Two weeks ago, a founder I have never met sent me a diligence report. Nine sections. Technical architecture. Token economics. Market structure. Ecosystem position. Regulatory posture. Team and governance. Risk matrix. Narrative and expectation gap. Supply-chain transmission.
Every field was blank. Not redacted. Blank. The template had been filled in with nothing and forwarded upward anyway, the way a signed, unwritten cheque moves across three desks before anybody checks the amount. The 2017 token sale I once pulled apart had a long disclosure document, too. The multisig structure was described in prose and never specified in code, and the space between those two things was where the entire risk lived. I found it by reading, which tells you how little reading was happening elsewhere.
An empty field is a data point. It tells you which questions an industry has quietly stopped asking.
Right now the field nobody is filling in is blob rent — the price of the data-availability space every Layer 2 buys, and the largest controllable cost in a rollup's economics.
Hold onto one number for the next two thousand words. The blob base fee can rise roughly 19x in five minutes and about 360x in ten, because the EIP-4844 fee market is permitted to adjust by up to 12.5% every twelve-second block. There is no priority fee to outbid. There is no auction to win. Rollups wait, or they pay.
That structure sits underneath every sub-cent Layer 2 claim in this bull market. Almost nobody writes it down.
Dencun landed on March 13, 2024, and EIP-4844 came with it. Blobs are 128-kilobyte chunks of data that a rollup posts to Ethereum so that anybody can reconstruct the chain's state later. Before Dencun, that same data travelled as calldata and competed for gas with every swap, mint and transfer on the network. After Dencun, the cost fell by one to two orders of magnitude and stayed there. Base, Optimism, Arbitrum and the rest pushed median transaction fees into fractions of a cent and kept them there through two years of a market that mostly went up.
The detail that matters is that blobs did not get cheaper gas. They got a separate market. A parallel fee mechanism with its own supply, its own target, and its own base fee, floating free of execution fees. At launch the target was 3 blobs per twelve-second block with a hard ceiling of 6. Pectra raised that to 6 and 9 in May 2025 under EIP-7691. Fusaka followed with PeerDAS, which lets validators sample pieces of blob data instead of downloading all of it, and with blob-parameter-only forks — a mechanism for lifting the blob count without shipping a full upgrade cycle. If the published schedule holds, the first of those lifts the target to 10 and the ceiling to 15, with a second step further out.
Two consequences follow from the design, and both are underappreciated. The first is that blob fees have no escape valve. Execution gas has a priority fee: if you need to land a transaction in the next block, you outbid the queue. Blobspace has nothing analogous. The base fee is the price, it is identical for everyone, and the only lever a rollup holds is how much data it chooses to post. The second is that the alternative data-availability layers — Celestia, EigenDA, Avail — are not really competing on price. They are competing on a security trade. Ethereum blobs inherit Ethereum's consensus for availability; an alternative layer inherits its own committee assumptions or its own sampling game. A rollup that switches is buying a lower fare with a different insurance policy, and it will be asked about that difference the first time something goes wrong.
Everything about how these numbers are priced is worth understanding, and I mean that literally rather than as a figure of speech. In 2020, during DeFi summer, I stopped reporting price action for a month and ran live sessions dissecting how Uniswap V2's automated market maker actually worked, because five thousand people were terrified of a mechanism nobody had explained to them properly. "Everybody knows this" is almost always false. The same is true of blob pricing. So: one blob consumes 131,072 units of blob gas. The fee for a blob is the blob base fee multiplied by 131,072. The floor on that base fee is one wei per unit of blob gas, which is why the first year of blobspace felt free. It was not free. It was uncontested.
In the ashes of Terra, we did not learn the lesson we thought we learned. We learned that an algorithmic stablecoin can break. We did not learn that a cost structure can break — that something priced near zero only because nobody was competing for it can reprice violently the moment somebody is. Blobspace in 2024 was exactly that something.
Blobs are not rented per rollup. Think of them as a bus that leaves every twelve seconds with a fixed number of seats, every rollup on the network riding it, all of them paying the same fare, and the fare set by how full the bus is. There is no way to reserve a seat. There is no way to pay more to guarantee one. If the bus is oversubscribed for a sustained period, the fare rises for everyone at once, on a schedule that has nothing to do with how badly any individual passenger needs to travel.
