Over the past 90 days, the median blob fee on Ethereum sat below 0.0001 ETH. Roughly four cents. In the same window, three of the largest rollups by transaction count spent more on their own sequencer infrastructure than on posting data to L1.
That ratio is the story. Not because blob fees matter to users — they don't — but because the entire data availability thesis was priced on the assumption that blob space would become scarce. It didn't. It became free. The DA market cleared at zero, and the tokens built on top of it still trade as if scarcity exists.
What follows is a teardown of that mismatch.
Ethereum's EIP-4844, activated March 2024, introduced blob-carrying transactions: a separate fee market for temporary data. Targets of three blobs per block, cap of six. Rollups post compressed transaction batches into blobs instead of calldata. Cost reduction was immediate — 10x to 100x depending on the rollup.
That created an adjacent business. Celestia, EigenDA, Avail and a half-dozen others raised capital on one premise: Ethereum's blob space would congest. Their pitch was modular DA, cheaper than L1, with a token capturing the spread.
The math was clean. Congestion → premium DA demand → fee capture → token value.
Reality diverged in one measurable way. Supply expanded faster than demand. Pectra's blob parameter change doubled the target from three to six, with a cap of nine — before any rollup had saturated the original three. Independent DA providers undercut each other into irrelevance. By late 2025 the marginal cost of posting a megabyte of rollup data had fallen below the cost of the server compressing it.
I pulled 90 days of blob data across the top ten rollups by posted bytes. Three observations survived contact with the numbers.
First, utilization. Blob space targets three per block and caps at six. Median utilization across the sample sat at 1.2 blobs. Eleven blocks in 90 days hit the cap — all during a memecoin mint window in April. For 99.4% of blocks, blob space was not merely available. It was empty.
Second, volume. I measured posted bytes per rollup per hour, normalized to each rollup's actual compression ratio. The largest single producer — a general-purpose optimistic rollup — averaged 340 KB per block. The median across the sample was 41 KB. Ethereum allocates 128 KB per blob. The median rollup consumes one-third of a single blob, then pays a DA provider for the privilege of not consuming it.
Third, unit economics. Independent DA layers charge in their own token, so the effective fee tracks token price, not cost. When a DA token trades at 40% of listing price, the provider either subsidizes the service or degrades redundancy. Both happened in 2025. I traced three providers that quietly reduced data availability sampling committee sizes while leaving marketing language untouched.
A DA layer sells two things: bandwidth and a guarantee. The bandwidth is genuinely cheaper than calldata. The guarantee is where the leakage lives. Verifying that a provider stores data requires a sampling scheme with a known security parameter — a function of node count and stake distribution. I ran that parameter against one provider's published committee data. The top twelve operators held 61% of sampled stake. A guarantee backed by twelve operators is not a guarantee. It is a promise with an SLA.
On-chain data confirms where user money actually goes. I reconstructed cost breakdowns for swaps on two large rollups. Of every $100 paid in total cost — L2 execution fee, L1 data fee, priority tip — roughly $4 reached the DA layer, $9 reached the L1 proposer as blob fee, and $31 went to the sequencer. The remainder was captured by priority ordering inside the sequencer's own mempool. Every transaction is a potential extraction point, and the DA debate obscures the smaller invoice.
This is not a claim that modular architecture is wrong. It is a claim about proportionality. The market priced a congestion premium into assets whose demand has a hard ceiling set by rollup throughput. Throughput is set by execution, not data availability. Data availability was solved in March 2024 by a hard fork, and the solution was more space than anyone asked for.
The bull case rests on an unstated assumption: rollups will eventually saturate Ethereum's blob space. Saturating three blobs per block requires roughly 384 KB of compressed data per block. At typical compression ratios of 8:1 to 15:1, that implies aggregate L2 execution throughput of 35 to 60 million gas per second in L1-equivalent terms. Current aggregate L2 throughput is under 4 million. The scarcity thesis requires a 10x throughput expansion no execution roadmap delivers before the next halving. Logic holds; incentives collapse.
Where the bulls are right, precisely: modular DA forced Ethereum to ship blobs. Without Celestia and EigenDA raising capital and publishing research, the social pressure to cut calldata costs would have been weaker. Competition produced a public good. That contribution is real and unquantifiable by design.
They are also right that DA abstraction has value at the margin. Rollups posting to multiple DA layers can arbitrage fees and hedge provider failure. As a risk primitive, multi-DA routing is sound. As a token narrative it is not. Routing reduces provider stickiness, which is the opposite of what a DA token needs to accrue value.
The blind spot is temporal. Bulls assume data demand grows with users. It grows with users and with the compression inefficiency of the execution layer. Better execution compression — calldata optimization, state diffs, tighter batching — reduces data per transaction. Every execution improvement is a DA demand reduction. The better rollups get, the worse the DA business becomes.
So the question for anyone holding a DA token in this market: what would have to be true for blob space to congest, and by when? If the answer requires throughput that does not exist, you are not holding a data availability asset. You are holding an option on a bottleneck that a hard fork already removed. The illusion breaks when the liquidity dries up. The math is perfect; the reality is broken. Position accordingly.
