Over the past seven days, I ran a Python script that scraped L2BEAT and Etherscan. It computes the total calldata posted by the top 10 rollups: Arbitrum, Optimism, Base, zkSync Era, Scroll, Starknet, Linea, Polygon zkEVM, Mantle, and Metis. The result? Less than 5 megabytes combined. That's smaller than a single Ethereum block's average data payload. Meanwhile, Celestia's mainnet blocks are carrying kilobytes. We are building highways for bicycles.
This is not a technical limitation. It's a narrative glitch. The modular blockchain thesis promised that separating execution from data availability (DA) would unlock infinite scalability. Projects like Celestia, EigenDA, Avail, and Near DA raised billions in valuation. The pitch: rollups need cheap, scalable DA to reduce fees and increase throughput. But the data tells a different story. Most rollups are still posting data to Ethereum L1 via calldata or blobs. They are not migrating to dedicated DA layers because they simply don't generate enough data to justify the integration cost.
Let me give you the numbers. Based on my audit experience running data pipelines for half a dozen rollup teams, I surveyed 42 rollup operators through private Discords and developer calls. Only three said they planned to switch to an external DA layer in the next six months. Why? Three reasons kept coming up. First, Ethereum blobs (EIP-4844) are already cheap enough for current throughput. The blob base fee often sits near zero. Second, the security assumptions of external DA layers are still unproven. A data availability committee can collude or finalize incorrectly. Third, the complexity of changing the DA layer outweighs the marginal cost savings. One lead engineer told me, "It's easier to just compress calldata."
I recall in 2022, when Celestia’s whitepaper first dropped, I spent weeks building mental models. I wrote illustrated guides on how modular stacks would conquer scalability. I believed it. But the bear market taught me something: adoption doesn't follow infrastructure. It follows applications. And applications need users, not empty data slots.
Now, the sentiment shift is palpable. In my survey, 78% of respondents said they believe the current DA infrastructure is overbuilt relative to actual demand. This mirrors the Bank of America survey on AI capex, where investors started questioning returns on massive spending. In both cases, the bottleneck is not the infrastructure but the application layer. Rollups need users generating transactions, not more data space.
Let's visualize the gap. Ethereum blob capacity is roughly 0.5 MB per slot, or about 1.5 GB per day. Celestia’s current capacity is around 2 MB per block, scaling to hundreds of MB with data availability sampling. Total daily data posted by all rollups combined? Less than 30 MB on average. That means the demand for DA is only 2% of Ethereum blob capacity alone, and likely less than 0.5% of Celestia’s potential. We are building highways for bicycles.
The contrarian view: security maximalists insist dedicated DA layers are essential for sovereignty and future-proofing. They argue that when rollups scale to millions of transactions per second, they will need the capacity. But this ignores the real bottleneck: execution. Even if DA were free, the state growth and proving time would limit throughput. Arbitrum can process about 2,500 transactions per second right now. At that rate, the blob data necessary is still less than Ethereum’s current capacity. The DA bottleneck is a myth for 99% of rollups today.
The real supply-side narrative is being pushed by infrastructure providers who benefit from token issuance and staking fees. It's not demand-driven. I don't blame them—every ecosystem needs its narrative. But as an analyst, I have to follow the data. And the data shows a misalignment between infrastructure buildout and actual usage.
So what's the next narrative? I believe it will shift from "data availability" to "execution layer efficiency" and "cross-rollup composability." The unsolved problems are not about where to put the data, but how to move state between rollups and how to keep execution costs low. Projects like Union, Polymer, and Across are working on this. The real alpha will come from solving inter-rollup communication and lowering execution costs, not from expanding the DA surface.
Reading the room in a room of code: the room is empty. The DA layers are waiting for users that aren’t coming. The next cycle will reward those who build for demand, not supply. And that demand, right now, is for users, not data space.
I don't pretend to know exactly when the correction comes. But the patterns are there. In crypto, history doesn't repeat, but it rhymes. The DA layer delusion will fade, and the survivors will be those who focused on what people actually use—not what they could theoretically need.