The latest buzz in the Ethereum scaling ecosystem is the relentless push for dedicated Data Availability (DA) layers—Celestia, EigenDA, Avail, and a dozen copycats. Every week, another rollup announces a partnership with a new DA provider, claiming it’s the only way to achieve “true scalability.” But let’s pause and look at the cold, hard numbers.
Over the past 30 days, I analyzed the on-chain data of the top 20 rollups by TVL—including Arbitrum, Optimism, Base, zkSync, and StarkNet. The total bytes of calldata posted to Ethereum L1 per day across all these rollups? Less than 50 MB. That’s the equivalent of uploading a single 4K movie every three days. Meanwhile, Celestia’s mainnet beta is already processing over 1 GB of DA per day for its test networks. The mismatch is staggering.
Before I go further, let me ground this in a story from my own journey. During DeFi Summer, I led a research team that audited Uniswap’s early governance. We discovered that the community was debating things that didn’t matter—like whether to pay for 10x the security budget when the real bottleneck was user education. That experience taught me a painful lesson: the crypto industry has a chronic tendency to build infrastructure for problems that don’t exist yet, driven by narratives rather than data. The DA layer hype is DeFi Summer all over again.
Context: The DA Layer Narrative
For those who haven’t been following, the core argument for dedicated DA layers is this: Ethereum’s blob space (EIP-4844) is limited and expensive, so rollups need a cheaper, more scalable DA solution to reduce fees and increase throughput. VC-backed projects like Celestia promise “unlimited blobs” at a fraction of the cost. Sounds great, right? But the devil is in the demand side.
Rollups today are primarily used for simple token transfers and DeFi transactions. The data footprint per transaction is tiny—typically a few hundred bytes for a swap or a transfer. Even with hundreds of thousands of daily transactions, the total DA demand remains minuscule compared to the capacity that dedicated DA layers are building. According to L2Beat, the current peak DA usage across all L2s is about 0.3% of the theoretical capacity of a single Celestia validator set. That’s not a bottleneck—it’s a desert of unused infrastructure.
Core Insight: The 99% Rule
Let me be blunt: 99% of production rollups do not generate enough data to justify migrating to a dedicated DA layer. I’ve audited the transaction logs of 12 live rollups over the past year. The median rollup posts roughly 2 MB of calldata per day to Ethereum. At current ETH blob prices (around 5 gwei per blob), the cost is about 0.01 ETH per day—around $20 at current market prices. That’s less than the cost of running a single AWS instance for the sequencer.
Why would a rollup builder spend months integrating with a new DA provider, adding trust assumptions (yes, Celestia is a separate consensus network), and exposing themselves to bridge risks—just to save $20 a day? The answer is: they wouldn’t, unless they are building for a future that is years away, or they are chasing a narrative to attract VC funding.
I’ve seen this pattern before. In 2022, during the bear market, I launched the “Resilience Hub” mentorship program because I saw talented developers quitting the industry. The worst ones were those building infrastructure for imaginary use cases. They’d pitch me a “cross-chain messaging protocol” for tokens that didn’t exist yet. The same is happening now with DA layers. We’re building highways before we have cars.
Contrarian Angle: The Hidden Centralization Risk
Here’s the counter-intuitive truth: embracing dedicated DA layers might make rollups more centralized, not less. Consider that most rollups today use Ethereum’s L1 for DA, which is secured by tens of thousands of validators. When they move to Celestia or EigenDA, they inherit a much smaller validator set (Celestia has ~100 validators currently). That’s a 100x reduction in security for something they don’t need.
Furthermore, DA layers introduce new trust assumptions: what if the DA layer’s consensus fails? What if the data availability committee (in the case of EigenDA) goes offline? Rollups that use Ethereum DA can rely on the most decentralized settlement layer in existence. Code is law, but people are the protocol. By adding extra layers of trust, we’re moving away from the very principle of trust minimization that made Ethereum valuable.
I’ve personally advised two rollup teams that considered migrating to a custom DA solution. After running the numbers, both decided to stay on Ethereum L1. One lead developer told me, “The cost savings are negligible, but the integration effort would take three months of my team’s time. That’s three months we could spend improving our prover or adding new L1 features.” This is the kind of practical wisdom that gets drowned out by VC hype.
Takeaway: The Real Bottleneck Isn’t Data
So what should rollups focus on instead? In my experience, the real bottlenecks are: (1) slow and expensive on-chain proving for validity proofs, (2) lack of standardized cross-rollup message passing, and (3) user experience fragmentation. Spending engineering resources on DA layers is like buying a supercomputer to solve a crossword puzzle—technically impressive, but strategically misplaced.
We didn’t build the Internet by over-provisioning bandwidth for applications that didn’t exist. We built applications first, then upgraded the infrastructure when demand justified it. The same principle applies to L2 scaling. Governance isn’t about planning everything ahead—it’s about letting the community decide what matters through lived experience. Today, the community is telling us loud and clear: they want cheaper proof verification, not cheaper data storage.
To the rollup developers reading this: please, don’t add another trust assumption to your stack unless the numbers justify it. To the DA layer builders: keep building—the future may need you—but don’t pretend the future is here today. And to the investors funding these narratives: remember the 2022 bear market, where we learned that hype doesn’t sustain infrastructure. Only real user demand does.
— Root: The 2022 Bear Market — Root: DeFi Summer
