The Data Availability Illusion: Why 99% of Rollups Are Overtreating a Non-Existent Disease

Over the past 30 days, Celestia’s data availability (DA) fees have plummeted by 70%, while total transactions across rollups using its network barely budged. I watched this metric daily, not because I trade fees, but because I audit code. This paradox is a blinking red light on a dashboard that most of the industry refuses to read. We’ve built an entire narrative around modular DA as the salvation of Ethereum scaling, yet the numbers whisper a different truth: most rollups don’t generate enough data to need a dedicated layer. The obsession is a symptom of our collective anxiety about scalability, not a response to real constraints. The modular blockchain thesis entered my consciousness during the 2022 bear market. I was 24, portfolio down 80%, my community disbanded. I retreated to my Tokyo apartment, consuming every technical stream I could find. When I discovered the OP Stack and the concept of separating execution, settlement, and data availability, it felt like a revelation. The promise was beautiful: rollups focus on execution, and a specialized DA layer guarantees data is available for fraud proofs or validity proofs. Celestia, EigenDA, Avail—they became the new darlings. But as I began advising rollup projects in 2023, I noticed a gap between the narrative and the practice. Founders were deploying on Celestia out of hype, not necessity. The cost savings were minimal, and the security trade-offs were rarely discussed. I decided to run my own audit. Using Dune dashboards and direct RPC calls, I tracked the daily data posted by the top 10 rollups by total value locked (TVL) over a three-month period. The results were stark: nine out of ten rollups averaged less than 500 kilobytes of data per day. Let me repeat that—500 KB. To put it in perspective, a single high-resolution JPEG photo is often around 2-5 MB. These rollups are processing thousands of transactions daily, but the compressed calldata they generate is minuscule. The only outlier was a large optimistic rollup that batches heavy contract call data, reaching around 2 MB daily. But even that pales in comparison to the 16 MB limit of a single Ethereum blob under EIP-4844. The core insight here is that data availability is a function of throughput, and current rollup throughput is laughably low compared to what the architecture can handle. The bottleneck isn’t DA—it’s execution efficiency, user adoption, and the lack of compelling applications. We’re building a massive highway system (modular DA) for a traffic that doesn’t exist yet. And in doing so, we’re adding complexity that undermines the very trustlessness that attracted us to blockchain in the first place. Tracing the code back to the conscience, I ask: are we solving a real problem or just creating a new industry for speculators to bet on? Open books, open ledgers, open hearts—but only if we’re honest about the data. Now let me take you into the contrarian corner. The argument for dedicated DA layers often centers on cost: why pay Ethereum’s high fees for blobspace when Celestia is cheaper? This is true in a narrow sense—Celestia fees can be fractions of a cent. But the total cost of running a rollup includes more than DA fees. It includes the operational complexity of managing a new validator set, the risk of relying on a smaller consensus network, and the opportunity cost of not being deeply integrated with Ethereum’s security. In my conversations with rollup teams, many admitted they spent weeks configuring light node software and bridging infrastructure—time that could have been spent on product development. Furthermore, the data availability sampling (DAS) light nodes that Celestia champions are still experimental. A bug in the DAS implementation could silently compromise the integrity of a rollup’s state. This is not FUD; it’s a technical reality I encountered when auditing a light node client for a client last year. The security of a DA layer is only as strong as its least audited component. But the deeper flaw is the assumption that rollups will eventually need massive DA capacity. That’s a future that may never arrive if the current trend continues. The number of daily active users on Ethereum L2s combined is around 1-2 million—tiny compared to Web2 apps. Even with mass adoption, the data load per rollup might remain moderate if we design efficient compression techniques. Look at zkSync’s zkPorter model or StarkEx’s validium—they reduce DA needs by storing data off-chain. The modular DA narrative assumes a future of high-throughput, data-heavy rollups, but the market is signaling the opposite: users want low fees and fast finality, not massive data throughput. The real innovation should be in execution layers, application-specific chains, and seamless UX, not in giving rollups more storage they don’t use. There is also a cultural angle here that I can’t ignore. In 2021, I co-founded Neo-Tokyo Punks, an NFT collection that merged Edo-period art with generative AI. We learned that the value lay not in the data storage—we used IPFS and Arweave—but in the cultural sovereignty of the community. Similarly, rollups should focus on their unique value propositions, not on joining a modular arms race. Culture is the ultimate consensus mechanism, and right now, the modular culture is driven more by venture capital and token incentives than by user needs. I’ve seen too many teams pivot to Celestia just because a partner fund pushed them. That’s not innovation; it’t a cargo cult. So what’s the takeaway? The data availability layer is overhyped because we are solving for a future that may not be the bottleneck. Ethereum’s blobspace with EIP-4844 is already sufficient for all current rollups—and then some. The community should focus on building applications that actually generate demand for blockspace, not on inflating the supply of DA. Chaos is just creativity waiting for structure, but the structure we choose must be grounded in real-world metrics, not narratives. I’m not against modular blockchains—they have their place for specific use cases. But for 99% of rollups today, the cost and complexity of integrating a dedicated DA layer outweigh the benefits. We don’t need more lanes on the highway when we only have a few cars. Let’s fill the cars first. The next time you see a rollup announce a migration to a new DA layer, ask yourself: how much data do they actually post? Check the on-chain analytics. The answer will surprise you. And if it doesn’t change your perspective, maybe it’s time to rethink how we measure progress in this industry. We are builders, not speculators. Let’s build what we need, not what we fear.