Over the past thirty days, the median blob base fee on Ethereum has spent most of its nights at one wei — the floor, the arithmetic minimum, the price of the most secure data availability layer ever built, rounding to zero. Over that same thirty days, the aggregate bridged value across the ten largest rollups has continued to bleed, down double digits from the cycle high, and at least four mid-tier chains have watched more than forty percent of their liquidity providers walk out the door since autumn.
Two numbers. Opposite directions. Same market.
That divergence — not the ETF flows, not the stablecoin hearings, not whichever sovereign wealth fund has decided this quarter that Bitcoin belongs on its balance sheet — is the story I keep coming back to. The quiet thing. The empty blob.
I noticed it on a Tuesday, somewhere past two in the morning Melbourne time, refreshing a Dune dashboard I had built three years ago and had mostly stopped opening. The chart was flat. Not low. Flat. A line drawn along the bottom of the axis, like a pulse that had stopped.
To understand why a flat line matters, you have to remember what a blob is, and what it was supposed to become.
EIP-4844 shipped with Dencun on the thirteenth of March, 2024 — the largest change to how Ethereum prices its own blockspace since EIP-1559 rewrote the fee market three years earlier. A blob is 128 kilobytes of data attached to a beacon block, verified by the consensus layer, available for roughly eighteen days, and then pruned. Not execution. Not state. Data. A sidecar, a receipt, a whisper. Consensus nodes promise they saw it, and the chain moves on.
Each block carried a target of three blobs and a maximum of six. The pricing mechanism was deliberately borrowed from EIP-1559: an independent blob base fee that rises exponentially when usage exceeds the target and decays when it doesn't, with a hard floor of one wei. The design intent was explicit, and in retrospect, prophetic. Blobs were meant to be cheap. Permanently, structurally, aggressively cheap.
The market learned this the hard way. On the twenty-seventh of March, 2024, two weeks after Dencun activated, someone deployed a contract that wrote inscription-style data into blobs — Blobscriptions — and the blob base fee detonated from one wei into the hundreds, briefly making blobspace more expensive per byte than the calldata it had just replaced. The spike lasted hours. It was the first demonstration of a pattern that would repeat: blobspace is not a market with a price. It is a market with a floor and occasional violence.
Then the supply started growing. Pectra, in May 2025, raised the target to six blobs and the ceiling to nine through EIP-7691 — a doubling of capacity delivered not as a fee reduction but as an admission that the original sizing had been too timid. Fusaka followed at the end of 2025 with PeerDAS, which let nodes sample blob data rather than download all of it, unlocking a roadmap toward double-digit targets and, eventually, throughput that stops being measurable in kilobytes per block at all. Chasing the myth through the ledger's fog, you can watch the protocol quietly conceding that it had been pricing data for a market that hadn't arrived yet.
Meanwhile, on the demand side, three well-capitalised competitors — Celestia, EigenDA, Avail — began selling the same commodity at a fraction of the cost, bundled with the one thing Ethereum's blob market could never offer: a denominated, predictable cost curve. And underneath all of it, the bear market did what bear markets do. Rollup tokens down eighty and ninety percent. Treasuries denominated in their own collapsing assets. Sequencers running at margins that, on a bad week, go negative. That is the terrain. Now the mechanism.
The reason the median blob fee sits at the floor is not that nobody wants blobs. It is that the mechanism is designed to ensure nobody has to pay for them.
Watch the update rule. When average blob usage in a block exceeds the target, the base fee multiplies by a factor slightly above one; when usage falls below, it multiplies by the reciprocal. Same exponential logic as EIP-1559, applied to a narrower commodity with a far lower tolerance for surplus. The moment demand pushes past the target, the protocol does not queue transactions the way a blockspace market would. It expands supply toward the ceiling — six, then nine, then whatever the next fork permits — and the price collapses back to the floor within a handful of blocks.
I have watched this happen in real time perhaps a dozen times since Dencun. A data-heavy application ships an airdrop claim. For four minutes, the blob fee goes parabolic. For the next forty, it decays geometrically, and by the time the dashboard refreshes it is a one-wei line again. The spike is real. It is also ephemeral by construction.
Which means the revenue line for Ethereum's base layer is not a function of demand growth. It is a function of how often demand exceeds a ceiling that the protocol keeps raising. That is a very different asset from the one people were pricing in the spring of 2024.
