I remember the moment the illusion shattered. It was three in the morning, and I was staring at a Dune Analytics dashboard for a rollup that had just raised $80 million. The network had been live for six months. Its proud team boasted of a “dedicated data availability layer,” a custom-built solution secured by a separate set of validators. They called it a “modular breakthrough.” But the numbers told a different story — a story of emptiness masquerading as innovation. In those six months, the rollup had posted a grand total of 4.2 megabytes of transaction data. That’s less than a single JPEG of a bored ape. The dedicated DA layer was processing less data per day than a single Ethereum L1 block did in 2018. And yet the market had blessed it with a valuation that could fund a small country’s infrastructure.
This is not an outlier. Over the past three years, I have audited the on-chain footprints of over 40 rollups and validiums. The pattern is so consistent it has become a scar on my conscience: 99% of these projects generate so little data that a dedicated data availability layer is not just overkill — it is a costly, performative distraction. The industry has built a cathedral of infrastructure for a congregation that has not yet arrived. And we are all complicit in the deception.
Let me be clear: I am not against modular architectures. I have spent countless nights arguing for the philosophical beauty of separation — of sovereignty through separation, as I wrote in my 30,000-word analysis of Celestia’s architecture back in 2022. The idea of decoupling execution, settlement, consensus, and data availability is intellectually elegant. It promises a future where blockchains can specialize and scale without compromise. But the gap between the promise and the current reality is not a crack — it is a canyon. And the longer we pretend otherwise, the more we risk building a house of cards that will collapse when the market’s euphoria inevitably turns to curiosity.
The Context: How We Got Here
To understand why dedicated DA layers are overhyped, we must first understand the problem they claim to solve. In Ethereum’s original monolithic design, every node stores all transaction data — forever. This ensures that anyone can reconstruct the state, but it also imposes a hard limit on throughput. Rollups were born as a scaling solution: they batch hundreds of transactions into a single compressed blob, post that blob on Ethereum, and rely on fraud proofs or validity proofs to ensure correctness. The key constraint is that Ethereum’s blob space (post-EIP-4844) is limited. Each blob is roughly 128 kilobytes, and Ethereum can handle about six blobs per slot. That’s enough for a few rollups, but not for a future with thousands.
Enter dedicated data availability layers — Celestia, Avail, EigenDA, and others. Their pitch is seductive: we can store rollup data on a separate, high-throughput network, with its own consensus and security model. This frees Ethereum from the burden of storing endless blobs. Rollups can use a fraction of the cost, and the DA layer can scale to many megabytes per second, potentially enabling new use cases like fully on-chain games or high-frequency trading.
I bought into this vision. I really did. In 2023, I spent three months working with a small team to integrate a rollup with Celestia’s testnet. I wrote the smart contracts, I tuned the namespaced merkle trees, I even wrote a blog post titled “The Dawn of Modular Sovereignty.” But the moment I saw the actual data usage patterns of production rollups, my excitement curdled into dismay.
The Core Discovery: 99% of Rollups Don't Generate Enough Data
Let’s do some back-of-the-envelope math. A typical rollup transaction on Arbitrum or Optimism is about 200 bytes. If a rollup processes 100,000 transactions per day — a respectable number for most L2s — that amounts to roughly 20 megabytes per day. Over a month, that’s 600 MB. Now, Ethereum’s blob capacity today is about 786 kilobytes per slot (six blobs at 128 kB each). With 7,200 slots per day, the total blob capacity is about 5.7 gigabytes per day. That’s enough for nine rollups doing 100k TPD each, or one rollup doing nearly a million transactions per day, before hitting blob limits.
But here’s the uncomfortable truth: very few rollups actually do 100k transactions per day. I looked at the top 20 rollups by TVL in February 2026. The median daily transaction count was 12,000. The average transaction size was 180 bytes. That’s 2.16 MB per day. A single blob every 60 seconds would suffice. The dedicated DA layers I have audited were processing between 2 and 10 MB per day across all their rollup customers combined. That’s less than the data generated by a single Telegram group sharing memes.
During my audit of one layer-2 project that shall remain nameless, I discovered that the team had deployed a custom DKIM-based DA layer with a dedicated validator set of 21 nodes. They had spent over $500,000 on infrastructure and cloud hosting. When I analyzed the on-chain data, I found that the rollup had posted exactly 47 blobs in three months, totaling 1.2 MB. The team’s own dashboard showed “data throughput” as a graph with a peak of 8 bytes per second. I felt a profound sadness. Not anger — sadness. Because I knew the team was brilliant. They had built a beautiful, redundant, secure system for storing the digital equivalent of a shopping list.
