The blob base fee has been sitting at 1 wei for most of the past quarter. That single number โ the floor of Ethereum's dedicated data-availability market โ is the quietest and most consequential fact in rollup economics right now.
One wei is the protocol minimum, the price at which the market clears when demand barely exists. It means any rollup can post a 128 KB blob for a rounding error. It means the entire data-availability line item that dominated every 2023 pitch deck has been functionally zeroed out. And in the seven days before I wrote this, a mid-sized general-purpose rollup I track spent more on zero-knowledge proofs than it collected in sequencer fees โ for the fourth consecutive month.
It is not insolvent. Insolvency is a legal event, triggered by a missed payment. What this operator has is something more durable and harder to repair: a cost base denominated in hardware, engineering salaries, and electricity, set against a revenue base denominated in a token that has fallen hard against the currency you need to buy the hardware.
Cheap blobs were supposed to save this business. Reading the code that writes the culture, they did the opposite. Cheap data availability did not make rollups profitable. It made them legible โ it removed the last variable, and exposed that data availability was never the binding cost.
To understand why, you have to go back to the pre-Dencun arithmetic.
Through 2022 and 2023, a rollup's dominant expense was Ethereum L1 calldata. Batches were compressed, state diffs minimized, and still โ during gas spikes โ posting a single batch could cost five figures. Teams hired compression specialists. They published blog posts about byte-packing. The entire intellectual energy of the scaling debate was pointed at one target: get the data cost down, and everything else follows.
EIP-4844 delivered on that promise in March 2024. Blobs arrived with their own fee market, their own EIP-1559-style mechanism, a target of three per block and a maximum of six, and a hard floor of 1 wei for the base fee. Pectra raised the ceiling again in May 2025, moving the target to six and the maximum to nine. The result is what the people who model these markets expected and what the marketing departments did not: blobspace is, in most blocks, essentially free.
That should have been a windfall. Instead it revealed a three-part cost structure that had been hiding behind the DA number all along. Proving. Settlement. Sequencing.
Settlement is the cheapest of the three and, on most optimistic rollups, the most theatrical: a bonded assertion, a challenge window, a fraction of a cent. Sequencing is where the revenue is โ priority fees, order flow, and the extractable value that comes with being the only party who sees a transaction before it lands. Proving is where the money dies.
And that is the irony of this cycle. The market celebrated the collapse of DA costs as if it were margin expansion. What it actually did was strip away the one expense that scaled with usage and leave standing the one that scales with complexity. Navigating the storm to find the steady current means reading the shape of the remaining cost curve, not the size of the deflated one.
Start with proving, because it is the least understood line item on any rollup's books.
A zero-knowledge proof over a general-purpose virtual machine is not a fixed cost. It is a function of cycles executed, and cycles executed is a function of how much logic you insist on proving. A simple token transfer inside a zkEVM runs in the tens of millions of cycles and costs fractions of a cent to prove on modern GPU hardware. A multi-hop swap routed through three pools, with oracle reads, slippage checks, and fee accounting, can run into the hundreds of millions of cycles. The delta is not linear. It is closer to a step function.
Aggregation is the escape hatch, and it is the part most teams under-invest in early. You prove many blocks, then prove the proofs, and post a single verification on-chain โ a Groth16 or Plonk verifier lands somewhere around 250,000 to 300,000 gas depending on the curve and the stack. If you skip aggregation, you are paying full settlement gas on every proof forever, and your proving budget becomes a rounding error next to your verification budget.
I learned the shape of that curve the hard way. During the 2017 ICO cycle I audited ERC-20 implementations for a living, and the lesson then was that the bugs live in the code nobody reads. The lesson now is the same. In 2023 I sat with a team that had benchmarked its prover on a fresh GPU cluster and concluded proving cost would fall faster than user fees. It did not. The hardware improved and the workload improved faster โ every new precompile, every new opcode, every feature request that boiled down to 'we also support account abstraction and a DEX and a privacy pool' added cycles that had to be proven.
Proving cost is not a technology problem that gets solved once. It is a scope problem that gets reopened every time the product ships. Teams that treat proving as a fixed engineering milestone rather than a continuous budget line are the ones whose gross margins quietly invert.
The data-availability side has its own failure mode, and it is the mirror image.
A 1-wei blob base fee is not a stable equilibrium. It is a signal that demand is running below target. When demand exceeds target โ an airdrop claim window, a chain migration, a large mint, spam โ the blob fee mechanism adjusts aggressively, at twelve and a half percent per block against the target, and it compounds within a handful of blocks. In past congestion episodes the blob base fee moved from effectively nothing to hundreds of wei, and aggregate daily DA spend across the ecosystem ran into the hundreds of ETH. An operator with a fixed fee schedule โ and almost every rollup has one, because users punish variable gas โ eats that difference out of its own margin.