March 2024 made this concrete. Within two weeks of Dencun, inscription-style minting arrived that wrote its data directly into blobs rather than calldata — an obvious arbitrage once you understood that blobspace was priced near zero and execution gas was not. Blobs filled. The base fee came off the one-wei floor and moved by two orders of magnitude. Every rollup's data cost moved with it, whether or not that rollup had anything to do with the mints. Nobody had done anything wrong. The bus was simply full.
Now run the arithmetic. One blob, 131,072 blob gas. Fee per blob equals base fee times 131,072. At the floor of one wei, a blob costs roughly 0.00000000000013 ETH. At a thousand wei, still nothing. At one gwei — a blob base fee of 1,000,000,000 wei — a single blob costs about 0.00013 ETH, call it forty cents at a three-thousand-dollar ether. A rollup posting one blob every block is posting 7,200 blobs a day, which at that fare is around $2,900 a day. Not catastrophic. Not zero either — and the number before it was zero, and the number before that was a rounding error next to calldata.
Then remember the shape of the adjustment. With every seat full, the fare rises about 12.5% per block. Twelve seconds. It doubles every seventy-two seconds and multiplies roughly 360-fold inside ten minutes. Congestion is not a slow squeeze. It is a step function, and rollups are price takers standing on the wrong side of it.
Which leaves a rollup with three unpleasant options under sustained congestion, none of which is a solution. It can wait, accepting that user transactions sit in a queue while the fare is high. It can post less often, batching more state into each blob, which cuts cost and degrades the latency every rollup markets as its differentiator. Or it can move data availability off Ethereum entirely, which changes the trust model in a way its documentation will describe as a design choice and its critics will describe as a downgrade. Producing a fourth option — paying more for a guaranteed seat — is not permitted by the protocol. Certainty is not for sale.
Here is the part the diligence templates leave blank, and the reason the blank matters. Blob rent is paid by the rollup's batcher, the address that compresses and posts state. The user fees a rollup collects go to its sequencer. The sequencer is, in every economically significant case, operated by a company. Coinbase runs Base. Offchain Labs runs the Arbitrum chains. OP Labs and its partners run the OP Stack. The spread between what users pay and what blobspace costs accrues there, in private ledgers, or in the case of a public company, in a footnote nobody reads past.
The token does not touch it. The token votes. Test that claim against the record: go through the last four years of governance proposals on any major rollup and find the one that changed the sequencer's take. You will find parameter tweaks, grant programs, incentive councils, working groups, and a great deal of process designed to look like accountability. You will not find a dividend. A governance token is a non-dividend share in a business whose cash flows are controlled by someone else, and the only exit is a later buyer who believes the same story — a mechanism that is not structurally distinguishable from the ones we all agree to call something else.
I am not saying this to be unkind to holders. I spent the spring of 2022 running a confidential peer-support network for people who lost money in the Terra collapse, coordinating with clinicians and blockchain ethicists, because panic is a mental-health event before it is a market event, and I have no appetite for the sneer. I am saying it because the risk is mispriced, and mispricing is a factual problem before it is anything else.
In the ashes of Terra, we rebuilt the interfaces before we rebuilt the accounting.
Now the part that is genuinely forecastable, and the field I most want the template to have. Every increase in blob supply so far has been consumed within months of shipping. That is not a flaw in the design; it is a property of elastic demand. When you make blockspace cheaper, applications appear that could not have existed at the old price. That is Jevons, and it has held for every capacity increase in the history of this industry. Post-Dencun blobs at 3 target and 6 max sat mostly empty until they did not. The move to 6 and 9 in 2025 bought headroom that is being drawn down now, in a bull market, with rollups competing on fee levels and new consumers arriving through ETF-adjacent distribution channels.
The supply side has a real answer, and it is real engineering. PeerDAS raises the ceiling without requiring every validator to download every blob. Blob-parameter-only forks let the target move without a full hard-fork cycle. Twelve and twenty-four blobs are not fantasy. They are on a calendar.