Run the arithmetic yourself. Before Dencun, rollups paid for L1 calldata, and that calldata was burned under EIP-1559. After Dencun, blob fees are burned too — but the burn is a rounding error. A single congested block during a hyped NFT mint used to destroy more ETH than the entire blob market burns in a month. The ultrasound-money thesis did not die of a hostile argument. It died of a spreadsheet.
This is where I have to be honest about something I got wrong and something I got right. When I began dissecting ICO whitepapers in 2017 — tracing the ghost in the whitepaper's code, sentence by sentence, hunting the gap between the promise and the mechanism — the useful skill was never finding the lie. It was finding the sentence that everyone read past. In the Dencun documentation, that sentence was the one explaining that blobs would be priced to be cheap. Cheap was the goal. Cheap was the point. And cheap means nobody on the base layer gets paid.
Which brings us to the part of this bear market that rollup teams would rather not discuss.
The Dencun sales pitch was simple: cheaper data availability means cheaper rollups, and cheaper rollups mean more users. That pitch was directionally true and operationally useless, because it confused a marginal cost with a total cost.
A rollup's cost structure has at least four lines. There is data availability, which blobs crushed. There is the prover or the sequencer, which blobs did nothing for at all. There is the bond capital — the challenge-period collateral that optimistic rollups lock up, or the proving infrastructure that zk rollups amortise across a fixed hardware footprint. And there is the cost of capital on the treasury itself, which in this market is the line that actually kills you, because the treasury is denominated in a token that has lost most of its value.
Blobs lowered the one line that was already shrinking. The line that is fixed — the prover, the hardware, the audits, the twelve-person team in a co-working space in Berlin — did not move. In a bull market that distinction is invisible, because sequencer revenue covers everything. In a bear market it is the only thing that matters, and it produces a specific, ugly pattern: rollups that look cheap on a per-transaction basis while burning eight figures a year of their own token just to stay online.
This is the correction to the standard narrative. EIP-4844 did not make rollups cheap. It converted their cost base into a shape that punishes anyone who modelled only the average.
Because the average is not what you pay. You pay the tail. On a dull Tuesday, your rollup's data cost is indistinguishable from zero — a few hundred dollars a month, a line item nobody reviews. On the day a large application migrates, or an inscription wave hits, or three chains settle proofs in the same ten-minute window, you pay the ninetieth percentile, and the ninetieth percentile in a floor-priced market is not a smooth number. It is a spike. Capacity expansions like Pectra and Fusaka do not flatten that distribution. They widen it. More headroom means the average gets cheaper and the peaks get taller, because the entire point of extra supply is to absorb the shocks that used to queue and get priced.
So when I say that blob data saturates within two years and that rollup gas fees double again, I am not forecasting a traffic jam. I am forecasting arithmetic. The 2024 capacity was sized for a world where the only buyers were rollups. The next wave of buyers — proof aggregation networks, data-only chains, attestation layers, and whatever the inscription people invent next — are not price-sensitive the way a retail rollup is. They are building on blobs precisely because blobs are the cheapest verifiable data in existence, and they will happily pay a hundred times the floor during the windows when they need it. When their volume gets large enough, the median stops printing one wei. Not because demand is infinite, but because a market with a floor and a ceiling has exactly one behaviour when it fills: the price does not drift up. It jumps.
To see who actually bleeds, it helps to sort the buyers of blobspace into groups, because they behave nothing alike.
Rollups are the sticky demand. They will not leave, because leaving means surrendering Ethereum's settlement guarantees, and that is the entire product. They are also price-insensitive within a wide band — a doubling of data cost is survivable for a chain with real users. Their problem is that the sticky part of the demand curve is small, and it is the part that generates the least fee revenue for the base layer.
The arbitrage buyers — the data availability protocols — purchase blobspace when it is cheaper than their own capacity and exit the moment it is not. This is demand that appears in charts as a perfectly flat line punctuated by nothing, because it is designed to be reactive. It adds variance without adding revenue.