I have a spreadsheet from my own research. I call it the “Empty Blob Registry.” Out of 40 rollups, 38 had blob usage that could easily fit within Ethereum’s existing blob capacity without any contention. Only two — a DeFi protocol running a high-frequency order book and a gaming rollup with 200k daily active users — actually approached the limits. And even those two could be accommodated by Ethereum’s blob count with careful planning.
The Hidden Cost: Performance Theatre
Why do projects build dedicated DA layers when they don’t need them? I believe it is primarily a marketing decision, driven by the bull market’s hunger for novelty. In a frothy market, “we have our own DA layer” sounds more impressive than “we post data to Ethereum blobs.” It signals technical sophistication and modular thinking. Investors reward it. Token buyers reward it. But there is a hidden cost beyond the financial waste.
Every new DA layer introduces a new trust assumption. The security of the rollup is now tied to the economic security and decentralization of that DA layer. Most dedicated DA layers launched with fewer than 50 validators and staking values under $100 million. In my analysis, I found that an attacker would need to bribe or compromise only three of the top five validators to temporarily censor or reorganize the DA layer, leading to potential loss of funds in the rollup. Compare that to Ethereum’s 1 million+ validators. The trade-off is rarely presented to users.
I recall a specific conversation with a CTO who argued that their DA layer was “secure enough” because it used delegated proof-of-stake with a “robust” slashing mechanism. I pointed out that their top three validators were operated by the same cloud provider. He said, “We’re working on that.” That was eighteen months ago. The validators are still on AWS.
Contrarian Angle: Maybe We Need Dedicated DA After All
Now let me play the devil’s advocate. I have argued this point with many brilliant engineers at Celestia, Avail, and Eigen Labs. Their counterargument is powerful: “You are looking at today’s usage. We are building for tomorrow. When hundreds of millions of users are using crypto daily, when every cent of cheap data matters, dedicated DA will be essential.” They remind me that the early internet had empty backbones too. Just because the pipes are empty now doesn’t mean they should not exist.
I accept that logic up to a point. Yes, scaling can be anticipatory. But there is a difference between building infrastructure for a known demand and building infrastructure for a hypothetical demand with no clear roadmap. The internet boom had a clear trajectory of increasing usage — first email, then web browsing, then video. What is the equivalent next step for L2s that would generate ten times today’s data? Fully on-chain games? Real-time trading? I have looked at the data for the top on-chain games. The most successful one, a battle royale game, generates 50 kB per game, with 10,000 games per day. That’s 500 MB per day. Still manageable on Ethereum. High-frequency trading? The top DEX on Arbitrum does 1,500 trades per minute. That’s 300 kB per minute — under 500 MB per day. We are orders of magnitude away from exhaustion.
Furthermore, I worry about the fragmentation of liquidity and composability. When rollups use different DA layers, cross-layer communication becomes more complex. Each DA layer is a new execution environment with its own latency and finality. I have seen many rollups attempt to build bridges between their DA layer and Ethereum, only to introduce additional points of failure. In my ethical code audits, I have found at least four vulnerabilities in bridge contracts that rely on DA layer state attestations. These are not hypothetical risks — they have led to actual exploits.
The Takeaway: A Call for Honest Infrastructure
I do not know if dedicated DA layers will one day be necessary. Perhaps in ten years, when every car and phone is a node generating proofs, we will look back at today’s skepticism as short-sighted. But in this moment — in this bull market — we owe it to ourselves and to the newcomers to be honest about the current state. The numbers are not controversial. I have shared my spreadsheet with the teams I audited. Some thanked me. Others ignored me. A few accused me of being a maximalist stuck in the past.
I am not against modularity. I am against performative engineering that consumes capital and attention while delivering no real benefit. The ethos of decentralization has always been about substance over form — about code being law, not marketing being reality. If we build castles in the air, we should not be surprised when the market wakes up and realizes the foundation is made of conference slides.
So here is my request, from one builder to another: before you deploy a dedicated DA layer, ask yourself one question honestly. Run the numbers. If your rollup is generating less than 10 MB of data per day, you do not need a separate DA layer. You need more users. Use Ethereum’s blobs. Use a data availability committee. Use a simple off-chain submission with fraud proofs. Save the modular architecture for the day when your daily data output exceeds the capacity of a single YouTube video. That day may come. But it is not today.
Signatures
⚠️ Deep article forbidden without attribution — this one took 40 rollup audits and a lot of uncomfortable conversations.
⚠️ Deep article forbidden without attribution — the spreadsheet is real, and I will share it with anyone who asks.
⚠️ Deep article forbidden without attribution — the CTO on AWS still hasn’t changed validators, but maybe this article will remind him.
⚠️ Deep article forbidden without attribution — the per-byte cost of vanity infrastructure is measured in lost trust.
⚠️ Deep article forbidden without attribution — I wrote this at 4 a.m. after another dashboard revealed empty blobs.