EIP-4844 gave rollups a data lane of their own, and with it a separate market for space. Before the change, a rollup’s batch data competed with ordinary mainnet traffic for the same gas. After it, the batch pays a blob fee that moves on its own curve. That decoupling is why a rollup can quote a fee in fractions of a cent while mainnet gas does something else, and why that fee can rise while mainnet is quiet.
What blobs actually changed
EIP-4844 introduced a data type that rollups post to Ethereum and that the protocol prices separately from execution gas. The headline effect is a discount. The structural effect is that data cost became a market with its own supply and its own fee, and that fee responds to how full recent blocks were.
For a rollup, that turns data cost into a variable it can influence but does not set. When the lane is quiet, posting a batch is cheap and the user fee can sit near zero. When several rollups want the lane at once, the price of posting rises and the operator has to decide what to do about it.
Two levers, and both reach the user
An L2 facing a higher blob fee has two options.
The first is to pay. The rollup absorbs a higher data cost and either narrows its margin or raises the fee it charges. Most rollups set that fee by policy rather than by auction, so pass-through is delayed and lumpy. Users see a change days after the blob market moved.
The second is to post less often. A batch can cover more transactions if the operator waits longer before posting, which spreads one expensive posting across more user activity. The cost of that choice is time. The data behind your transaction reaches mainnet later, and the window in which your L1 claim depends on the operator’s word gets longer.
Neither lever is visible in a fee number by itself. One shows up as a higher fee, the other as a cheap fee on a chain that settles more slowly.
The fee page is the output, not the input
Which lever a rollup pulled is partly readable in the fee it publishes. Ethereum.org’s L2 directory, in its 1 to 6 October 2026 snapshot, listed Base at $0.001 to $0.002 per transaction and Arbitrum One at $0.004 to $0.005, while OP Mainnet displayed $0.00, Scroll showed $0.002 to $0.003, Starknet $0.005 to $0.008 with gas paid in STRK, and Linea about $0.024.
A spread that wide is not explained by blob fees alone: fee policy, proof system, batch cadence and payment asset all differ. What the comparison does establish is that the user-facing price is a decision layered on top of a data cost, not a readout of it.
Scale is the other half. The same directory, in its October 2026 reading, put Base at $14.6B in value and Arbitrum One at $12.3B, with OP Mainnet at $1.65B, Starknet at $389M, Linea at $352M, ZKsync Era at $227M and Scroll at $45.7M. A larger chain spreads a fixed posting cost across more transactions, which is one reason the biggest names quote the smallest per-transaction numbers.
| Chain | Fee per transaction, Ethereum.org, 1 to 6 October 2026 | What the figure does not show |
|---|---|---|
| Base | $0.001 to $0.002 | Batch cadence behind the fee |
| Arbitrum One | $0.004 to $0.005 | Length of the dispute window |
| OP Mainnet | $0.00 displayed | Whether the fee is subsidised |
| Scroll | $0.002 to $0.003 | Thin flow under a small base |
| Starknet | $0.005 to $0.008, paid in STRK | Token price exposure |
| Linea | About $0.024 | Why it sits at the top of the range |
What the fee number does not carry
Three things sit behind a published figure.
Batch frequency. A cheap fee can come from posting rarely rather than from cheap data. Cadence is the number that tells you which, and it is usually not on the fee page.
The exit. A fee is the cost of entering an economy, not of leaving it. Base and OP Mainnet withdrawals to mainnet take 7 days, Arbitrum’s dispute window is 45,818 blocks, about 6.4 days, and ZKsync Era quotes a 3-hour minimum, per Ethereum.org’s L2 directory in October 2026. On a cheap chain, that is where the real cost usually sits.
The payment asset. Starknet’s fee is denominated in STRK in that snapshot, so the dollar figure depends on a token price as well as on gas.
What to do next
Read the fee as a lagging indicator. It is set after the data market moves, so the number on screen describes the last decision, not the next one.
Then check the two things the fee page leaves out: how often the chain posts, and how long the exit takes. The fee to revenue question decides whether a low fee is a subsidy or an efficiency, and what a sequencer does explains who sets the price you are quoted.
Compare chains by exit, not by entry. Entering is cheap almost everywhere. Leaving is where the models differ.
0 comments
No comments yet — ask a question and we will answer it here.