Blob base fee: one wei.
That is the floor — the minimum the protocol permits — and in every sample I have pulled of mainnet blocks this month, more than nine slots in ten have closed there. Ethereum's data availability market is not merely cheap. It is unpriced, because it is unsold.
Dencun shipped EIP-4844 on March 13, 2024, and handed rollups a dedicated data channel: blobs, 128 KB each, three per block targeted, six maximum. Pectra raised the ceiling again on May 7, 2025, to six targeted and nine maximum. That is roughly 1.1 MB of block space per twelve-second slot at the top end — about 96 KB per second of pure rollup data throughput, priced by its own fee market, burned separately from execution gas.
The sell-side story wrote itself overnight. Rollup fees collapsed. A swap on an OP Stack chain that cost $0.30 in February 2024 cost roughly $0.01 by April. Analysts called it the largest UX unlock in Ethereum's history. They were right about the fees. They were wrong about almost everything downstream.
Context
Here is what the blob market actually looks like from a surveillance desk.
Since Dencun, blob base fees have cleared at the 1-wei minimum in the overwhelming majority of slots. The exceptions are a handful of spikes — and when I trace back to what caused them, it was never broad rollup demand. It was a single actor flooding blobspace: inscription minters, a searcher stress-testing the fee curve, one chain performing a bulk state migration. In each case the fee curve spiked for a few hours, then decayed back to the floor within a day. That is not a shortage. That is a market with one buyer who occasionally walks in, purchases everything, and leaves.
The supply side, meanwhile, only got more generous. Pectra doubled the target. Fusaka and PeerDAS will push blob count higher again. Ethereum is running a scaling schedule that assumes demand it has not yet observed.
Against that backdrop, a parallel industry raised capital on the opposite premise. Celestia, EigenDA, Avail, and a long tail of modular DA networks sold a thesis: Ethereum blobs will be scarce, expensive, and unreliable, so buy dedicated data availability instead. Restaking collateral, modular settlement layers, and DA token launches turned that thesis into a multi-billion-dollar asset class. Several chains took the deal — Manta migrated to Celestia, Mantle purchased EigenDA capacity. The announcements were written as cost-savings stories.
They were cost-savings stories in 2023. In 2026, they are marketing.
Core
Let me put numbers against the thesis.
A post-Dencun OP Stack chain posting a single blob every twelve seconds — continuous, no gaps, no downtime — consumes one third of Ethereum's pre-Pectra target and one sixth of the current one. To saturate the blob target today, you would need somewhere between four and six high-volume rollups running at peak simultaneously, every block, indefinitely.
There are not four to six rollups running at peak. There is Base, and then there is everyone else.
Aggregate blob demand across the ecosystem has consistently run well below target capacity, and the market's own price discovery confirms it — a commodity trading at the protocol floor is a commodity nobody is bidding for. Every sustained blob fee expansion since Dencun traces to a single non-rollup actor, not to a broad-based rise in rollup throughput.
Now follow the cost line, because this is where the DA narrative quietly dies.
For an optimistic rollup in 2023, L1 calldata was roughly 80 to 90 percent of operating cost. Posting a state root was cheap; posting the batch data was ruinous. That is why the market believed DA was the binding constraint — because for eighteen months, it was.
Post-blobs, that line item collapsed. For a typical OP Stack chain, DA is now a low single-digit percentage of operating cost. Compressed batches are nearly free to publish because blobspace is nearly free to consume. The cost center moved. What remains is sequencer infrastructure, L2 execution, L1 verification, and the rapidly growing bill for proving — which for ZK rollups is now both the dominant expense and the dominant latency source.
So the pitch for alt-DA reduces to this: buy our product because Ethereum's version of it might become expensive someday. That is a hedge, not a need. And hedges get repriced when the underlying never moves.
I have watched this pattern before. In 2020 I logged every gas fee and slippage error from the Curve-to-Sushiswap stablecoin arb, and the lesson was not that the trade printed. It was that the trade printed because a temporary inefficiency existed, and it stopped the instant the inefficiency closed. Modular DA is the same shape. It is a spread against Ethereum blob pricing. The yield was sweet, but the exit was sharper — and in this case the spread closed at one wei.
There is a structural problem no DA provider solves, and a second one no one wants to name. Rollups are not bottlenecked on data publication. They are bottlenecked on proving, on sequencing, on forced-inclusion guarantees, on cross-chain settlement, and on the intent-routing layer that quietly relocated MEV extraction off-chain and into solver auctions. Data availability is the one part of the stack that actually got fixed.
The unnamed problem is worse. EIP-1559's blob market adjusts by a factor of 1.125 per block above target and decays symmetrically below it. That mechanism snaps back to the floor within hours of any spike. And because blob fees are burned, near-zero blob fees mean near-zero ETH burn from rollups — which means the base layer's own revenue depends on rollups consuming more blobs, not fewer. Ethereum has a direct economic interest in keeping DA cheap forever. You cannot sell a competing DA product against an issuer that is structurally motivated to undercut you.
Contrarian
Here is the angle nobody on the modular conference circuit wants to run.
The DA layer is not an infrastructure category. It is a financing structure. The scarcity claim was never derived from observed blob utilization. It was derived from a 2023 fee chart and projected forward without a demand curve attached. Token launches need a scarce resource to price. Proving costs are difficult to tokenize. Sequencer decentralization is regulatory and political, not economic. Cross-chain messaging has no natural token. DA is legible, it is quantifiable in megabytes, and in 2023 it was genuinely expensive. That made it the ideal narrative to raise against.
But you cannot sell data availability to a market whose data availability is free. Listen to the whispers, but trust the ledger — and the ledger says blob utilization, ex-spikes, sits in the low double digits as a share of target capacity.
The real question is not whether alt-DA is technically inferior. In several dimensions it is better: higher throughput, faster finality assumptions, cheaper at genuine scale. The question is whether anyone needs those dimensions in 2026. For a rollup doing five transactions per second, no. For Base at peak, marginally. For the four to six chains that would need to saturate Ethereum's target in concert, entirely hypothetical.
Chaos is just data waiting for a pattern. This pattern is a supply glut wearing a shortage's name tag.
Takeaway
Watch blob utilization, not blob announcements. Two falsifiable signals matter: sustained blob base fees above the 1-wei floor across a full week of slots, and at least four distinct rollups averaging more than one blob per block. Neither has occurred.
Until they do, the modular DA trade is a leveraged short on Ethereum's own adoption curve — a bet that the base layer fails to scale while you sell the alternative. In a bear market, that is not a hedge. That is a directional position with a friendly name, and it is the kind of trade that looks safe right up until the counterparty publishes a roadmap.
I run a 24-hour desk. In a twenty-four-hour cycle, sleep is a liability — but so is paying for capacity you will never use.