Understand what those increases do and do not change. They lift the ceiling on supply. They do not change the ceiling on the per-block price adjustment, which stays a hard 12.5%. They do not create a mechanism for a rollup to buy certainty. And they do not alter the direction of demand, because the applications consuming blobspace most aggressively are the ones least sensitive to its price: appchains, L3s, on-chain data markets, and increasingly autonomous agents.
I spent part of 2026 inside a cross-disciplinary working group drafting a transparency standard for autonomous agents with AI ethicists and protocol engineers — five decentralized exchanges have adopted it — and the thing that struck me is how little an agent cares about a cost shock. A human trader facing a 40x fee spike overnight stops trading, opens a ticket, and waits for a human to answer. An agent recomputes the route and keeps going, because its cost sensitivity is a parameter and its patience is infinite. The most efficient consumers of blobspace are also the least polite about it. They will not leave headroom for your rollup, because nobody told them to, and they have no reason to care.
Put the pieces together: demand that grows elastically with every fee reduction, supply that expands on a scheduled cadence, and a fee curve that can move 360-fold in ten minutes inside any given window. The conclusion is not that rollup fees explode tomorrow. The conclusion is that the floor under rollup fees is not zero. It is the blob base fee, it is a bus fare, and it is set by whoever else shows up.
If I were running diligence on any rollup in this market, the first column of the spreadsheet would be blob intensity — blob gas consumed per dollar of fee revenue, plotted weekly against the median blob base fee. The second would be the share of that rollup's transactions that only clear because data availability is currently underpriced, which in a bull market is most of them.
Neither number appears in a single published report I have read this year. What appears instead is total value locked, active addresses, and a diagram.
Which brings me back to the empty template, and to the story that template was written to accommodate. The dominant Layer 2 narrative of this cycle is liquidity fragmentation. Liquidity, we are told, is scattered across forty chains, users suffer, and therefore we need intent layers, solver networks, cross-chain messaging standards, and a new token attached to each one. Note how the problem is defined: in precisely the shape of the products currently raising. Fragmentation is not a diagnosis. It is a market segment. Ask who funds the research that names it, and ask what happens to that research budget if the problem is ever solved.
Watch what the framing leaves out. No interoperability layer reduces blob rent. If anything it increases it, because routing users across more chains means posting state across more chains, and every one of those posts is a seat on the same bus. The interop stack gets paid on volume and fees. The user gets a smoother interface layered over a cost structure that has not moved an inch. And the rollup team gets to avoid the two questions that actually determine its future: who owns the sequencer, and who pays for data availability when the fare triples.
Institutions ask differently, and I know this from the run-up to the spot Ethereum ETF approvals, where I interviewed twelve portfolio managers across major asset managers about the risk frameworks they actually apply, and wrote the results up so retail readers could use them rather than admire them. Not one of them led with fragmentation. They led with sequencer centralization, upgrade keys, and data-availability cost pass-through. That third one has almost no retail coverage at all — a remarkable gap for a variable that can move a chain's unit economics by an order of magnitude inside a single quarter.
The last time this industry agreed that a cost was structurally too small to matter was 2020, when the argument ran that gas was cheap and would stay cheap, right up until a single mint cost more than the artwork it produced.
In the ashes of Terra, we promised each other we would read the documents next time. Mostly, we read the summaries.
So here is what I am watching into the back half of this bull market. The weekly median blob base fee, published prominently, because it is the closest thing this industry has to a cost-of-goods line. The blob-parameter-only fork calendar, because every date on it is a bet on how quickly demand will eat the new capacity — and the last three bets lost. And the first rollup that puts blob rent per transaction on its own dashboard, beside the fee it charges, because that team will have earned the right to describe its fees as sustainable rather than subsidized.
Everyone else will keep shipping the same report with the same nine fields, signed off and forwarded upward. When the fare doubles, and then doubles again, the interesting question will not be how many chains survive it. It will be which one tells its users before the invoice arrives — and whether anyone is still reading closely enough to notice.