There is a number in the Ethereum protocol that has spent most of its life equal to one. Not one dollar, not one gwei โ one wei, the smallest unit the network can express, a quantity so small that it functions less as a price than as a placeholder for the absence of scarcity. It does not appear on the dashboards that track total value locked. It does not trend. It does not move when a large holder sells into weakness, and it does not appear in the quarterly letters that institutional allocators circulate to their investment committees. Yet for an entire class of businesses โ the rollups that now settle a majority of retail crypto activity โ this number is the largest single variable in their cost structure.
The number is the blob base fee. For most of the two years since Ethereum activated EIP-4844, it has sat at or near its floor, and the floor is one wei. But on a handful of occasions the floor has lifted, briefly and violently, and each time the reaction from the market has been the same: a shrug. Fees rose for a few hours. Transactions became marginally more expensive. Then the auction unwound and the number returned to one, and everyone went back to describing layer 2 blockspace as effectively free.
I want to argue that this shrugging is a mistake โ that the blob market is the clearest example in this cycle of a structural risk concealed inside what looks like a structural improvement, and that a bear market, when subsidy budgets are shrinking and attention is elsewhere, is precisely the wrong moment to be relaxed about it.
The Context: What Dencun Actually Did
To understand why the blob fee matters, it helps to remember what rollups were paying before March 2024. Every rollup has to do two things: execute transactions off-chain, and then publish enough data to layer 1 that anyone can reconstruct the state and challenge a fraudulent or invalid claim. Before EIP-4844, that publication step meant writing raw calldata into the Ethereum block, which meant competing for the same scarce execution gas as every swap, mint, and liquidation on the network.
The consequence was a peculiar inversion. A rollup's own compute was cheap โ it was running on commodity hardware, often in a data centre the team controlled โ but its data availability bill was denominated in the most congested fee market in the industry. Rollups were, in effect, wholesale buyers of the most expensive commodity in crypto, and they were buying it at retail auction prices that spiked precisely when they needed it most. During the 2021 and 2022 congestion episodes, calldata costs routinely consumed the overwhelming majority of a rollup's operating expenditure.
EIP-4844, activated with the Dencun upgrade, introduced a separate resource. Blob-carrying transactions attach a fixed-size chunk of data โ a blob โ to a block, priced by its own auction, inaccessible to the EVM, and pruned from nodes after roughly eighteen days. The design intent was explicit: give rollups a cheap, purpose-built data lane, and let the execution gas market serve the applications that actually need computation.
It worked, and it worked so well that it became the foundation of a narrative. Rollup fees collapsed. Within weeks, the cost of a layer 2 transaction fell by an order of magnitude or more across the major networks, and the framing that took hold in retail discourse was that layer 2 blockspace had been permanently re-based to something close to zero. The subsequent Pectra upgrade raised the blob target from three per block to six, and the maximum from six to nine, which only deepened the impression. More supply, same demand, lower price. The direction of travel looked obvious.
But here is where I want to slow down, because the framing embeds two assumptions that were never tested. The first is that the blob market behaves like a normal commodity market, where supply and demand meet on a slope and prices drift. The second is that a subsidy-driven level shift in price is the same thing as a change in the underlying cost curve. Neither assumption survives contact with the mechanism.
Listening to the Silence Between the Data Points
The blob fee market is built on an EIP-1559-style auction, but the parameters are far more aggressive than the ones most analysts are used to reasoning about, and the consequences are counter-intuitive.
Ethereum targets a number of blobs per block. When usage sits at or below that target, the base fee drifts down. When usage exceeds it, the base fee rises โ but the rate at which it rises is governed by an update fraction that permits roughly a twelve percent increase per block, and blocks arrive every twelve seconds. On the way down, the same mechanism allows a roughly equivalent decay per block whenever usage falls back below target.
Sit with the asymmetry for a moment. Twelve percent compounding per block means the fee can multiply by a factor of ten in about twenty blocks, or roughly four minutes, and by a factor of a thousand in under an hour of sustained saturation. It also means the fee can collapse back toward the floor almost as quickly once the congestion passes. The blob fee market does not have a slope. It has a step, and the step is climbed at compounding speed.
This is why the "one wei" observations are misleading as a measure of risk. A market that spends ninety-nine percent of its time at the floor and one percent of its time at a hundred times the floor is not a cheap market with occasional blips. It is a market with extreme variance and a very low median, and variance is a distinct hazard from level. A business that can plan around a high but stable input cost is a fundamentally different business from one that has to plan around an input cost that is negligible for thirty days and material for six hours.
