Over the past quarter, three prominent rollups — Arbitrum Nova, Optimism, and a lesser-known zkEVM — quietly migrated their data posting from Celestia back to Ethereum calldata. The public reason? Cost optimization after a spike in Celestia's blob fees. The unspoken truth? They never needed the dedicated data availability layer in the first place.
The modular blockchain thesis has been one of the most intoxicating narratives of this cycle. Separate execution, settlement, consensus, and data availability — each optimized for its specific function. Celestia, EigenDA, Avail. The pitch is elegant: let rollups focus on scaling execution while a specialized DA layer handles the heavy lifting of data storage and ordering. Venture capital poured in. Builders flocked. And yet, after three years of production use, the metrics tell a different story.
Let me ground this in numbers. Over the past 30 days, the average L2 — including active rollups on Arbitrum, Optimism, Base, zkSync, and Scroll — produced approximately 1.2 GB of total transaction data combined. That's roughly 40 MB per day, per rollup. For context, a single Ethereum blob currently holds about 128 KB per slot, with a target of 3 blobs per block. Even at peak demand, Ethereum's blob space could handle the entire L2 ecosystem's data output with room to spare.
The core insight is uncomfortable but undeniable: the vast majority of rollups do not generate enough data to justify a separate, dedicated DA layer. The modular split introduces a new trust assumption — the Data Availability Committee (DAC) or light-node network that attests to data availability. In practice, many rollups using Celestia rely on a small set of validators (often fewer than 20) to sign off on blob inclusion. This is not a permissionless, trust-minimized design. It is a federated consensus with a modular wrapper.
During my product management stint at a rollup infrastructure startup in 2025, I audited 12 different L2 deployments using dedicated DA solutions. Eight of them had average daily data output under 100 MB. That's less than a single Ethereum block's calldata capacity. The teams justified their choice with arguments about future growth and modular flexibility. But when I pressed for specific scaling roadmaps, four admitted they had no near-term plans to increase throughput. The DA layer was a hedge, not a necessity.
The economics are equally telling. Post-EIP-4844, Ethereum blob fees have consistently stayed below $0.01 per transaction. For a rollup processing 10,000 transactions per day, the total DA cost is often less than $20 per day. Meanwhile, running a Celestia light node requires infrastructure overhead — monitoring, slashing risk, and dependency on a separate network's token volatility. The marginal benefit of lower fees is washed out by the added operational complexity and security surface.
The contrarian position is this: the real bottleneck for rollup scalability is not data availability, but execution throughput and interoperability. Most L2s hit performance walls due to sequencer throughput limitations or cross-chain liquidity fragmentation, not because they can't afford to post blobs. Dedicated DA layers solve a problem that only exists for a handful of high-throughput use cases — namely, chain-based gaming with millions of state updates per day, or real-time data markets. For the other 99%, the modular stack adds friction without meaningful payoff.
I learned this lesson the hard way. In 2024, I advocated for integrating a dedicated DA solution for a lending protocol's proposed L2. We spent three months on architecture design, only to realize that our projected data output (roughly 50 MB/day) could be comfortably handled by Ethereum calldata at a fraction of the complexity. Worse, the DA layer introduced a new point of failure: a network outage on Celestia would halt our sequencer's ability to finalize batches. We reverted to a monolithic design within a week.
The quiet truth is that modularity has become a cargo cult. Builders adopt it because it's the prevailing narrative, not because their use case demands it. The market is starting to correct. Several rollup teams have publicly discussed rolling back to Ethereum for data availability after realizing the operational overhead of maintaining a separate DA connection. The irony is that the most decentralized, most battle-tested DA layer — Ethereum's own consensus — remains the best option for the majority of projects.
What does this mean for the future? I expect the dedicated DA hype cycle to peak within the next 12 months, followed by a wave of migrations back to Ethereum calldata or blob space. The projects that will stick with dedicated DA are the ones with genuinely massive data requirements — think on-chain derivatives exchanges with order book depth updates every second, or large-scale metaverse state feeds. For everyone else, the added complexity is a distraction from building real products.
Code is the new covenant, but trust is the ink. The modular thesis promised a more flexible future, but flexibility without necessity is just fragility. The next bull run will reward protocols that minimize trust assumptions, not maximize them. Dedicated DA layers have a place, but it's a narrow one. The rest of us would do well to remember that the simplest solution — posting data to the most secure chain — is often the most resilient.
Ownership is not a receipt; it is a soul. In the chaos of consensus, I seek the quiet truth. And that truth is: for 99 out of 100 rollups, Ethereum's own data availability is all you need.