The quiet hum of a data center in Shanghai was broken by a notification: Celestia’s native token had surged 45% in a week, while Ethereum rollups collectively shed 12% of their total value locked. I sat back, sipping cold tea, and wondered—are we building infrastructure for a problem that doesn’t yet exist? This is the ghost in the machine of trust: we invest in narratives of scarcity before we measure the actual demand for data.
Context: The Modular Thesis and Its Discontents The modular blockchain thesis emerged from a simple truth—monolithic chains cannot scale without sacrificing decentralization. Rollups promised to offload execution, while data availability (DA) layers like Celestia, EigenDA, and Avail claimed to be the "shared settlement" fabric for a multi-chain world. The pitch was seductive: dedicated DA reduces cost, increases throughput, and allows sovereign rollups to run without Ethereum’s bottleneck. By mid-2026, over 30 rollup projects had integrated or announced plans to use a third-party DA layer, with total venture funding exceeding $2.5 billion.
But I’ve been listening to this quiet hum for six years. When I first analyzed Arbitrum’s whitepaper in 2020, I saw a different truth: most rollups were designed for low-throughput applications—NFT bridges, perp exchanges, or prediction markets. They did not need terabytes of blobspace; they needed instant finality and cheap execution. The narrative of "infinite scalability" captured hearts, but the data tells another story.
Core: The Data Availability Overhype—A Quantitative Look Over the past 90 days, I tracked the actual blob usage of the top 15 rollups (including Arbitrum, Optimism, Base, zkSync, and StarkNet). On average, each rollup publishes less than 500 kilobytes of compressed data per batch to Ethereum L1. To put that in perspective: a single JPEG of a cat from 2015 is larger. Even at peak usage—during a memecoin frenzy on Base in March—the daily blob output barely exceeded 2 megabytes per rollup.
Now compare that to the capacity of dedicated DA layers. Celestia’s testnet in 2025 demonstrated a theoretical throughput of 6 megabytes per second. EigenDA boasts 10 megabytes per second with 50 nodes. We are building firehoses for puddles. The user demand for data availability is not the bottleneck; execution complexity and liquidity fragmentation are.
Based on my audit experience of five modular rollup stacks in 2024, I found that most projects overestimate their data needs by an order of magnitude. Their typical transaction volumes—a few hundred trades per minute—can easily be settled on Ethereum L1 at a cost of $0.10 per transaction. The real cost driver is not data publishing but the L1 gas for fraud proofs or validity proofs. By moving to a separate DA layer, they save pennies on blob fees but introduce a new trust assumption: that the DA committee is honest and available.
Consider the math: A rollup doing 10 transactions per second (TPS) at 200 bytes per tx produces 2,000 bytes per second, or 172.8 megabytes per day. Ethereum’s current blob target is 1 megabyte per slot (every 12 seconds), or over 7,000 megabytes per day. Even with all rollups aggregated, they use less than 3% of Ethereum’s blob capacity. The narrative of "DA scarcity" is a shadow projected by venture capital, not a technical reality.
Moreover, the claim that dedicated DA layers reduce finality time is misleading. Ethereum blobs finalize in ~15 minutes (economic finality); Celestia’s light node finality is ~10 minutes, but with higher liveness risk. The improvement is marginal for most use cases. Meanwhile, the complexity of bridging to a different DA layer introduces additional latency and security assumptions. The ghosts in the machine multiply.
Contrarian: The Real Bottleneck Is Execution, Not Data The contrarian angle here is uncomfortable for the modular maximalists: we are solving the wrong problem. The limiting factor for rollup adoption today is not how fast they can publish data, but how quickly they can settle cross-rollup liquidity and execute complex smart contracts. Three of the five rollups I audited had transaction confirmation times exceeding 30 seconds due to decentralized sequencer latency—not data availability.
Furthermore, the symbiotic relationship between DA layers and rollups has a vulnerability: if a rollup’s throughput never grows beyond 50 TPS, the economic security of the DA layer cannot be sustained by fees alone. Most DA tokens rely on inflation to reward validators. When inflation drops, so does security. We are mapping a future where 20 DA layers compete for a fraction of the total blob demand, leaving most underutilized and insecure.
There is also an ethical dimension. By convincing teams to adopt a separate DA layer, we are fragmenting the social consensus that makes cryptocurrencies resilient. Ethereum’s security comes from millions of stakers validating the same state. When a rollup moves to a DA layer with 50 validators, it inherits the risk of cartel behavior. This is a return to the bank-like trust we supposedly escaped.
Takeaway: The Next Narrative—Execution and Interoperability Listening for the quiet hum of the second layer, I predict that the DA narrative will cool within 12 months. The market will realize that 99% of rollups do not generate enough data to need dedicated infrastructure. Instead, attention will shift to shared sequencers (like Espresso or Astria) and cross-rollup messaging protocols—where the real friction lies. Weaving code into the fabric of physical reality means building systems that serve human needs, not speculative theses.
The signal in the noise of 2026 is clear: stop chasing the firehose of data and start mending the pipes of execution. The ghosts in the machine of trust are not in the blobs—they are in the sequencers.