In my own work building cost models for rollup operators, this is the number I have learned to watch least and think about most. The average blob fee tells you almost nothing. The distribution tells you everything, and specifically the thickness of the right tail.
What Dencun Repriced, and What It Did Not
Here is the part of the story that the fee-collapse headlines obscured. The blob fee was never the only component of a rollup's layer 1 cost, and it was arguably not even the most important one after Dencun.
When a rollup posts a batch, it pays several things. It pays blob gas, which after Dencun became nearly free. It also pays execution gas for the layer 1 transaction that carries the blob commitment, the cost of verifying proofs on layer 1, and โ depending on design โ the cost of any state roots or bridging messages posted outside the blob. These components did not disappear when blob pricing was introduced. They were simply re-based to a smaller share of a much smaller total.
That has an important consequence for anyone modelling where fees are going. When blob gas costs nearly nothing, the binding constraint shifts to everything else. A rollup that settles frequently pays for proof verification and layer 1 execution gas on a schedule it controls. A rollup that batches aggressively amortises those costs across more transactions but accepts longer finality. The compression algorithms that teams spent eighteen months optimising โ the zero-byte encoding tricks, the signature aggregation schemes, the state diffs โ they remain the largest lever on per-transaction cost, and they are the levers that a cheap blob fee did not touch.
So when I read that layer 2 fees fell by ninety percent after Dencun, I read it as a one-time re-basing rather than a trend. The post-Dencun fee level was a level shift, not the beginning of a curve, and treating it as a curve is the analytical error that the current cycle will punish.
I have made a version of this mistake before. In 2020, while dissecting the risk parameters of Aave during the DeFi Summer, I spent six weeks modelling the systemic fragility of over-collateralised lending under volatility, and I built my conclusions on the assumption that the incentive structure I was observing would persist long enough to matter. It did not. The mechanism was sound; the subsidy underneath it was not permanent. What survived my analysis was the structural finding and nothing else: when an input is priced below its cost of provision, something else is paying, and that something else eventually stops. The blob market is the same shape of problem with better plumbing.
The Elasticity Illusion
Advocates of the cheap-blockspace thesis lean on a supply argument. Blob capacity has already been expanded once, with Pectra, and further expansion is on the roadmap. PeerDAS and the data-availability sampling work that accompanies it are designed to let nodes store only fragments of blob data, which is what allows the blob count to rise again without imposing an impossible bandwidth burden on validators. Proposals circulating around the Fusaka-era configuration have discussed targets in the range of ten to fifteen blobs per slot, depending on how the sampling implementation performs in testnets.
That is a real and meaningful expansion, and it is the strongest version of the bull case for cheap layer 2 fees over a multi-year horizon. But it is a flattering framing of a specific problem, and the problem is timing.
Blob capacity expands on a protocol upgrade cadence: months of development, devnets, testnets, client coordination, a hard fork. Blob demand expands on an application cadence: a launch, an airdrop, an inscription surge, a new rollup onboarding, a viral contract. The two schedules have nothing to do with each other, and the asymmetry runs in the wrong direction. Demand can double in a week. Supply doubles roughly once a year, if that.
This is why I have written, and continue to write, that blob space will be saturated within roughly two years of any given expansion, and that the industry's response will be a doubling of rollup fees rather than a graceful glide. Not because the engineers will fail โ the engineers have been remarkably competent โ but because the demand curve is driven by speculative attention, and speculative attention has a much shorter cycle than a hard fork.
And in a bear market, the composition of that demand changes in a way that makes the tail risk worse rather than better. Speculative activity thins, but the activity that remains is disproportionately the activity that consumes the most blockspace per unit of economic value: airdrop farming campaigns, points programmes, sybil-resistant proofs, state-heavy gaming interactions, and the long tail of incentive-driven volume that exists only because a token is being distributed. Cheap liquidity and cheap blockspace arrived in the same window, and the two are constantly mistaken for one another. The separation matters, because one of them is policy-driven and reversible, and the other is a fixed commodity on a twelve-second clock.
Where the Subsidy Actually Sits
This brings me to the question that I think most layer 2 token models have not answered honestly, and it is a question that has become acute now that the market is in a drawdown.
