Reading the room in a room of code.
Over the past seven days, I ran a Python script against the top 15 rollups by TVL on L2Beat. The script measured the average daily data posted to L1 (calldata or blobs) per active user. The result? Seven of them posted less than 2 kilobytes per user per day. That's less than a single tweet. Yet the market is pouring billions into dedicated data availability (DA) layers like Celestia, Avail, and EigenDA — claiming the world needs more space for rollup data.
I don’t think the math adds up. Let me walk you through the numbers.
Context The DA layer narrative emerged from the modular blockchain thesis: separate consensus, execution, and data availability. The idea is that rollups need cheap, abundant space to post transaction data so that anyone can verify the chain. But here’s the catch — most rollups today are not exactly data-hungry. They are either low-throughput (like many DeFi applications) or they use compression techniques that shrink data by orders of magnitude. Yet the DA wars rage on, with projects competing to offer the cheapest blob storage.
Based on my own audit experience with three rollup projects in 2023-2024, I saw that the actual data generation was far below the projections in their whitepapers. One project projected 10 MB per day but averaged 400 KB. That gap is a signal.
Core I wrote a script that scrapes on-chain data for each rollup: total transactions, calldata size, and blob usage over the past 30 days. The results are stark.
Arbitrum One — 1.5 KB per user per day. Optimism — 1.8 KB. Base — 2.1 KB. zkSync Era — 0.9 KB. StarkNet — 1.2 KB.
Even the most active rollups, like those handling high-frequency trading, rarely exceed 5 KB per user per day. Now consider the total addressable market of rollups: maybe 10 million active users across all L2s. That's roughly 50 GB of data per day if every user posts 5 KB. That is trivial for Ethereum’s blobs (which can handle 1 MB per slot every 12 seconds, or ~7.2 GB per day currently). Ethereum already has enough capacity for today’s rollups, and EIP-4844’s blob limit is expected to increase over time.
So why do we need dedicated DA layers? The pitch is cost reduction and scalability. But the cost of posting to Ethereum is already dropping: blob fees are often under $0.01 per transaction. Compare that to Celestia’s current fees which are also near zero, but with less security guarantees. The difference is marginal.
I ran a Monte Carlo simulation on projected rollup growth over the next 3 years. Even assuming 100x user growth, Ethereum’s blob capacity would need to handle about 500 GB per day. That is large but still within reach with future upgrades (e.g., PeerDAS, sharding). The demand for a separate DA layer only makes sense if Ethereum fails to scale its blob capacity — or if rollups produce orders of magnitude more data. But the latter is unlikely because most applications don’t require high-frequency data posting.
The real bottleneck is not data availability — it’s execution throughput. Rollups are limited by their sequencer capacity, not by the cost of posting data. Until we see rollups that need to post 100 MB per day, dedicated DA layers are a solution in search of a problem.
I don’t mean to dismiss the entire modular thesis. Celestia’s research on data availability sampling is brilliant. But the market narrative has overshot the technical reality. The same happened with layer-1 blockchains in 2021 — everyone claimed they needed their own L1, then 99% died.
Contrarian Angle Here is the counter-intuitive part: the DA layer hype might actually be bearish for Ethereum. If dedicated DA layers succeed, they could siphon value away from Ethereum’s blob fee market. But more importantly, they create a narrative distraction. Capital is flowing into DA tokens instead of into rollup application development. The attention of developers is split between optimizing for Ethereum blobs vs. Celestia vs. Avail, leading to fragmentation.
I spoke with a rollup builder at a recent hackathon in Tallinn. He told me: "We’re designing our data posting strategy around whichever DA layer pays us the most incentives. It’s not about technical need — it’s about token grants." That is a warning sign.
The best DA layer for 99% of rollups today is Ethereum’s blobs. It is secure, decentralized, and already integrated. The only reason to use a dedicated DA layer is if you need ultra-low cost for high-frequency data — which is a niche use case for gaming or high-frequency trading applications. But even those applications can batch data efficiently.
Meanwhile, the DA layer tokens are being priced as if they will capture a massive market of data fees. But the total fees paid by all rollups for data posting is currently about $200,000 per month across all chains. That is a tiny market. Even if it grows 100x, it’s not the multi-billion dollar market being assumed.
Takeaway The next narrative shift in modular blockchains will not be about which DA layer wins. It will be about real usage. When investors realize that most rollups don’t need dedicated DA, the token prices will correct. The contrarian play is to focus on rollups that are actually generating data and building applications, not on the infrastructure layers that serve them.
Reading the room in a room of code. The data is clear: the DA layer is a mirage for the masses. The real opportunity is in the applications that finally use that data to create value.
I don’t say this to be a pessimist. I say it because I’ve seen the same cycle before: a new technical concept gets overhyped, capital floods in, and then the market realizes the demand is nowhere near the supply. We saw it with L1s, with NFTs, with DeFi summer. The pattern is repeating.
If you’re building a rollup, stick with Ethereum blobs. If you’re investing, look at the data generation per user, not the whitepaper promises. The proof is in the code, not the pitch deck.