On 41 of the last 60 days, Ethereum's blob base fee closed at its floor. Not near the floor. At it. For anyone who spent 2023 building a valuation model around data availability, that single line of chain data is the whole story.
I started tracking blob fee distributions the week EIP-4844 went live, because I wanted to know whether the market for data blockspace behaved like the market for execution blockspace. It does not. Execution blockspace gets expensive when people want to do things. Blobspace gets cheap when the protocol decides it should โ and the protocol decides that roughly every twelve months.
The market has not repriced the suppliers. That gap, between a fee market that structurally converges to zero and a cohort of DA tokens still trading on 2023 narratives, is where my research hours have gone this quarter.
Most readers know the broad shape. EIP-4844 introduced blobs as a separate, cheap data channel for rollups, decoupling the cost of posting batch data from the cost of posting execution transactions.
What gets lost is the governance detail. Blob capacity is not set by an auction. It's set by a target-and-maximum parameter pair that core developers adjust. Dencun shipped with a target of 3 blobs per block and a max of 6. The next upgrade lifted that to a target of 6 and a max of 9. PeerDAS, now on the roadmap, moves the ceiling again โ this time by an order of magnitude rather than a multiple.
That is an administered supply curve with a price-triggered escalation clause. When blob fees rise, the response is not that demand is strong and should be allocated accordingly. The response is that the ceiling goes up.
Meanwhile a parallel DA market formed. Celestia, EigenDA, Avail and others pitched the same product: cheaper, more scalable, more flexible data availability than the L1 could offer. They raised real capital and paid real incentives to attract rollup data. The pitch assumed DA would be a scarce resource that modular rollups would bid for. When I audited the parameter change history against the funding timelines, the two lines moved together far too neatly.
The supply side has a structural flaw I don't think has been audited properly. When I audited the blob schedule against historical fee prints, the pattern is unambiguous: capacity expansions are triggered by price, so price cannot persist above the marginal cost of the cheapest available posting venue. A fee market in which the seller unilaterally triples supply whenever the clearing price gets interesting is not a fee market. It's a public utility with a float.
That alone would compress DA fees toward zero. Demand does the rest. Rollups do not generate data at the rate the modular thesis assumed. A busy rollup posts compressed batch data โ state diffs, proofs, calldata โ on the order of hundreds of kilobytes per hour, not per second. The trend line is flat or falling per unit of throughput, because compression got better. Modern rollups post far less data per transaction than the 2022 versions did. ZK proving compresses further. Even as L2 activity grows, blob consumption grows sublinearly.
I checked that against the posting patterns of the largest rollups by value locked. None of them consistently consumes more than a single-digit percentage of per-block blob capacity, and more than one has shifted a share of its posting to cheaper external venues โ which tells you those teams treat DA as a substitutable input, not a dependency. Supply steps up in multiples. Demand grows with a sublinear coefficient. Those two curves have exactly one intersection, and it sits at the floor. The floor is not a temporary artifact of a quiet market. It is the equilibrium.
I ran a version of this analysis in 2020 on Uniswap and Curve pools, trying to separate organic liquidity depth from incentive-driven depth. The metric was simple: how much depth remains when emissions stop? The DA equivalent is instructive. Strip out the points programs, the retroactive airdrops and the foundation grants, and residual demand for third-party DA is a rounding error against available capacity. What Celestia and its peers were measuring was mercenary data โ bytes that arrived because they were paid for, and that leave the same way LP capital left Curve pools in 2021.
There is a macro layer. In a sideways liquidity regime, capital does not fund infrastructure. It funds distribution. The DA cohort raised at infrastructure valuations during a rate environment that no longer exists, and the market has been slow to mark that down because the tokens still have functioning order books.
The plumbing question matters more. When I audited custodial infrastructure ahead of the spot Bitcoin ETF launch in 2024, the operational risk was never data availability. It was settlement finality, proof of reserves, and who holds the keys at 4 a.m. on a Sunday. Institutional capital does not pay a premium for cheaper batch posting. It pays a premium for unambiguous finality and auditable custody. The DA layer solved a problem retail rollups had, not a problem allocators have.
The reflexive read on zero blob fees is bearish for Ethereum: the burn goes quiet, the ultrasound-money arithmetic breaks, the L1 loses a revenue line. I think that's a category error. Blob fees were never designed as revenue. They were rent control on a public good โ deliberately cheap, deliberately expandable, deliberately not a business.
The correct read is a transfer. Value is moving from DA suppliers to application layers. If posting data costs nothing, the economics of an application-specific rollup improve materially, and the binding constraint shifts to proving cost, sequencing latency and cross-domain settlement. Watch where the marginal dollar of rollup spend goes next.
The blind spot in the modular debate is that everyone is watching the price of the commodity at the base of the stack while the actual bottleneck relocated two layers up. Monolithic designs did not win because modularity was wrong. They won because DA was never the scarce input โ and the modular stack is still priced as though it were.
I'd put one question to anyone holding a DA token into the next cycle: if the base commodity of your stack is administered at zero, what exactly are you selling in twelve months? Watch for the pivot. The first DA supplier to reframe itself as a proving and sequencing business rather than a blockspace business will be telling you it has already run these numbers.