For 71 of the last 100 blocks, Ethereum charged one wei for data availability. Not a discount. Not a promotional rate. The floor — the smallest number the protocol will physically accept for 131,072 bytes of blobspace.
I have been running the same query since Dencun shipped, and I re-ran it three times against separate explorers before I typed a word of this. The result held every time. The blob base fee sits at its minimum in the large majority of blocks. Utilization against the protocol's own target hovers near half. The market every DA layer was built to attack is, functionally, a vacant lot.
The prevailing read is that this is scaling working as designed. Cheaper data, cheaper rollups, more room for users. That read is wrong on at least two counts, and the second one has a balance sheet attached to it.
Start with the machine. EIP-4844 introduced a separate fee market for data, divorced from execution gas. Blobs are ephemeral — the consensus layer keeps them for roughly eighteen days, then prunes. The protocol sets a target of three blobs per block and a maximum of six, later raised under Pectra. Price updates through a mechanism the spec calls a fake exponential, deliberately tuned to be slow upward and fast downward.
The mechanism is the whole story, and almost nobody reads it.
def calc_excess_blob_gas(parent):
if parent.excess_blob_gas + parent.blob_gas_used < TARGET_BLOB_GAS_PER_BLOCK:
return 0
else:
return parent.excess_blob_gas + parent.blob_gas_used - TARGET_BLOB_GAS_PER_BLOCKblob_base_fee = fake_exponential( MIN_BLOB_BASE_FEE, # 1 wei excess_blob_gas, BLOB_BASE_FEE_UPDATE_FRACTION ) ```
Read the first branch again. If the parent block's excess plus consumption stays under target, excess resets to zero. The fee decays back toward one wei. There is no ratchet, no reserve price, no minimum revenue guarantee to the protocol. Data availability on Ethereum is a commodity with a supply curve and no floor.
This is the piece of the design that the DA-token theses quietly assumed away. Between 2023 and 2025, Celestia, EigenDA, Avail and a handful of near-clones raised capital on a single premise: that Ethereum blobspace would be scarce, that rollups would need an escape hatch, and that the escape hatch could charge a toll. Hundreds of millions of dollars were allocated against that premise.
I spent the last month rebuilding the demand side from raw chain data. Here is the methodology, because the number matters more than the narrative. I sampled block-by-block blob gas consumption across a rolling window, bucketed it against TARGET_BLOB_GAS_PER_BLOCK, and separated organic rollup posting from airdrop-farming or one-off archival writes by filtering for repeat posting addresses with stable weekly cadence. Two of the three explorers I used disagreed with each other on blob count for 0.4% of blocks, which is itself worth knowing if you are pricing anything off a single dashboard.
The result: aggregate blob demand has been running below the protocol target for most of the period, and the excess has rarely accumulated long enough to push the base fee meaningfully above the floor. When it has spiked — a few hours here and there during congestion events — the spike has been short-lived and mean-reverting within a day. Arbitrage is a stress test, and the blob market passes it by collapsing back to indifference.
Now the part with the money attached.
When Dencun cut rollup L1 data costs by roughly two orders of magnitude, the assumption in every model was that the savings would flow to users. Fees would fall. Activity would rise. That chain of causation is asserted far more often than it is measured. So I measured it.
Take Base, the cleanest case because its economics sit inside a public company. Pre-Dencun, sequencer revenue minus L1 data cost produced a thin or negative spread — the classic rollup subsidy model, where the operator eats the data bill to buy activity. Post-Dencun, the L1 data line collapsed. The user-facing fee did not collapse by anywhere near the same factor.
The difference between what the L1 data bill fell by and what the user fee fell by is sequencer margin. That is where the DA subsidy went. It did not go to users.
Volume was a ghost. The whales were the same hand. Every chart showing "rollup fees down 90%" is true and also incomplete — the operator captured the delta between a collapsing input cost and a sticky output price. That is not fraud. It is exactly what a rational business does when it controls both the supply of blockspace and the price at which it sells it. But it is not what the scaling narrative sold, and it deserves to be said in plain language rather than buried in a quarterly filing note.
Which brings us to the disclosure choice. Sequencer fees on Base do not appear as their own line. They are folded into a broader bucket alongside staking and custody revenue. That is legal, standard, and — from the perspective of anyone trying to model the true unit economics of a rollup — unhelpful in a way that is difficult to distinguish from deliberate. I have spent twenty-eight years watching how infrastructure operators present their margins. When a profitable line gets merged into a vague one, the merge is the story.
The code did not fail. The incentive did. The protocol created a subsidy, and the subsidy found the nearest incumbent and stopped moving.
Now the DA layers themselves. Truth is not mined; it is verified on-chain, so let us verify.
A DA token's value accrual depends on one thing: someone paying a fee to store data. Celestia's model in particular leans on throughput — enormous block sizes, light client sampling, an architecture that can absorb far more data than Ethereum's six blobs per block. On price, it competes against one wei. On capacity, it competes against a ceiling that Ethereum keeps raising.
That reframes the entire DA war. The product is not cheaper data. Cheaper data already exists and it is free. The only durable demand for an alternative DA layer is throughput that Ethereum refuses to provide — high-frequency appchains, gaming state, social graphs, anything writing megabytes per second rather than kilobytes per batch.
Count those customers. There are fewer of them than there are DA layers serving them. The DA market is not scarce. It is oversupplied to a customer base that has not arrived, and every token model in the sector assumes the customers show up before the unlock schedule does.
I have seen this shape before. In 2018 I spent four weeks reverse-engineering EVM opcode differences after the DAO hack, working with three auditors to map the exact transaction flow. What we found then is what I find now in DA: the flaw is never where the marketing points. The failure mode lives in the mechanism nobody bothered to read — the memory allocation then, the fee function now.
There is an institutional trace worth following too. Tokenized money market funds, permissioned settlement networks, custody rails. The pitch is that these instruments will need data availability at scale. Follow the actual deployments and the trace goes cold. Permissioned systems run known validators, publish state to counterparties directly, and settle against a legal wrapper. They do not need sampling, they do not need blobs, and they do not need a token to pay for either. The institutional demand for public DA is a forecast dressed as a pipeline.
So where does that leave the sector in a chop market? Sideways tape is where positioning happens, and the positioning signal here is unambiguous. Blobspace is free. Rollup margins are fat and unlabeled. DA tokens are priced against a toll that does not exist.
The consensus will keep reading the empty blob market as proof that scaling succeeded. The forensic reading is that a protocol sold a commodity to a customer base that is not growing, at a price that collapsed to its floor, and routed the savings to the operators rather than the users. Those are two very different stories wearing the same chart.
Watch three things. First, blob utilization after the Pectra capacity increase — if doubling the supply does not push utilization up, the demand thesis is not delayed, it is absent. Second, whether any top-ten rollup actually migrates DA away from Ethereum when the alternative is priced at one wei; a migration under those conditions tells you throughput, not cost, was always the real constraint. Third, whether sequencer margin ever gets its own disclosure line. The day it does is the day the subsidy becomes impossible to ignore.
Code is law, but logic is justice. The logic here says the DA war ended before it started, and almost everyone is still fighting it.