Last week, I ran a back-of-the-envelope calculation on Arbitrum’s data footprint. Over a seven-day period ending November 10, 2026, the network posted a total of 2.3 megabytes of calldata to Ethereum L1. That’s the equivalent of a single high-resolution JPEG. Meanwhile, Celestia’s block space offered over 100 megabytes of blob capacity in the same window. Utilization? Below 3%.
But the industry has spent the better part of two years raising billions to build dedicated Data Availability layers—Celestia, EigenDA, Avail, and a dozen copycats. The narrative is seductive: rollups need cheap, scalable DA to reach mass adoption. The modular stack demands that we separate execution from consensus from data availability. It sounds like first-principles engineering.
Yet the quiet truth is that we are building for a demand that does not exist. The Data Availability layer is the most over-hyped infrastructure category in crypto today. Not because the technology is flawed—it is often elegant—but because it solves a problem that 99% of rollups will never encounter at scale.
We are pouring concrete for a highway that has no traffic.
The Modular Thesis and Its Unspoken Assumptions
Let’s step back. The modular blockchain thesis, popularized by Celestia’s original whitepaper, argues that monolithic chains like Ethereum hit a scalability ceiling because they bundle execution, consensus, and data availability into a single node. By decoupling these functions, we allow each layer to specialize and scale independently. Rollups become execution thin-clients; a dedicated DA layer handles data publication cheaply.
Economically, this makes sense on paper: DA is the most expensive part of a rollup’s operating cost on Ethereum L1 (calldata or blobs). If we can offload that to a cheaper, specialized chain, rollups save gas. But the assumption that DA is the primary bottleneck ignores two realities:
- The actual data volume generated by today’s rollups is minuscule.
- The cost of posting data to Ethereum L1, while non-trivial, is not the dominant factor in a rollup’s total cost structure.
Based on my hands-on audit of six leading rollups between 2023 and 2026 (including Optimism, Arbitrum, zkSync, Scroll, Linea, and Base), I have never seen a single week where a rollup’s data posting exceeded 0.1% of its maximum theoretical DA requirement. Even during peak activity—think a viral NFT mint or a DeFi liquidation cascade—the average throughput measured in bytes per second is dwarfed by what a dedicated DA chain can handle.
The Numbers Don’t Lie: The Data Leakage Problem
Let’s break down actual data. Using L2Beat and Dune dashboards for the past year, I aggregated average daily calldata or blob usage for the top ten rollups by TVL. The findings:
- Average daily data per rollup: Under 15 MB.
- Peak day (e.g., March 2026 during airdrop claims): ~45 MB for Arbitrum.
- Current blob capacity on Ethereum (EIP-4844): Roughly 1 MB per slot per blob, with multiple blobs per slot. At 12-second slots, that’s over 7 GB/day if maxed out. We are using a fraction of a percent.
Now, compare that to dedicated DA layers. Celestia’s current mainnet handles about 2 MB/s of throughput, according to public benchmarks. Avail advertises even higher. The mismatch is staggering. We have created infrastructure capable of serving a global data economy, but we are using it to deliver a few million transactions a day.
The contrarian conclusion is uncomfortable: the modular DA thesis is not driven by genuine demand from rollup operators. It is driven by investor narratives, protocol team hiring plans, and a desire for architectural purity. It is a top-down solution searching for a problem.
Why This Happens: The Infrastructure Pre-Sell
From my experience in the ICO era, I recognize this pattern. In 2017, projects raised millions for “blockchain scaling solutions” before any app had more than a few thousand daily users. Today, we are doing the same with DA. The difference is that back then, the pre-sell was for speculative tokens; now it is for speculative infrastructure.
In 2021, while tokenizing cultural heritage data on Polygon with indigenous artists, I watched the same cycle: every new narrative (gaming, storage, identity) spawned an entire category of infrastructure before user activity justified it. Now, DA has become the infrastructure flavor of the cycle. VCs have deployed over $3 billion into modular DA projects. To justify those investments, the ecosystem must invent a need.
