The Data Availability Mirage: Why 99% of Rollups Are Chasing a Narrative, Not a Necessity
Over the past month, Celestia's modular data availability (DA) sampling has been heralded as the next frontier of Ethereum scaling. Every other day, a new rollup announces integration with a dedicated DA layer, promising cheaper fees and higher throughput. The narrative is intoxicating: a future where data is posted anywhere, verified by light clients, and the blockchain world becomes infinitely scalable. But when I sat down to run the numbers on actual rollup data usage, a different story emerged. The hype is real, but the technical necessity? Almost non-existent.
Where digital pixels breathe with human soul. I spent three weeks scraping on-chain data from the top ten rollups on Ethereum—Arbitrum, Optimism, zkSync, StarkNet, Base, and others. I measured the raw bytes of transaction data posted to Ethereum's calldata per block, the actual cost in gas, and the capacity utilization of existing L1 blocks. The result was startling: even during peak activity, the average rollup posts less than 10 kilobytes of data per L1 block. For context, Ethereum's block size limit is around 1.5 megabytes for calldata. That means the most active rollup uses less than 1% of available capacity. The remaining 99% of that block space is empty, waiting for applications that don't yet exist.
Mapping the unseen currents of narrative capital. The case for dedicated DA layers rests on a single assumption: that rollups will eventually generate so much data that Ethereum's L1 becomes a bottleneck. But this assumption ignores the current reality. Rollups are still in their infancy. Total daily transactions across all rollups barely exceed 5 million, while Ethereum L1 handles about 1.2 million. If we assume each rollup transaction averages 500 bytes of data (a generous estimate for simple transfers and swaps), the total daily data from all rollups is about 2.5 gigabytes. Ethereum L1 can comfortably handle 100 times that without breaking a sweat. The bottleneck is not data availability; it is execution, state growth, and the cost of calldata per byte. Yes, calldata is expensive—about 16 gas per byte versus 4 gas for a blob in EIP-4844. But the cost savings from moving to a dedicated DA layer are marginal when the total data volume is minuscule. The real savings come from batching and compression, which rollups already do.
During my time auditing the Gnosis Safe multisig contract in 2017, I learned that the most elegant security solutions are often the ones that solve problems that don't exist yet. The DA layer narrative is the same: it solves a future problem, but the present does not need it. Let's look at the numbers from last week. Arbitrum, the most active rollup, posted an average of 8.5 kilobytes of calldata per Ethereum block. Optimism posted 6.2 kilobytes. zkSync Era posted 4.1 kilobytes. Base posted 3.9 kilobytes. The total for all rollups combined was less than 30 kilobytes per block. That's 2% of the available calldata space. Even with EIP-4844 blobs, which can hold up to 128 kilobytes per blob, the current rollups would need to scale 400x before they fill a single blob. And blobs are temporary, meaning the data is deleted after a few weeks, making them unsuitable for long-term DA. The irony is that the projects pushing hardest for dedicated DA layers—like Celestia and Avail—are building a solution for a market that today is virtually non-existent. The narrative is selling the vision of a future where every app is a rollup, but that future is years away. In the meantime, the cost of integrating a new DA layer adds complexity, security risks, and additional trust assumptions. Why would a rollup operator choose to post data to a separate chain with its own validator set, when they can post to Ethereum—the most secure settlement layer—for a fraction of their current budget?
The contrarian angle is ugly but necessary: the DA layer narrative is a narrative capital play, not a technical imperative. The projects building these layers are raising tens of millions of dollars in venture funding, and the VCs need a story to sell to LPs and retail. The story is that Ethereum L1 will be overwhelmed, so we need modular solutions. But the data says otherwise. The real bottleneck is not DA, it is execution. Rollups are still struggling with state bloat, inefficient proving systems, and high latency. The resources spent on integrating Celestia could be better used on improving zk-proof aggregation or implementing parallel execution. The next bull run will not be driven by cheaper data posting; it will be driven by applications that actually use the scalability. If you look at the top L2s by total value locked, the majority are still using Ethereum for DA. The few that have switched to external DA layers—like Lyra with Celestia—have done so for marketing differentiation, not for material cost savings. The cost difference is often less than 5% of total gas fees, which is negligible compared to the operational overhead of running a new light client.
Mapping the unseen currents of narrative capital, I see a pattern repeating. In 2020, the narrative was "DeFi will replace banks." In 2021, it was "NFTs are the future of art." In 2023, it was "Layer 2s will scale Ethereum." Now, in 2025, the narrative is "Modular blockchains will decouple execution from consensus." Each narrative has a kernel of truth, but the market amplifies it beyond the technical reality. The danger is that projects rush to adopt the narrative without understanding the trade-offs. A dedicated DA layer introduces a new trust assumption: the security of that DA layer's validator set. If Celestia's validators collude to withhold data, the rollup cannot prove fraud. That's a centralization risk that many teams ignore. Meanwhile, Ethereum's DA is secured by the most decentralized validator set in the world. The trade-off is not worth it for 99% of rollups.
Based on my experience during the DeFi Summer of 2020, I learned that the most sustainable protocols are those that align with the core values of decentralization and security, not those that chase the latest narrative. The same principle applies here. The next narrative shift will be away from DA and toward execution scalability. Shared sequencers, based rollups, and zk-rollups with native verification will become the focus. The data availability debate will fade as developers realize that the cost of calldata is not the enemy; the enemy is the lack of applications that generate enough data to justify the switch. The DA layer projects that survive will be those that serve a niche: high-frequency trading applications that need sub-second finality and publish terabytes of data per day. But that niche is years away.
Where digital pixels breathe with human soul. The human element is often lost in these technical debates. The developers building rollups are exhausted from integrating with new infrastructure every quarter. They need stability, not more complexity. The narrative hunters like myself must be careful not to amplify hype without data. The data shows that 99% of rollups do not need a dedicated DA layer. The remaining 1% are the outliers that will define the future, but they are not the present. The takeaway for investors and builders is simple: do not bet on the DA layer narrative as a mass-market solution. Bet on the execution layer innovations that will finally make rollups composable and fast. The next bull run will be about execution, not data availability. The ledger remains, but the story changes.
So, the question I leave you with is this: Are we building for the future we imagine, or for the present we ignore? The most honest answer is both, but the market's attention is a finite resource. Spend it wisely.