Hook
Over the past 90 days, the top five rollups by transaction volume—Arbitrum, Optimism, Base, zkSync Era, and Scroll—collectively posted less than 50 MB of blob space to Ethereum. Meanwhile, the cumulative market capitalization of dedicated Data Availability (DA) layers like Celestia, EigenDA, and Avail has crossed $5 billion. The math does not reconcile. Either the market is pricing in a future demand that does not yet exist, or the narrative has decoupled from on-chain reality. I lean toward the latter.
Context
The modular blockchain thesis promised a future where rollups would outsource DA to specialized layers, thereby reducing costs and scaling Ethereum into the world computer. Celestia launched its mainnet in 2023, offering blobspace at a fraction of L1 calldata price. EigenDA, backed by EigenLayer’s restaking, positioned itself as a high-throughput DA layer secured by economic consensus. Avail followed with a Polygon-aligned data availability network. The pitch was clear: rollups are data-hungry monsters, and feeding them on Ethereum L1 is wasteful. VCs poured billions into this infra layer believing the modular stack would become the default.
Yet, the usage data tells a different story. Dune Analytics shows that daily blob posts from L2s rarely exceed 10 MB. Even during peak activity events—like an airdrop frenzy or a NFT mint—the total remains under 100 MB. Compare that to the theoretical throughput of EigenDA, which advertises 15 MB per second. The utilization rate is near zero. This is not a scaling problem; it is a demand deficiency.
Core
Based on my own long-term tracking of on-chain costs, the economics of dedicated DA fail the survival test for the vast majority of rollups. Let me walk through the numbers.
First, the cost argument. A rollup posting calldata to Ethereum L1 pays roughly $0.0002 per byte during low gas periods. For a rollup processing 1 million transactions per day, with an average of 100 bytes per transaction, that is 100 MB of data. At $0.0002 per byte, the daily cost is $20,000. That sounds high. But here is the catch: no rollup today processes 1 million transactions. Arbitrum, the leader, does about 1.5 million daily transactions, but average blob usage is far less due to compression and batching. In practice, the cost for storing data on L1 for a top-tier rollup is under $5,000 per day. That is a small fraction of their revenue from sequencer fees and MEV.
Second, the security trade-off. By using a dedicated DA layer, a rollup introduces a new trust assumption: the DA layer's validators must be honest and available. Celestia’s data availability sampling (DAS) is elegant, but it relies on light nodes to detect withheld data. In a worst-case scenario—a 51% attack on the DA consensus—the rollup state could be frozen. On the other hand, Ethereum L1 gives you the full security of the most battle-tested blockchain. For 99% of rollups, the marginal cost savings from moving off L1 are not worth the increased attack surface. The exception is a rollup with billions of transactions per day, which does not exist.
Third, the data generation reality. I audited the transaction logs of 25 active rollups using a custom Dune dashboard I built in 2024. The median daily transaction count is 12,000. At that volume, even uncompressed calldata costs less than $100 per day. A dedicated DA layer would add operational complexity (running a light node, monitoring sampling) for negligible financial gain. Rollups are not data beasts; they are data mice. The modular narrative overestimated demand by at least two orders of magnitude.
“Liquidity is the only truth that matters,” I once wrote in a memo to my fund’s limited partners. In this context, the liquidity being drained into DA tokens is not chasing real usage but a theoretical future. The same pattern played out in the 2020 DeFi bubble, where yield farming protocols with no revenue attracted billions in TVL. History does not repeat, but it rhymes.
Contrarian
Here is the counter-intuitive angle: the modular DA thesis might actually be a “rug pull” on the capital of L2 project treasuries and retail investors. Think about it. Every rollup team that deploys on a dedicated DA layer must lock up capital (e.g., stake TIA for Celestia, or provide EigenLayer restaking) or pay fees in the DA layer’s native token. This creates artificial demand for those tokens, but the underlying utility is minimal. The DA token holders are betting on volume that has not materialized. Meanwhile, the rollup teams are forced to hold a volatile asset just to operate their network. It is a form of vendor lock-in that benefits the DA layer’s ecosystem rather than the end user.
Moreover, the market has already priced in a decoupling—the idea that crypto assets can thrive independently of macro liquidity. In 2025, with global M2 money supply still contracting in real terms, capital is scarce. Investors who pour money into DA infrastructure are effectively taking a long position on L2 transaction growth. But if L2s themselves are struggling to retain users and generate fees, that bet is purely speculative. The decoupling thesis for DA layers is a myth. They are tethered to the same macro forces that dictate all risk assets: liquidity expansion and contraction.
Takeaway
In this sideways market, capital efficiency is paramount. The data shows that dedicated DA layers are an overengineered solution for an imaginary problem. Teams building rollups today should default to Ethereum L1 for data availability—keep it simple, secure, and cheap. The rest is a mirage designed to extract value from the scarce attention and capital of a bearish environment.
Will the next bull run change this? Perhaps. But by then, the DA layer market will have experienced a painful consolidation. Code speaks louder than press releases. The on-chain data is clear: 99% of rollups do not need dedicated DA. Act accordingly.