocks, achieving 21.33 MB/s throughput. This represents 16x the current mainnet capacity.

Storage Cost Reduction​

One of the most overlooked economic changes: Matcha reduced the minimum data pruning window from 30 days to 7 days + 1 hour (CIP-34).

For bridge nodes, this slashes storage requirements from 30TB to 7TB at projected throughput levels. Lower operational costs for infrastructure providers translate to cheaper data availability for rollups.

Token Economics Overhaul​

Matcha also improved TIA token economics:

  • Inflation cut: From 5% to 2.5% annually
  • Validator commission increase: Max raised from 10% to 20%
  • Improved collateral properties: Making TIA more suitable for DeFi use cases

Combined, these changes position Celestia for the next phase: scaling toward 1 GB/s throughput and beyond.

Rollup Economics: Why 50% DA Market Share Matters​

As of early 2026, Celestia holds approximately 50% of the data availability market, having processed over 160 GB of rollup data. This dominance reflects real-world adoption by rollup developers who prioritize cost and scalability.

Cost Comparison: Celestia vs Ethereum Blobs​

Celestia's fee model is straightforward: rollups pay per blob based on size and current gas prices. Unlike execution layers where computation dominates, data availability is fundamentally about bandwidth and storage—resources that scale more predictably with hardware improvements.

For rollup operators, the math is compelling:

  • Ethereum L1 posting: At peak demand, batch submission can cost $1,000–$10,000+ in gas
  • Celestia DA: Sub-dollar costs per batch for equivalent data

This 100x+ cost reduction is why rollups are migrating to modular DA solutions. Cheaper data availability directly translates to lower transaction fees for end users.

The Rollup Incentive Structure​

Celestia's economic model aligns incentives:

  1. Rollups pay for blob storage proportional to data size
  2. Validators earn fees for securing the DA layer
  3. Bridge nodes serve data to light nodes and earn service fees
  4. Light nodes sample data for free, contributing to security

This creates a flywheel: as more rollups adopt Celestia, validator revenue increases, attracting more stakers, which strengthens security, which attracts more rollups.

The Competition: EigenDA, Avail, and Ethereum Blobs​

Celestia's 50% market share is under siege. Three major competitors are scaling aggressively:

EigenDA: Ethereum-Native Restaking​

EigenDA leverages EigenLayer's restaking infrastructure to offer high-throughput data availability for Ethereum rollups. Key advantages:

  • Economic security: Secured by restaked ETH (currently 93.9% of restaking market)
  • Tight Ethereum integration: Native compatibility with Ethereum's blob market
  • Highest throughput claims: Though previous versions lacked active economic security

Critics point out that EigenDA's reliance on restaking introduces cascade risk: if an AVS experiences slashing, it could propagate to Lido stETH holders and destabilize the broader LST market.

Avail: Universal DA for All Chains​

Unlike Celestia's Cosmos focus and EigenDA's Ethereum orientation, Avail positions itself as a universal DA layer compatible with any blockchain architecture:

  • UTXO, Account, and Object model support: Works with Bitcoin L2s, EVM chains, and Move-based systems
  • Modular design: Separates DA from consensus entirely
  • Cross-ecosystem vision: Aims to serve as the neutral DA layer for all blockchains

Avail's challenge? It's the newest entrant, lagging in live rollup integrations compared to Celestia and EigenDA.

Ethereum Native Blobs: EIP-4844 and Beyond​

Ethereum's EIP-4844 (Dencun upgrade) introduced blob-carrying transactions, offering rollups a cheaper data posting alternative to calldata. Current capacity:

  • Target: 6 blobs per block (~0.75 MB)
  • Maximum: 9 blobs per block (~1.125 MB)
  • Future expansion: PeerDAS and zkEVM upgrades targeting 10,000+ TPS

However, Ethereum blobs come with trade-offs:

  • Short retention window: Data is pruned after ~18 days
  • Shared resource contention: All rollups compete for the same blob space
  • Limited scalability: Even with PeerDAS, blob capacity maxes out far below Celestia's roadmap

For rollups prioritizing Ethereum alignment, blobs are attractive. For those needing massive throughput and long-term data retention, Celestia remains the better fit.

Fibre Blockspace: The 1 Terabit Vision​

On January 14, 2026, Celestia co-founder Mustafa Al-Bassam unveiled Fibre Blockspace—a new protocol targeting 1 terabit per second of throughput with millisecond latency. This represents a 1,500x improvement over the original roadmap targets from just a year prior.

Benchmark Details​

The team achieved the 1 Tbps benchmark using:

  • 498 nodes distributed across North America
  • GCP instances with 48-64 vCPUs and 90-128GB RAM each
  • 34-45 Gbps network links per instance

Under these controlled conditions, the protocol sustained 1 terabit per second data throughput—a staggering leap in blockchain performance.