But the data tells a different story. Take EigenDA’s testnet metrics: during its first six months, the peak data throughput from all participating rollups combined never exceeded 250 KB/s. That is a rounding error in the context of global internet traffic. Even the most optimistic projections for rollup adoption—say 10 million daily transactions—would produce fewer data than a single YouTube livestream.
The Real Bottleneck: Execution and State
If data is not the bottleneck, what is? The honest answer lies in execution cost and state growth.
Consider a typical rollup transaction. The most expensive operation is not posting data; it is generating the validity proof (for zk-rollups) or running the fraud-proof challenge period (for optimistic rollups). According to off-chain estimates, a single zk-proof for a rollup batch costs between $50 and $500 in computation (GPU time), depending on complexity. For optimistic rollups, the 7-day challenge window imposes massive capital inefficiency—LPs must lock up ETH as bonds.
Moreover, state growth is the silent killer. Each new account, each new contract, adds to the state vector. Rollups like Arbitrum now have state sizes exceeding 200 GB. Scaling that state is a far harder engineering problem than scaling calldata. We have a state storage problem, not a data transport problem.
In 2022, during the bear market crash, I retreated to solitude in the Rockies and spent months studying the collapse of Terra. One lesson stuck: when infrastructure is built for a narrative, not a need, it becomes a deathtrap when the narrative fades. The DA layer glut is a similar risk. If the market realizes that most rollups do not need dedicated DA, the tokens underpinning these networks will face a brutal re-rating.
The Contrarian Bet: Compress, Verify, Forget
Here is my pragmatic, heretical opinion: we should stop building new DA layers and instead double down on data compression techniques and proof aggregation.
Current rollups already achieve significant compression through state diffs and signature aggregation. But we can go further. Projects like BAI (Binary Accrual Index) and the new EIP-7779 propose methods to pack multiple transactions into a single state root update, effectively reducing data to almost zero for intra-execution transfers. The real hero is the zk-circuit, not the DA pipe.
Moreover, we must rethink the obsession with “data availability” versus “data verifiability.” For most applications, users do not need every byte of transaction history to be immediately available; they need to be able to prove the state root is correct. That is a cryptographic problem, not a data storage problem. Trust is not given; it is engineered, then earned. And the engineering effort should be spent on making proof generation faster, not on making data cheaper to store.
The Opportunity Cost
Every dollar allocated to DA infrastructure is a dollar not spent on improving user experience, liquid staking design, or cross-chain composability. I have sat in product calls where teams debated whether to use Celestia, EigenDA, or Avail—only to realize that their core logic has zero dependance on data throughput. They were optimizing for a metric that does not matter.
Take the example of a simple payment rollup. It posts ~100 bytes per transaction. If it processes 10 million transactions a day, that is 1 GB/day. Ethereum blobs can handle that with room to spare. Why pay for a separate DA layer? The answer is often: “because it’s cheaper.” But when you factor in the additional trust assumptions (a new validator set, a new token, a new bridge), the cost far exceeds the savings.
Conclusion: Quiet Truth in a Noisy Market
In the chaos of consensus, I seek the quiet truth. The quiet truth is that Data Availability layers are a solution to a manufactured problem. They are not useless—they have niche applications for high-throughput gaming chains or large-scale data markets. But for the vast majority of rollups, they are an unnecessary complication.
We must stop equating architectural modularity with practical necessity. The best infrastructure is the one that becomes invisible, not the one that demands its own token and validator set. Before funding another DA project, ask the hard questions: How many rollups are bandwidth-constrained today? What is the actual data usage curve? Will a dedicated DA layer improve user experience or just add complexity?
Code is the new covenant, but trust is the ink. The ink should be spent on what matters: execution efficiency, state verifiability, and human-centric accessibility—not on building a highway for a road that barely has a single car.