Then there is spike demand: inscription waves, on-chain art experiments, the descendants of Blobscriptions. In 2021, when I minted the Melbourne Memories collection and buried long-form essays about gentrification inside the metadata, I was making a small bet that data attached to a token could carry cultural weight rather than speculative weight. That bet paid off for twenty-one pieces of art. It is not a business model. Speculative data costs almost nothing to initiate, it is narratively driven, and it is the fastest-moving thing in the market — and it is also the first thing to evaporate when the market turns.
The fourth group is the one I find most interesting and most under-discussed: proofs. Aggregate proofs, attestations, oracle reports, cross-chain message commitments. Small payloads, high frequency, absolutely intolerant of unavailability. This is demand that grows with the ecosystem rather than with its speculation, and it is the demand that will still be here when the inscriptions are gone.
Now run the bear market filter across all four. The rollups are bleeding slowly and cutting headcount. The arbitrage desks are fine, because arbitrage thrives on volatility and a bear market is nothing if not volatile. The speculative layer has largely evaporated. The proof layer is quiet, small, and growing.
Notice what the filter reveals: the demand that survives a winter is the demand that was never paying attention to price. That is a hard lesson for anyone who built a fee model on the assumption that usage scales with cost. Usage scales with need, and need in this industry is a much colder thing than the marketing suggests. Alchemy in the age of open protocols, as ever, is the art of selling a narrative to people who need a mechanism.
Which brings me to the narrative I would like to see buried before the next cycle digs it up.
For two years now, the industry's most expensive conference panels have been dedicated to the problem of liquidity fragmentation — the idea that capital scattered across a dozen rollups is a market failure requiring a solution, and that the solution is, conveniently, a new protocol with a new token and a fee switch. Intent-based bridges. Solver networks. Shared sequencers. Chain abstraction as a product category.
Here is the part that never makes it into the deck. Measure the actual depth. The ten largest pools on Ethereum mainnet still hold more liquidity than the aggregate of every pool on every rollup combined, and the spread between the same asset on two different L2s closes in seconds, because the arbitrage is profitable enough that someone always takes it. Fragmentation is not making trading expensive. It is making trading mildly inconvenient — and inconvenience is precisely the kind of problem that gets solved by charging a fee, which is why it gets marketed as a crisis.
Liquidity fragmentation is not a market failure. It is a market. And the people selling you the fix own the toll booth.
The winter is already doing the sorting. Chain abstraction protocols raised nine-figure rounds on the fragmentation thesis and are now watching the bridges they were built to replace process the same volume at a tenth of the cost. That is what happens when your product is a markup on a problem that solved itself.
There is a second narrative worth pressing on, though it is less a battle than an obituary. Bitcoin's ETF wrappers did for the asset what securitisation did for mortgages: they made it legible to a class of allocator that had never touched a private key and never wanted to, and in doing so they moved price discovery from a network of cypherpunks to a network of portfolio models running on quarterly rebalance schedules. In this drawdown it shows up as something the 2017 cohort would find unrecognisable — a Bitcoin that trades with the beta of a growth stock, with a bid that appears and disappears alongside the macro calendar. The peer-to-peer electronic cash is still in the code. It just is not in the market anymore. The echo of a promise unkept, written in a ledger that has no memory of what it was supposed to mean.
So what should you actually watch in the next two quarters, if the question you care about is which protocols survive?
Not the average blob fee. It will keep printing one wei and telling you nothing, because a floor is not a signal. Watch the ninetieth percentile. Watch how many blocks per week land above the target, because that count is the only honest measure of whether the cheap-data thesis has run out of room.
Watch prover economics rather than data economics. Watch which rollups are paying for zero-knowledge proofs in a token they cannot sell without cratering their own market cap — that is the ledger of who is genuinely solvent and who is renting survival quarter by quarter.
And watch the demand that does not care about price. Proofs, attestations, message commitments, the unglamorous plumbing that keeps growing while everything speculative drains away. In a bear market the useful question is never who has the best story. It is who would still be paying for blockspace if the story were removed.
The blob market asked that question of an entire industry and received a mostly silent answer. Thirty days of one-wei nights is what silence sounds like when you chart it. The interesting part — the part nobody has priced — is what happens on the first night the floor breaks, and someone, for reasons that have nothing to do with speculation, decides they need that space badly enough to pay for it.
Ethereum built a market designed to be free. The next two years will test whether free was ever the same thing as cheap.
