Block 21,486,203. Blob base fee: 1 wei.
Block 21,486,215. Blob base fee: 1 wei.
Two weeks later, the same reading. Ethereum's blob fee market has spent almost its entire life pinned to the floor โ 1 wei per unit of blob gas, the hard minimum the protocol permits under EIP-4844's MIN_BLOB_GASPRICE constant. For anyone reading that chart, the argument is closed. Data availability is free. The DA wars are over. Every rollup roadmapped through 2026 is running on a cost basis of approximately nothing.
I priced the same load differently. I pulled 431 days of per-block blob gas consumption from an archive node, joined it against ETH spot and the London close on IBIT, and recomputed the identical DA workload in dollars instead of wei.
Glitch detected. Source traced.
The base fee never moved. The bill did.
EIP-4844, shipped in Dencun, did not make data cheap. It made data separately priced. That distinction is carrying a lot of weight right now, and almost nobody has audited what it implies.
Before Dencun, a rollup's data went to calldata. It consumed the same gas as a swap, competed for the same blockspace, and paid the same EIP-1559 base fee as everyone else. After Dencun, blob-carrying transactions โ type 3 โ draw on a second, parallel fee market with its own excess_blob_gas accumulator, its own target, and its own absolute floor.
The mechanism is EIP-1559 with a different exponent. Blob base fee equals MIN_BLOB_GASPRICE multiplied by e raised to (excess_blob_gas / 3338477). Below target, excess decays and the fee slides back toward 1 wei. Above target, excess accumulates and the fee compounds. Blobs are burned โ no priority fee, no tip, no validator cut. It is a pure protocol tax paid by whoever signs the blob transaction, which in practice means the sequencer.
That floor is the reason the chart looks dead. Between Dencun and the first genuine saturation episodes, blob demand sat below target almost continuously. The fee fell to 1 wei and stayed there for months. Press releases called it scalability. What it actually was: a market with a minimum price and no demand pressure.
EIP-7691, in Pectra, raised the target and the max. It did not touch the update fraction โ 3338477 stayed fixed. That detail matters more than the headline number, and I'll come back to it.
Here is the arithmetic nobody quotes.
Blob space is priced at the tail, not the mean. Each block at full saturation adds exactly one target's worth of excess_blob_gas, and the fee multiplies by e raised to (target / 3338477). Pre-Pectra, with a target of 393,216 blob gas, that multiplier came to 1.125 โ roughly 12.5% per block. Post-Pectra, the target doubled to 786,432 while the update fraction stayed put.
The per-block ramp is now approximately 26.6%.
Ten consecutive saturated blocks: 1.266 to the tenth power, roughly 10.6x. Twenty: about 105x. Forty: about 11,000x. This distribution is not a bell curve with a fat right tail. It is a floor with a cliff attached.
Put a number on it. A mid-size rollup posting 1,200 blobs a day pays, at 1 wei per blob gas, roughly 1.6e-10 ETH. Call it nothing. Move the base fee to 50 gwei and the same day costs 7.9 ETH. Move it to 500 gwei and it costs 78.6 ETH. At a $5,000 ETH, that is a jump from sub-cent to roughly $393,000, and the protocol delivers it in two minutes of full blocks. Not a quarter. Not a governance cycle. Ten blocks.
This is not theoretical. The March 2024 inscriptions wave pushed individual blob costs above $500 before decaying back to 1 wei within a day. Several airdrop claim windows did the same thing on a smaller scale. In every case the spike lasted hours, and in every case the rollups that got caught were the ones whose fee logic quoted the last observed blob base fee rather than a forward estimate.
Every rollup I have audited prices DA at the median. The sequencer estimates cost from a trailing window, adds a margin, publishes a gas price to the L2. The user pays that quoted price. If a saturation event lands inside the window, the sequencer eats the delta. It cannot pass it through retroactively โ the user transaction is already sequenced, and the blob is already posted. The loss is booked and invisible.
Liquidity draining. Logic broken.
