Over the past six months, I have been tracking blob utilization across Ethereum's post-Dencun network. The data is uncomfortable for the modular thesis: average blob usage hovers around 30 percent of target capacity, while the combined valuations of dedicated data availability layers have climbed into the billions. Signal in the noise. We are watching a narrative cycle consume more capital than the infrastructure it claims to serve. This is not new. But the cost structure this time is different, and the difference will decide which tokens survive the chop.
The modular thesis emerged from a real constraint. Ethereum calldata was expensive, and rollups settling directly to L1 were paying premium fees for a commodity. In 2022, Celestia proposed a radical fix: decouple data availability from execution and consensus entirely. The whitepaper was elegant, the timeliness impeccable. The market agreed. Fifty-five million dollars, then a hundred million more at a valuation that made a chain with negligible usage a top-tier network by market cap. EigenDA followed, selling DA through restaked ETH. Avail, a Polygon offshoot, entered. The story was compelling precisely because it was hard to falsify: DA is invisible infrastructure, and invisible infrastructure can sustain a narrative longer than visible applications. The DA wars became the defining infrastructure narrative of this cycle.
History repeats, but the code evolves. The actors change; the arithmetic does not. My first taste of this pattern was 2017, when I audited more than fifty ICO whitepapers. The most dangerous projects were not the obvious scams; they were the ambitious ones whose business models assumed the future would arrive faster than their burn rate. Dedicated DA layers are the 2025 version of that ambition โ beautifully engineered solutions whose revenue model depends on usage that has not yet materialized.
I have spent enough time auditing rollup architectures to say this plainly: the industry's most publicized problem is not a problem at current scale. It is a problem at a scale we have collectively decided to believe in.
Consider what a rollup actually posts. An optimistic rollup batches transactions, compresses them, and commits the result to a settlement layer. A typical batch from a live rollup contains a few thousand transactions at most. Let me be generous: two thousand transactions per batch, each consuming two hundred bytes post-compression. That is four hundred kilobytes per batch. At one batch every ten minutes, the rollup generates 2.4 megabytes per hour, roughly 57 megabytes per day.
Now the other side of the ledger. EIP-4844 gave Ethereum a blob target of three per block, expanding to six, each holding up to 128 kilobytes. At a twelve-second block time, Ethereum's blobspace absorbs roughly 2.8 gigabytes per day at target issuance, and up to 5.6 at maximum. That single lane can serve dozens of hypothetical active rollups of the kind I just described. And we are not close to target. Since Dencun, weekday blob usage routinely drifts between 20 and 50 percent of target. On most days, the blob fee sits at its minimum of one wei โ which, in any functioning market, is the clearest possible signal that supply exceeds demand. The price of blockspace is the most honest oracle in this industry, and it is currently telling us that the modular future is oversupplied.
The median active rollup posts daily transaction volumes in the low thousands, not millions. Its entire daily data output fits in one blob. One. A 128-kilobyte slot on the mainnet whose problem was supposedly severe enough to justify a new network, a new token, and a new validator set. During my audits, I reviewed rollups paying for Celestia blockspace while their entire on-chain activity could have been replicated in a spreadsheet. This is not an exaggeration. It is the median case.
Here is the uncomfortable truth the modular thesis refuses to model: data availability is a fixed-cost problem, not a marginal-cost problem. The cost of posting data to Ethereum falls as utilization rises; the marginal cost of an additional user is nearly zero. The cost of running a dedicated DA layer โ validators, token incentives, ecosystem grants, liquidity programs โ is fixed and unforgiving. At current volumes, every rollup posting to a dedicated DA layer is paying the economics of a skyscraper to live in a studio apartment. When I run the unit economics across the rollups I have audited, I consistently find that the DA bill outweighs every other operational cost combined. The product-market fit is inverted: those who need the service cannot afford it, and those who can afford it do not need it.
There is also the trust question, which marketing rarely mentions. The moment a rollup opts for an external DA layer, it must answer: who can reconstruct this chain if the DA layer withholds data? With Ethereum, the answer is anyone running a node โ a permissionless set measured in tens of thousands. With an external DA layer, the answer is anyone who runs a node on that specific network โ a much smaller set, often dominated by the same validators who secure the token and therefore benefit from its price. You have not eliminated the trust assumption. You have traded one trust assumption for a new token's validator set. In my cybersecurity training, that is not decentralization; that is a re-centralization with extra steps.
Data availability sampling is genuinely elegant. The cryptography is sound, and the design is among the most original contributions to blockchain systems since the fraud proof. But elegance does not equal demand. We are building highways for a city with one car.
The contrarian reading is not that DA layers are worthless. It is that we are pointing the wrong consumers at this technology, and the real winners have already been determined by a force that throughput charts cannot measure.
The demand will come from agents, not rollups. This cycle, the first autonomous economic actors are beginning to experiment with on-chain identity, provenance, and micro-transactions. When an AI agent needs to verify that another agent's stated action actually occurred, it needs a data layer that answers questions without trusting a centralized operator. That is a genuinely new demand profile, and it maps onto data availability sampling far better than the rollup thesis ever did. The DA market is not a fiction; its current customer base is. These agents do not care about narrative cycles; they care about cost, latency, and provability โ precisely the profile a DA layer should target.
Second, the DA war has already been won, and it was won by Ethereum โ not because Ethereum's roadmap is technically superior, but because the only data availability that matters to a settlement layer is the data availability of the settlement layer itself. Rollups settle where the economic security is, and economic security is where the largest pool of value is locked. Follow the protocol, not the influencer. The billion-dollar DA tokens are competing for a market whose anchor tenant is the very platform they set out to disintermediate.
This brings me to the cultural dimension, which I find more instructive than throughput benchmarks. We have seen this sequence before. Soulbound tokens spent three years as a celebrated concept because no one actually wants a permanent credit record on-chain. The gap was never technical; it was demand. The DA narrative is identical: the gap was never architectural; it was a user base. And the same narrative displacement arrived with the 2024 ETF era โ Bitcoin became Wall Street's custody toy while the vision of peer-to-peer electronic cash quietly became a footnote. Narratives mature into compliance documents. The DA wars are now maturing into consolidation.
For investors positioned in this sideways market, the signal is precise. The chop rewards patience, but not passivity. When I evaluate a modular project today, I ask one question: who consumes this data, and how many bytes do they actually generate? If the answer is other infrastructure projects subsidized by venture treasury, I move on. If the answer is economic agents whose incentives are independent of token price, I pay attention. That distinction is the difference between owning a toll road in a ghost town and owning one on the edge of a growing suburb. That is the difference between narrative exposure and structural exposure.
The next narrative is data consumption, not data availability. Watch for the first agent-to-agent verification economy that pays for its own bandwidth. Watch for settled DA layers repricing from growth multiples to utility multiples. And watch the smartest funds quietly exiting the infrastructure race before the public realizes the race has no finish line.
History repeats, but the code evolves. Signal in the noise: the builders who internalize this are already rotating from selling blockspace to selling what blockspace makes possible. The question is not whether dedicated DA layers survive. The question is whether you are still holding tokens for the problem โ or positioned for the problem after the problem.