If blob fees rise materially, a rollup has three options. It can pass the cost to users, which is politically expensive because the entire value proposition of the network has been built on the promise of cheap transactions. It can absorb the cost out of sequencer revenue and operator margin, which works until the margin is gone. Or it can absorb the cost out of token issuance โ that is, pay for blockspace with a subsidy funded by dilution, and hope the market rewards the growth.
That third option is the one that gets chosen, because it is the one that preserves the narrative, and it is structurally identical to the mechanic I have spent years criticising in the liquidity mining context. A yield that is paid out of a treasury is not a yield the protocol has earned; it is a transfer. A fee that is suppressed by token issuance is not a fee the network can sustain; it is a marketing expense. The test is simple and rarely applied: turn off the subsidy and see which users remain, and then ask whether those users are enough to cover the cost of serving them.
What makes the current environment unusually unforgiving is that the subsidy itself is being repriced. In a bear market, treasuries are worth less in hard currency, emissions are worth less to recipients, and the cost of capital for anything that depends on continued fundraising has risen. A rollup that could credibly fund cheap blockspace out of a rising token in 2024 is making a much harder calculation in the present environment. This is not a prediction of collapse. It is a prediction of repricing, which is a slower and more confusing event, and one that tends to be discovered in the composition of revenue rather than in the headline fee.
The Contrarian Case: Cheap Blobs May Be Bad for Rollups
The consensus reading is that cheap blobs are unambiguously good for layer 2s. I think the opposite is at least as defensible, and I want to state it plainly because I have not seen it argued with much conviction.
When an input is priced near zero, it stops being a source of differentiation. Before Dencun, the teams with the best compression and the best settlement cadence enjoyed a genuine cost advantage, and that advantage showed up in fees users could see. After Dencun, everyone's data availability bill approached zero, and the marginal cost advantage of operational excellence largely vanished with it. What remains as differentiation is distribution, ecosystem, and brand โ which is to say, the things that are hardest to build and least amenable to a roadmap.
There is a second, subtler effect. Navigating the paradox of decentralised trust has always required accepting that value in these systems can accrue one layer away from where it is generated. If blob space is abundant and cheap, the network that supplies it captures very little. If blob space becomes scarce and expensive, the network that supplies it captures a great deal. The direction of that transfer is not a detail of tokenomics; it is the central question of whether layer 2 tokens are claims on fees or claims on attention. Cheap blobs push the answer toward the latter.
And there is a governance layer that I have found almost nobody is willing to discuss, which is what happens to the entities that must vote on these transitions. Many of the protocols in this sector, including several infrastructure and layer 2 projects, operate with no legal wrapper at all. The token holders and the multisig signers who approve a change to fee structure or treasury policy may believe they are voting on software. In several jurisdictions, the analysis is considerably less comfortable, and the personal exposure of the individuals who execute the decision is not limited to the value of their holdings. Unmasking the vacuum behind the hype is usually framed as a critique of valuations. My concern is narrower and more concrete: when a fee transition forces an economic decision, it also forces a legal event, and the participants in that event are frequently unaware of the exposure they have accepted.
I raise this not to frighten but because the ethical friction is real and it sits exactly where the industry prefers not to look. Efficient markets are extremely good at pricing assets and extremely poor at pricing responsibility. The costs that are not on the balance sheet are still paid.
Takeaway: Three Things to Watch
As this cycle works itself out, I will be watching three things, none of which appear on a price chart.
The first is the shape of blob usage, not its average. A market that sits at the floor for months and spikes for hours is telling you something about how much of the demand is episodic and incentive-driven. The right tail of the blob fee distribution is the honest measure of how fragile the cheap-fee model is.
The second is the composition of layer 2 revenue, split between fees actually paid by users and value transferred from token issuance. The teams that begin disclosing that split are the ones I will trust, and the ones that never do are answering the question by avoiding it.
The third is the settlement cadence of the largest rollups. If operators begin batching more aggressively and accepting longer time-to-finality in order to save on blob and proof costs, that is the market discovering the new cost curve in real time โ quietly, without an announcement, and months before anybody writes the post-mortem.
Peering through the haze of speculative value, the question worth holding onto is not whether layer 2 transactions will remain cheap. It is whether cheapness was ever a product, or whether it was a subsidy that has not yet been invoiced โ and how a market accustomed to the first answer will behave when it is handed the second.