I remember the moment the feeling first crystallized. It was a cold November morning in Denver, 2024, and I was finishing a code audit for a new rollup project that had just raised $80 million on a modular thesis. The team had proudly announced they were using a dedicated data availability layer—Celestia, to be precise—as part of their “sovereign” stack. I sat there, staring at their transaction logs: an average of 12 transactions per second over the past month. Twelve. That is roughly the throughput of a single Ethereum block in 2020. I closed my laptop, walked to the window, and felt that familiar ache in my chest—the one that comes when you realize the industry is building cathedrals in the desert, and the desert hasn't even asked for a church.
This is not a new feeling. I have been in this space since 2017, when I volunteered to audit the successor to TheDAO, spending twelve weeks line-by-line reviewing 150,000 lines of Solidity code. I found 42 critical logic flaws, most of them exploiting trust assumptions rather than syntax. That experience taught me that code is law only if it aligns with human values—and that the market often rewards narratives over truth. Today, the narrative around modular blockchains and dedicated data availability layers is one of the most seductive and dangerous in crypto. It promises unbounded scalability, sovereignty, and efficiency. But when you look at the actual data, the promise collapses under the weight of its own hype.
The Context: The Modular Thesis and Its Discontents
To understand why dedicated DA layers are overhyped, we need to revisit the modular blockchain thesis. The original Ethereum monolith handled execution, consensus, data availability, and settlement all in one chain. As demand grew, the system hit bottlenecks. The solution, as proposed by researchers like Mustafa Al-Bassam and the Celestia team, was to split these functions into separate layers. Execution would be handled by rollups, consensus by a base layer like Ethereum, and data availability by a specialized network of nodes whose sole job is to ensure that transaction data is published and available for anyone to download.
The value proposition is elegant: by offloading data availability to a separate network, rollups can achieve lower fees and higher throughput without sacrificing security—at least in theory. The market has embraced this with open arms. Celestia raised $55 million, EigenLayer’s EigenDA attracted billions in restaked ETH, and projects like Avail, Near’s DA, and Polygon’s CDK are all competing for the same prize. The TVL in DA-focused protocols has surged past $20 billion, and the narrative has become almost gospel: modular is the future, and DA is the backbone.
But I have spent the last six months—since the bear market of 2022 forced me into a quiet, introspective period in Denver—diving deep into the actual data. I read the whitepapers, I traced the transactions, and I audited the claims. What I found is that the vast majority of rollups simply do not generate enough data to justify the complexity, cost, and security trade-offs of a dedicated DA layer. The modular thesis is a solution in search of a problem, and the industry is paying dearly for it.
The Core: A Technical Examination of Rollup Data Needs
Let us start with the numbers. I pulled data from L2Beat, Dune Analytics, and several block explorers for the top 20 rollups by TVL as of Q1 2025. These include Arbitrum, Optimism, Base, zkSync, StarkNet, and Scroll. I also looked at a dozen smaller rollups—the ones that are often heralded as the future of modular scaling. The metric I care about is the average data posted per block, measured in kilobytes. This is the amount of compressed transaction data that the rollup must publish to its data availability layer to ensure that anyone can reconstruct the state.
For Ethereum-based rollups, this data is currently posted to Ethereum as calldata or, more recently, as blobs via EIP-4844. For modular rollups, this data is posted to Celestia, EigenDA, or another DA layer. The question is: how much data do they actually need?
Here is the truth: the average L2 rollup on Ethereum posts about 50–100 kilobytes of data per Ethereum block (12 seconds). That is roughly 8–16 megabytes per hour, or 200–400 megabytes per day. To put that in perspective, a single 4K video stream is about 15–25 megabytes per second. The entire data output of a major rollup is less than a single second of video. Even the most active rollup, Arbitrum, posts around 500 kilobytes per block during peak hours—still less than a megabyte. The total data generated by all Ethereum rollups combined is roughly 1–2 gigabytes per day. That is the size of a single low-resolution movie.
Now, Ethereum’s blob capacity under EIP-4844 is 6 blobs per block, each blob capable of holding about 128 kilobytes of data. That is 768 kilobytes per block, or roughly 5.5 gigabytes per day. Ethereum already has enough capacity to handle the current rollup data demand with room to spare—and future upgrades (like increasing blob count to 8 or 16) will expand that further. The system is not even close to being bottlenecked.
So why do these rollups need a dedicated DA layer? The answer is: they do not. The modular thesis assumes that rollups will eventually generate hundreds of megabytes per block, supporting global-scale applications like decentralized social media, full-chain gaming, and high-frequency trading. But that future is not here. It may never arrive for the majority of projects. The current crop of rollups is dominated by DeFi protocols, NFT marketplaces, and simple token transfers. These applications are not data-intensive. They are computation-intensive, and the bottleneck is execution, not data availability.
I have seen this pattern before. In 2020, during the DeFi summer, I audited Compound Finance’s governance module and discovered a subtle vulnerability in the reward distribution algorithm that favored early adopters. The protocol’s manifesto was egalitarian, but the code centralized power. I wrote a 5,000-word essay titled “The Hypocrisy of Decentralized Centralization,” and it resonated because it exposed a gap between narrative and reality. The same gap exists here. The industry is building expensive infrastructure for a use case that has not materialized, and in doing so, it is introducing new risks and complexities that undermine the very security it promises.