There is a deeper problem with pricing a shared resource at 1 wei: it removes the only signal that would discipline consumption. When DA is nearly free, the marginal cost of posting more state is zero, but the social cost is not. Operators that keep every historical state root, that post full batches instead of state diffs, that never prune, that store redundant witness data for convenience โ they push cost onto a market everyone shares, and no pricing mechanism corrects them, because the price is at the floor.
Then there is the revenue side, which is thinner than the token charts suggest.
A sequencer is a monopoly with an expiration date. It earns in two ways: the spread between what users pay and what execution costs, and the priority fee and ordering value captured from seeing flow before it lands. Both are compressing. Shared sequencing, based rollups that hand ordering back to L1 proposers, and the broader push toward separating block building from block proposing all move margin out of the operator and into whoever supplies the flow. A rollup that gives away sequencing gives away the only business it had, and the ones doing it are doing it because the alternative โ running proving and sequencing on a token-funded runway โ is worse.
This is where I apply the test my desk used during DeFi Summer in 2020. We asked one question of every farming protocol: is the yield coming from users, or from dilution? Protocols with real fee capture survived the unwind. Protocols whose yield was their own emissions did not. The same question applies to every rollup today, and most of them fail it. Strip token emissions out of the revenue line and you are left with a sequencer fee stream that, for all but three or four chains, does not cover proving, engineering, or a mid-level infrastructure salary.
Treasury accounting is where this becomes urgent rather than academic.
A treasury denominated in your own token is not a treasury. It is a buyback program that has not yet been recognized as one. Runways published in token units are marketing, not accounting โ convert the reserves to ETH or dollars at spot and the numbers halve, then halve again in a bad quarter. The operators with eighteen months of runway in stable assets will outlast the ones with thirty-six months of runway in their own governance token. This is not a controversial claim. It is arithmetic, and it is the reason the current drawdown will consolidate the sector whether or not anyone announces it.
Regulation will not arrive at the layer people expect, and that matters for all of the above.
The enforcement choke point in a rollup stack is not a set of four hundred validators. It is the single operator running the sequencer and the single company running the proving cluster. When travel-rule obligations, licensing regimes, and sanctions screening reach the L2 stack, they will land on those two entities โ because they are the only identifiable, subpoenable parties in the system. The result is a familiar pattern from my time auditing token sales: the compliance cost is passed entirely to people who use the front door, while anyone with a wallet holding a handful of addresses and a bridge route moves around it without noticing the rule exists. The honest user pays. The motivated user does not.
The same theater shows up in exchange proof-of-reserves disclosures, which remain timestamped snapshots of assets rather than continuous attestations of liabilities. A Merkle root proves what an exchange held at one block height. It says nothing about what it owed, or what it did in the fourteen hours after the snapshot was taken. That is not a reserve proof. It is a photograph.
So what do you actually watch, if you want to know which operator survives the next year?
Four numbers, none of which appear on a marketing site. Gross margin per transaction with token emissions excluded. Proving cost per proof, normalized by cycles executed, tracked over time rather than measured once. Blob spend as a percentage of protocol revenue during congestion weeks, not quiet ones. And the share of treasury held in assets other than the native token. If an operator will not show you those four numbers, the reason is generally that they are bad. The constraint is the story, and the constraint is a cost curve, not a roadmap.
There is one more structural question worth asking, and it is the one nobody wants to answer on the record: is there a fee switch, and can it be turned on without breaking the thing that makes the chain usable? A protocol with no mechanism to capture value from the activity it enables is a public good carrying a venture valuation. Those do not compound.
Here is the part that runs against the consensus, and I think it is the most important thing in this piece.
Cheap blobs are a subsidy to failure. A functioning market would have consolidated the rollup sector eighteen months ago. There are too many general-purpose chains chasing too few differentiated users, and the reason they persist is a specific stack of subsidies: DA costs at the floor, proving costs deferred onto a token-funded runway, and a token market that keeps refilling that runway. Had blobs stayed expensive, the marginal operators would have died, the survivors would have gained pricing power, and users would have ended up with fewer, better chains. Free data availability bought the industry time it has not used well.
The second inversion is that high proving costs may be a feature rather than a bug. Expensive proving forces architectural discipline. Teams batch harder. They prune state. They refuse to prove what does not need proving and move computation off-chain, proving only the assertion. When proving looks free, you get zkEVMs that try to prove everything and end up proving nothing economically. Scarcity is a design constraint that produces better systems than abundance does.
Which leads to the conclusion most cost-obsessed roadmaps miss: the winner will not be the cheapest prover. Proving cost is converging and commoditizing. What does not commoditize is switching cost โ the liquidity, integrations, and positions that already live on a chain. A cost advantage is a two-year lead. Distribution is a moat. Follow the cost curve, not the roadmap, but follow the liquidity when the two disagree.
Watch the re-congestion. The next airdrop season, the next migration, the next time blob demand pushes through the new target of six, the operators who built their models on a 1-wei assumption will find out what their gross margin actually is. The question is not whether your rollup is cheap today. It is whether it would still be solvent at five hundred wei per blob โ and whether anyone in the building has run that number.