When EIP-4844 shipped in March 2024, rollups celebrated. Data posting costs dropped by orders of magnitude almost immediately, and for a while the blob market looked like a solved problem - cheap, abundant, reliable. Eighteen months later, that picture is more complicated.
Blob space on Ethereum is capped at a target of three blobs per block, with a maximum of six. During periods of high L2 activity - particularly when multiple major rollups are posting data simultaneously - the market hits that ceiling, and blob base fees spike sharply. The fee mechanism mirrors EIP-1559’s design: fees burn when demand exceeds target, and rollups that can’t absorb the cost either delay posting or pass it to users.
Optimism and Arbitrum both have public dashboards tracking their L1 data costs, and both have shown intermittent spikes in blob fees during periods of concentrated activity in 2025 and early 2026. The spikes are temporary, but they expose a structural tension: the blob limit was set conservatively to protect node operators, and increasing it requires either Ethereum to raise the cap through a future upgrade or rollups to compress data more aggressively.
The Compression Race
Rollups have responded by pushing harder on data compression. zkSync and StarkNet have both shipped improvements to how they batch and encode transaction data before posting blobs. The efficiency gains are real - in some cases cutting per-transaction data costs by 30 to 40 percent compared to naive encoding - but this is engineering effort spent compensating for a fixed resource, not expanding the resource itself.
Petra, Ethereum’s next major upgrade, is expected to include an increase to the blob count target. Ethereum researchers have discussed moving toward a 6/9 target/max configuration, roughly doubling throughput. Whether that ships in Petra or gets pushed to a later fork is still being debated among client teams as of mid-2026.
Who Actually Pays
The uncomfortable truth is that blob fee spikes are largely invisible to end users right now because most L2s absorb the variance internally, smoothing costs across batches. That subsidy comes from somewhere - either sequencer revenue or, in the case of decentralised sequencers, protocol treasuries.
As rollup sequencer decentralisation progresses and those hidden subsidies shrink, blob fee volatility will eventually surface at the user level. EIP-4844 solved the cost problem for 2024. The scaling work it deferred is just starting to show up on the ledger.