Rollup fees fell sharply once Ethereum introduced a dedicated area for temporary data. The saving came from correcting a mismatch between how long the data was stored and how long it was needed.
Rollups must publish their data somewhere
A rollup executes transactions off the main chain and posts enough information for anyone to reconstruct its state independently.
That publication is what makes the system verifiable. Without it, users would have to trust the operator's word about balances.
The cost of publishing dominated rollup fees, because the main chain charges for data by volume and rollups produce a great deal of it.
Calldata stored the data forever
Originally the data went into ordinary transaction fields, which are part of the chain's permanent history and stored by every archival node indefinitely.
Rollups only need the data available long enough for anyone to download it and for any challenge period to elapse, which is a matter of days.
They were therefore paying permanent storage prices for a temporary requirement, and competing for the same block space as ordinary transactions.
Blobs price a shorter guarantee
The newer mechanism attaches large data segments to blocks that nodes are only required to keep for a limited period before discarding them.
Execution cannot read the contents directly. The chain stores a commitment, which is enough to prove later that specific data was published.
Because the storage obligation is bounded, the price can be far lower, and this space has its own fee market separate from ordinary transactions.
A separate fee market changes the dynamics
Blob space has its own target and its own base fee that adjusts with demand, so rollups no longer bid against ordinary users for the same resource.
When demand for blobs is below target, the fee falls to a nominal level, which is why rollup costs collapsed initially and why they rise when many rollups publish at once.
The consequence is that rollup fees now move with aggregate rollup activity rather than with main chain congestion, which decouples two things that used to move together.
Data availability is now the competitive dimension
Some rollups publish to the main chain, while others use external systems that are cheaper and carry different security assumptions.
The difference determines what happens if the operator disappears: users can reconstruct their positions from published data, or they cannot.
Since publication is the largest cost, this is where rollup design choices concentrate, and where the trade between fees and self-sufficiency is actually made.