There is a structural reason this cannot be fixed on-chain. The EVM can read the current blob base fee through the BLOBBASEFEE opcode. It cannot read the fee that will apply when the batcher actually submits, five to sixty seconds later. Blob gas itself is inaccessible to execution โ only the versioned hash reaches the contract. There is no hedging instrument, no futures market, no on-chain conditional. The only available control is a human-set threshold in a batcher config file.
Most of those thresholds, in the sample I pulled, were written in 2024 and never revisited after Pectra shipped.
Now the number that genuinely surprised me.
Blob base fee is denominated in wei per blob gas. Rollup DA cost in dollars is that fee, times 131,072 blob gas per blob, times the ETH price. Two of those three variables move on their own.
ETH travelled from roughly $2,400 in the post-Dencun trough into the $4,000-$5,000 band during the recent institutional inflow cycle. Blob base fee, across the same stretch, sat at its floor for the substantial majority of blocks. The gwei chart drew a flat line. The dollar cost of the identical DA load roughly doubled.
Exchange volume anomaly flagged.
This is the whole failure mode of the "DA is free" narrative. Rollup economics get quoted in ETH by people who think in crypto and reported in dollars to the institutions funding them. Those are different numbers, and the gap is driven by a variable no rollup controls and few hedge. A treasury holding ETH is exposed to blob base fee risk on one axis and ETH price on the other. The second axis has been the dominant one for eighteen months and it does not appear on any blob dashboard I have seen.
The demand side is about to stop being rollup-only. My model assumed rollups were the sole blob consumers. That assumption has roughly twelve months left on it. Inscriptions and blob-metadata protocols were the first non-rollup demand. Alt-DA attestation bridges are the second โ a rollup running execution data on Celestia or EigenDA still posts a commitment to Ethereum, and those commitments batch into blobs. Blob-embedded proofs, which several proving teams are prototyping now, are the third.
None of these are large individually. All of them are indifferent to blob base fee in a way rollups are not. A metadata protocol posting 0.2 ETH of blobs does not care whether that costs 0.2 ETH or 0.8 ETH; the commitment is the product. That is inelastic demand inside a market with a 1 wei floor and a vertical ceiling โ the exact shape that produces tail events rather than steady-state price discovery.
Set that next to Pectra's target increase and you get the configuration I flagged to my desk last quarter: more elastic supply of blob space, more inelastic demand for it, and a ramp rate that doubled while nobody updated their sequencer thresholds. Payment-focused rollups are the most exposed, because basis-point spreads on stablecoin transfers do not survive a 10x DA move โ DA is a larger share of their unit economics than it is for a general-purpose chain, and their cost base assumed floor pricing forever.
The consensus view is that Celestia, EigenDA, Avail, and Ethereum blobs compete on price, and Ethereum is losing on the spot number. That framing collapses under a security-model audit.
Blob data posted to Ethereum inherits Ethereum's consensus and its sampling roadmap. Blob data posted to an alt-DA layer inherits that layer's validator set and that layer's assumptions โ and then the rollup posts a bridge contract to Ethereum anyway, which is where the real trust assumption lives. The two products are not the same good at different prices. They are different goods, and the market has been comparing stickers.
The genuine competition is not on price per byte. It is on the price of the tail. An alt-DA layer charging fixed fees is selling variance insurance as much as it is selling bytes, and almost no rollup has modelled the two against each other.
The second blind spot is the update fraction itself. When EIP-7691 doubled the target without adjusting 3338477, it doubled the speed at which the blob fee can run away from the floor. Nobody wrote a governance post about that, because the parameter that changed and the parameter that governs acceleration are not the same parameter. A fee market with a 1 wei floor and a 26.6% per-block ramp is not a cheap market. It is a market that looks cheap from the median and behaves like a sealed-bid auction during the only hours that matter.
Stop reading the blob base fee median. Read the 95th percentile over thirty days, converted to dollars, and then go check whether your batcher threshold was written before or after Pectra.
The next saturation event will not be an inscriptions wave. It will be a product launch, an airdrop, or an institutional onboarding window that somebody scheduled without checking what else was queued that hour.
The question worth asking is not whether blobs are cheap. It is who is holding the variance when they are not โ and whether that entity has any idea it signed up for the position.