ZODA Encoding: 881x Faster Than KZG​

At Fibre's core is ZODA, a novel encoding protocol that Celestia claims processes data 881x faster than KZG commitment-based alternatives used by EigenDA and Ethereum blobs.

KZG commitments (Kate-Zaverucha-Goldberg polynomial commitments) are cryptographically elegant but computationally expensive. ZODA trades some cryptographic properties for massive speed gains, making terabit-scale throughput achievable on commodity hardware.

The Vision: Every Market Comes Onchain​

Al-Bassam's roadmap statement captures Celestia's ambition:

"If 10KB/s enabled AMMs, and 10MB/s enabled onchain orderbooks, then 1 Tbps is the leap that enables every market to come onchain."

The implication: with sufficient data availability bandwidth, financial markets currently dominated by centralized exchanges—spot, derivatives, options, prediction markets—could migrate to transparent, permissionless blockchain infrastructure.

Reality Check: Benchmarks vs. Production​

Benchmark conditions rarely match real-world chaos. The 1 Tbps result was achieved in a controlled testnet environment with high-performance cloud instances. The real test comes when:

  • Actual rollups push production workloads
  • Network conditions vary (latency spikes, packet loss, asymmetric bandwidth)
  • Adversarial validators attempt data withholding attacks

Celestia's team acknowledges this: Fibre runs parallel to the existing L1 DA layer, giving users a choice between battle-tested infrastructure and cutting-edge experimental throughput.

What This Means for Rollup Developers​

If you're building a rollup, Celestia's DAS architecture offers compelling advantages:

When to Choose Celestia​

  • High-throughput applications: Gaming, social networks, micropayments
  • Cost-sensitive use cases: Rollups targeting sub-cent transaction fees
  • Data-intensive workflows: AI inference, decentralized storage integrations
  • Multi-rollup ecosystems: Projects launching multiple specialized rollups

When to Stick with Ethereum Blobs​

  • Ethereum alignment: If your rollup values Ethereum's social consensus and security
  • Simplified architecture: Blobs offer tighter integration with Ethereum tooling
  • Lower complexity: Less infrastructure to manage (no separate DA layer)

Integration Considerations​

Celestia's DA layer integrates with major rollup frameworks:

  • Polygon CDK: Easily pluggable DA component
  • OP Stack: Custom DA adapters available
  • Arbitrum Orbit: Community-built integrations
  • Rollkit: Native Celestia support

For developers, adopting Celestia often means swapping out the data availability module in your rollup stack—minimal changes to execution or settlement logic.

The Data Availability Wars: What Comes Next​

The modular blockchain thesis is being stress-tested in real time. Celestia's 50% market share, EigenDA's restaking momentum, and Avail's universal positioning set up a three-way competition for rollup mindshare.

  1. Throughput escalation: Celestia targets 1 GB/s → 1 Tbps; EigenDA and Avail will respond
  2. Economic security models: Will restaking risks catch up to EigenDA? Can Celestia's validator set scale?
  3. Ethereum blob expansion: PeerDAS and zkEVM upgrades could shift cost dynamics
  4. Cross-chain DA: Avail's universal vision vs. ecosystem-specific solutions

The BlockEden.xyz Angle​

For infrastructure providers, supporting multiple DA layers is becoming table stakes. Rollup developers need reliable RPC access not just to Ethereum, but to Celestia, EigenDA, and Avail.

BlockEden.xyz offers high-performance RPC infrastructure for Celestia and 10+ blockchain ecosystems, enabling rollup teams to build on modular stacks without managing node infrastructure. Explore our data availability APIs to accelerate your rollup deployment.

Conclusion: Data Availability as the New Competitive Moat​

Celestia's Data Availability Sampling isn't just an incremental improvement—it's a paradigm shift in how blockchains verify state. By enabling light nodes to participate in security through probabilistic sampling, Celestia democratizes verification in a way monolithic chains cannot.

The Matcha upgrade's 128MB blocks and the Fibre vision's 1 Tbps throughput represent inflection points for rollup economics. When data availability costs drop 100x, entirely new application categories become viable: high-frequency trading onchain, real-time multiplayer gaming, AI agent coordination at scale.

But technology alone doesn't determine winners. The DA wars will be decided by three factors:

  1. Rollup adoption: Which chains actually commit to production deployments?
  2. Economic sustainability: Can these protocols maintain low costs as usage scales?
  3. Security resilience: How well do sampling-based systems resist sophisticated attacks?

Celestia's 50% market share and 160 GB of processed rollup data prove the concept works. Now the question shifts from "can modular DA scale?" to "which DA layer will dominate the rollup economy?"

For builders navigating this landscape, the advice is clear: abstract your DA layer. Design rollups to swap between Celestia, EigenDA, Ethereum blobs, and Avail without re-architecting. The data availability wars are just beginning, and the winners may not be who we expect.

Sources:

Share on Twitter