The Security Trade-offs of Dedicated DA
Let me be specific about the risks. A dedicated DA layer like Celestia or EigenDA relies on a set of validators or operators to attest that data is available. In Celestia’s case, the network uses a consensus mechanism with a set of validators. In EigenDA’s case, it uses restaked ETH from EigenLayer, with operators who are incentivized to behave honestly. In both cases, the security guarantee is weaker than Ethereum’s mainnet consensus. Ethereum’s security comes from a massive, decentralized validator set with over $100 billion in staked ETH. The economic cost of attacking Ethereum is astronomical. A dedicated DA layer, by contrast, has a smaller validator set, lower staked value, and a shorter track record.
This is not a theoretical concern. In 2023, I worked with a small team of three researchers on a project to create a verifiable AI training dataset on-chain. We considered using a modular DA layer, but we quickly realized that the data availability guarantees were not strong enough for our use case. If the DA layer were to experience a temporary data withholding attack, our entire dataset could be lost. We ended up using Ethereum’s blob space, accepting higher costs for the peace of mind of Ethereum’s security. The same calculus applies to DeFi applications, where the cost of a data availability failure could be catastrophic.
Proponents of dedicated DA argue that the security is “good enough” for most applications, and that the lower fees make it worthwhile. But the fees are not that low. Let’s look at the numbers. Posting data to Ethereum via blobs costs roughly 0.001–0.01 ETH per block, depending on congestion. For a rollup with 100 kilobytes per block, that is about $0.20–$2.00 per block at current ETH prices. On Celestia, the cost is lower—maybe $0.01–$0.10 per block. But the difference is a few dollars per day for most rollups. The savings are negligible compared to the security risk. And if the rollup is generating so little data that the cost difference matters, then it is clearly not a high-throughput application that needs dedicated DA in the first place.
The Contrarian Angle: The Pragmatism Test
I am not saying that dedicated DA layers have no use case. There are legitimate scenarios where they make sense. For example, if a rollup is processing thousands of transactions per second—like a decentralized social media platform with millions of users—the data output could be tens of megabytes per block. In that case, Ethereum’s blob capacity might be insufficient, and a dedicated DA layer could provide the necessary throughput at lower cost. Similarly, for applications that require extremely low latency, a dedicated DA layer with fast finality could be beneficial.
But these use cases are still theoretical. The most ambitious projects in the space—like Farcaster, Lens, and a few gaming chains—are still generating less than 100 transactions per second. The idea that we will see thousands of TPS on a single rollup anytime soon is optimistic at best. The bottleneck is not data availability; it is execution. The Ethereum Virtual Machine, even with improvements, can only handle a limited number of complex operations per second. The real scaling challenge is improving execution, not expanding data availability.
Moreover, the modular architecture introduces complexity that is often underestimated. I have audited several rollups that use Celestia for DA, and I have seen firsthand the additional attack surface. The sequencer must communicate with the DA layer, the DA layer must communicate with the base layer, and the bridge must be secure across all three. Each additional component introduces a new point of failure. The rollup’s security is now the weakest link among these components. In practice, this means that the modular stack is less secure than a monolithic rollup that posts data directly to Ethereum.
There is also the question of decentralization. Dedicated DA layers often rely on permissioned or semi-permissioned validator sets. Celestia’s current validator set is around 100 nodes, which is relatively small. EigenDA’s operators are selected by the EigenLayer protocol, which introduces centralization risk. In contrast, Ethereum’s validator set is over a million nodes, and it is constantly growing. The modular thesis claims to improve decentralization by allowing anyone to run a rollup, but in practice, it centralizes the data availability function into a smaller set of actors.
The Takeaway: A Vision Forward
I have been in this industry long enough to know that narratives are powerful. They drive funding, they attract talent, and they shape the direction of development. But narratives can also be wrong. The modular blockchain thesis is not wrong in its entirety—it is a useful framework for thinking about scaling. But the current obsession with dedicated DA layers is a distraction. The industry is pouring billions of dollars into infrastructure for a problem that does not yet exist, while ignoring the real challenges: improving execution, reducing latency, and building applications that people actually want to use.
My advice is simple: focus on the data. If you are building a rollup, calculate your actual data needs. If you are generating less than 1 megabyte per block, use Ethereum’s blobs. The cost is low, the security is unmatched, and the complexity is minimal. If you are generating more than that—and you have a proven track record of demand—then consider a dedicated DA layer. But do not buy into the hype without doing the math. The industry cannot afford another cycle of building infrastructure for a future that never comes.
I think back to that November morning in Denver, staring at the transaction logs of a $80 million rollup that was processing 12 TPS. The team had been sold a vision of modular sovereignty, but they had built a castle on a foundation of sand. The castle will stand for a while—until the first real test of security, or the first bear market, or the first audit that reveals the hidden costs. I hope that when that test comes, the industry will have learned the lesson. Until then, I will keep auditing, keep writing, and keep asking the uncomfortable questions. The conscience of code demands nothing less.
— The Conscience of Code
— The Voice for the Conscience
— The Vulnerable Analyst