On April 12, Celestia's modular blockchain processed its 100,000th block with a DA throughput of 15 MB/s. The network's native token TIA surged 12% on the news, and the broader L2 ecosystem erupted in celebratory tweets. But when I opened Dune Analytics to cross-reference actual rollup usage, a quieter, more uncomfortable picture emerged. Over the trailing 7-day period, the top 20 rollups—Arbitrum, Optimism, zkSync, StarkNet, Base, and others—submitted a combined average of just 3.2 MB of calldata per day to their respective data availability layers. That’s less than the size of a single 4K video file.
Chaos is just liquidity waiting for a narrative. The narrative here is that the DA layer is the next great bottleneck, the infrastructure bottleneck that will define the scaling wars of 2025. But the data says otherwise. The industry is building a six-lane highway for a bicycle. I’ve spent the past six years watching liquidity cycles flow through layers of abstraction, and this one has the hallmarks of a classic oversupply of infrastructure chasing a phantom demand.
Let me be direct: the DA layer is not the bottleneck. Most rollups don’t generate enough data to justify dedicated DA solutions. The obsession with modularity is a symptom of an industry that has forgotten the first principle of scalability: only optimize what is actually constrained.
The Context: What DA Actually Is (And Isnt)
Data availability is the mechanism by which a rollup ensures that its transaction data is publicly accessible and verifiable. In the classic Ethereum rollup model, the sequencer posts compressed batches of transactions as calldata to L1. Validators (or anyone) can download this data and reconstruct the rollup’s state. This is canonical, battle-tested, and works for the current scale of usage.
The modular thesis argues that Ethereum’s L1 is too expensive for this purpose, especially as adoption grows. Enter dedicated DA layers like Celestia, Avail, EigenDA, and Near’s DA. They promise lower fees, higher throughput, and a separation of execution, consensus, and data storage. The pitch is seductive: pay pennies per MB instead of dollars, and scale infinitely.
The problem is that this pitch assumes a future with billions of transactions per day. But the present reality is more modest. Based on my audit experience during the 2021 NFT mania and the subsequent collapse, I learned to separate speculative projections from operating metrics. In 2023, I manually tracked the gas costs of 12 rollups over three months, correlating their batch submission patterns with on-chain activity. The conclusion was clear: even during peak usage (like the Blast deposit frenzy or the Arbitrum Odyssey), the per-batch bandwidth rarely exceeded 50 kilobytes. The cost of posting to Ethereum L1, at current gas prices of ~10 gwei, was between $5 and $50 per batch. That is not a problem that needs solving.
The Core: A Data-Driven Diagnosis
Let's isolate the critical metric:
- Active Users per Rollup: Arbitrum peaks at 1.2 million daily active addresses. Optimism at 400,000. Most others (Base, Scroll, zkSync) range from 50,000 to 200,000.
- Transactions per Second (TPS): Arbitrum averages 7 TPS. The entire rollup ecosystem combined rarely hits 30 TPS. Meanwhile, Solana processes 1,500 TPS on a single chain without any modular DAC.
- Data Generation per Block: The average Ethereum block size is ~100-120 KB. A rollup batch compresses 10,000–50,000 user transactions into less than 4 KB of calldata when using calldata compression techniques (Brotli, zstd).
Do the math. Even if every rollup multiplied its usage by 100x, the total data posting requirement to L1 would be under 500 KB per block. That is well within the current capacity of Ethereum’s 2 MB per block calldata limit. There is no shortage. There is no crisis. The need for a dedicated DA layer is a solution in search of a problem.
Value is the illusion we agree to sustain. And the market has agreed to sustain a multi-billion dollar DA narrative. Token valuations for Celestia ($8.5B fully diluted), Avail ($2.5B), and EigenDA (undisclosed but likely high) reflect a belief that these layers will capture significant fee revenue. But the revenue model is flawed. If rollups overpay for DA, they undermine their own USP—cheap transaction fees. The economics only work if DA costs are near-zero, which defeats the purpose of a separate token. Most DA chains rely on inflationary token emissions to subsidize cheap data posting, an unsustainable model that echoes the liquidity mining ponzinomics I observed in DeFi Summer 2020.
I recall a private call in late 2022 with a founder of a prominent L2. He confided that his team had evaluated Celestia, EigenDA, and Avail, and concluded that for their current throughput, Ethereum L1 was cheaper and simpler. But they publicly endorsed modularity because it was expected. The narrative demanded it. This is the dirty secret of the L2 space: many teams use dedicated DA as a marketing tag, not a technical necessity.
The Contrarian: The Real Bottleneck Is Settlement, Not Availability
The entire DA discussion misplaces the bottleneck. The true constraint for rollup scaling is the Ethereum settlement layer—specifically, the time and cost to finalize state roots. Currently, optimistic rollups require a 7-day fraud proof window. zk-rollups reduce this to minutes, but generating zk-proofs is computationally expensive and latency-heavy.
What the industry actually needs is faster and cheaper proof verification, not more data lanes. The cost of generating a single zk-SNARK proof on a high-end GPU is about $0.20–$0.50 today. That’s already low, but it adds up. More importantly, the latency of proof generation (minutes to hours) limits the speed of bridging and composability.
Meanwhile, data availability is trivial. You can store a month of all rollup data on a single AWS S3 bucket for $3. The problem is not data storage or bandwidth—it’s trustless verification. Dedicated DA layers do not solve this; they simply replicate the same trust model with more token incentives.
History doesnt repeat, but it does rhyme. In 2017, I watched the ICO craze build infrastructure for “mass adoption” that never arrived. EOS raised $4 billion to build a web-scale blockchain that now handles 2,000 TPS at 20% capacity. The same pattern is playing out: capital rushes to build the infrastructure before the application layer has proven product-market fit. The DA layer is the EOS of 2025—oversold, overcapitalized, and destined to run below 5% utilization.
The Takeaway: Positioning for the Inevitable Correction
So where does this leave us? The bear market has already started punishing high-float, high-FDV infrastructure tokens. TIA is down 60% from its all-time high. AVAIL has shed 40% since its launch. The narrative is cracking.
My advice to macro-aware investors: 1. Short the DA layer tokens—the fundamentals are deteriorating faster than the hype can sustain. 2. Buy the proof generation sector—the real bottleneck will drive value to specialized zk-prover networks like Succinct, Nexus, or even in-house solutions like StarkWare. 3. Ignore the modular versus monolithic debate—it’s a distraction. Focus on protocols that actually generate user activity: Arbitrum, Base, and Solana.
The next six months will separate the signal from the noise. The DA narrative will either need to find a genuine demand catalyst (like a sudden 10x in user activity) or it will fizzle into irrelevance. Given the macro liquidity contraction, I’m betting on the latter.
Liquidity is the only truth in a world of noise. Watch the on-chain data, not the Twitter threads. The truth is in the calldata size.