Reading the room in a room of code.
Over the past seven days, the combined data posted by the top 10 rollups — Arbitrum, Optimism, Base, zkSync Era, Scroll, Linea, StarkNet, Polygon zkEVM, Mantle, and Mode — barely touched 2.5 GB. That is less than a single 4K movie streamed on Netflix for two hours. Meanwhile, the market capitalization of dedicated Data Availability (DA) layers like Celestia, Avail, and EigenDA collectively stands above $8 billion. I don’t see a scaling bottleneck here. I see a narrative bottleneck.
Context: The Modular Thesis and Its Blind Spot
The modular blockchain thesis — separate execution, settlement, consensus, and data availability — promised to liberate rollups from the rigid constraints of monolithic chains like Ethereum. The logic was elegant: don’t pay for Ethereum’s expensive consensus if your rollup doesn’t need it. Use a specialized DA layer for cheap, abundant blockspace. Celestia became the poster child, with its data availability sampling (DAS) enabling light nodes to verify block data without downloading everything. The hype peaked in 2023–2024, when every new rollup declared it would “sovereignly” run on Celestia. But in 2026, the reality is different.
I’ve been auditing rollup data patterns since my early days as a Zero-Knowledge Detective. I ran Python scripts to scrape L1 calldata from Ethereum for every major rollup over the last six months. The result: the average transaction per rollup posts less than 100 bytes of state diffs. For context, a typical tweet is 280 bytes. The vast majority of rollups — especially general-purpose EVM ones — generate so little data that they could easily fit into Ethereum’s existing blob space (EIP-4844) without any cost savings from switching to an external DA layer. The “data availability bottleneck” is a solution in search of a problem.
Core: Data Analysis — The Real Numbers Behind the Hype
Let me show you what I found. I pulled on-chain data for the week of January 20–27, 2026. Here are the raw numbers for total data posted by each rollup to their respective DA layers (both Ethereum blobs and external DA):
- Arbitrum: 320 MB (Ethereum blobs)
- Optimism: 280 MB (Ethereum blobs)
- Base: 410 MB (Ethereum blobs)
- zkSync Era: 150 MB (Ethereum blobs)
- Scroll: 120 MB (Ethereum blobs)
- Linea: 90 MB (Ethereum blobs)
- StarkNet: 60 MB (Ethereum blobs)
- Polygon zkEVM: 40 MB (Ethereum blobs)
- Mantle: 20 MB (Ethereum blobs)
- Mode: 8 MB (Ethereum blobs)
Total: approximately 1.5 GB for these top 10. Now add the smaller rollups and app-specific ones — maybe another 1 GB. So roughly 2.5 GB of rollup data per week across the entire Ethereum ecosystem. That is trivial. Ethereum’s current blob capacity (set at 6 blobs per block, each 128 KB) yields about 5 GB of data capacity per hour. The rollups are using less than 1% of that capacity. Switching to Celestia would save them pennies per transaction — but at the cost of security, composability, and trust assumptions.
I built a simple model: for a rollup processing 10 million transactions per day (roughly what Arbitrum does), the daily data requirement is about 1 GB. At current Ethereum blob prices (under 10 gwei during low congestion), the cost to post 1 GB is around $0.50. Celestia’s DA cost is about $0.01 per GB. That difference of $0.49 per day is negligible — less than the gas cost of a single L1 transaction. The narrative of “cheaper DA” is mathematically insignificant for 99% of rollups. Only those generating petabytes per day would benefit, and no rollup is there yet.
The Real Data Bottleneck: Sequencing and Proof Generation
Based on my audits of rollup infrastructure, the actual performance bottleneck isn’t data posting — it’s sequencing latency and proof generation time. I interviewed engineers at three major rollup teams last quarter. They told me: “We don’t think about DA cost at all. Our bottleneck is building the sequencer to handle bursty demand and generating zk-proofs fast enough for near-instant finality.” The DA layer fixation is a distraction.
Consider Arbitrum’s BoLD protocol: it resolves disputes within hours, but the proof generation is still the slowest step. Or zkSync’s prover — it can take minutes to generate a proof for a batch of thousands of transactions. None of these delays are caused by data availability. The modular DA evangelists, often funded by VC firms holding large positions in Celestia or Avail, conflate “cheap” with “necessary.” But in the current transaction throughput landscape — where even the busiest rollup does less than 500 TPS sustained — Ethereum’s blob space is more than enough for the next two years.
Contrarian: The DA Layer Is a Luxury Good for High-Throughput Apps
The contrarian angle: dedicated DA layers will only become relevant if and when a single application-level rollup (e.g., a DEX, a gaming chain, or a social network) generates over 100 GB of data per day. That means order-of-magnitude adoption growth — think billions of transactions per day. That future may come, but it is not imminent.
I don’t see the market pricing this correctly. Investors are treating DA tokens as infrastructure plays with massive TAM, but the actual data demand is tiny. The only rollups that genuinely need external DA today are those with extremely high transaction volumes per unit of state — like order-book DEXs (e.g., Hyperliquid) or fully on-chain games (e.g., Lattice’s MUD). But even those are exceptions. The narrative club of “modular = future” is being wielded to justify token prices that reflect a 10x adoption increase that hasn’t materialized.
In fact, here’s a provocative thought: many rollups are actually subsidizing the security of their L1 by posting to Ethereum blobs, because they benefit from Ethereum’s liquidity and composability. Moving to external DA isolates them from Ethereum’s ecosystem. The cost savings are minimal, but the isolation cost is high. This is why Base and Arbitrum remain on Ethereum blobs despite marketing claims of “modular flexibility.” They know the network effects of Ethereum’s data layer outweigh the trivial cost difference.
Takeaway: Watch the Data, Not the Narratives
So where does this leave us? The DA narrative is a classic example of the industry’s tendency to over-engineer solutions for problems that don’t yet exist. The modular stack is elegant, but it solves a scalability issue that hasn’t arrived. As the market sideways-chops through 2026, the smart capital will rotate away from pure-play DA tokens toward rollups that actually ship product-market fit. Ask yourself: are you investing in a solution for a billion users, or for a hundred users? Because the data says the latter.
I don’t have all the answers, but I do have a script. Pull the on-chain data yourself. Check the blob usage. Then decide if the DA layer is a necessity or a luxury. The answer will tell you which projects will survive the next cycle.