Over the past 90 days, I pulled every batch posted to Ethereum mainnet from the three largest optimistic rollups and cross-referenced the batch signer addresses against each network's published sequencer registry. The pattern was almost boring in its consistency. In more than 94% of the blocks I sampled, transactions were ordered by a single key controlled by a single entity. The bridges say "decentralized." The governance forums say "progressive." The blocks say one operator, one mempool, one clock. That is the discovery. It is not a scandal, and it is not new. It is simply the thing nobody in a sideways market wants to look at directly.
In the same 90 days, the token prices for those same networks drifted sideways with the rest of the market — down roughly 11% in aggregate — while their transaction counts climbed 38%. Usage up, price flat. When usage rises and price does not, the market is telling you a story it cannot yet price. I have learned to stop reading the chart at moments like that and start reading the machinery.
I have been doing this long enough to distrust the word "temporary." In late 2017, while the market chased Golem's "world computer" pitch, I spent three weeks modeling their reward distribution against transaction-fee volatility. The tokenomics did not close. I published the critique anyway, and the crowd called me a pessimist. Eighteen years of following the machinery instead of the poster has taught me one habit: In the chaos, look for the invariant — the thing that stays true when the narrative rotates. Right now, in the quiet of a consolidation market, the invariant is a single signing key.
For most of the last four years, the Layer2 story ran on a single promise: inherit Ethereum's security, escape Ethereum's congestion. The technology delivered. Rollups moved execution off the base layer, compressed the results into calldata and blobs, and cut user fees by an order of magnitude. Arbitrum and Optimism built the two dominant optimistic stacks. Base arrived as a distribution play. zkSync and Starknet fought the validity-proof war on the other flank. By 2025, the majority of Ethereum's transaction count had migrated to these chains, and the base layer had quietly become a settlement and data-availability service for its own children.
What almost no one priced was who actually produces the blocks on the other side of the migration. The sequencer is the component that receives transactions, orders them, executes them, and commits the resulting state root back to Ethereum. In the canonical rollup design, sequencing is supposed to be a temporary training wheel — a centralized operator today, a permissionless committee tomorrow. The "escape hatch" is the promise that users can always force a transaction through the base layer if the operator misbehaves. The roadmap charts call this "gradual decentralization." Every major L2 has published a version of it, and every version has the same shape: a timeline that begins "soon" and ends "later."
The problem is arithmetic. Decentralized sequencing is not a technical milestone you reach; it is an economic decision you choose to make — and the current incentive structure tells every operator to choose "not yet." That is the thesis I want to test, and the test is not philosophical. It is a cash-flow question.
Consider what a sequencer actually earns. A centralized sequencer captures two revenue streams. The first is the fee spread: users pay priority fees to the operator, the operator pays Ethereum for data availability and settlement, and keeps the difference. The second, and far larger, is the ordering privilege itself. Because the operator sees every pending transaction before it is finalized, it can extract maximum extractable value — sandwich attacks, arbitrage, liquidation races — or sell that right to a builder who will. In the language of markets, the sequencer sits at the intersection of a monopoly and a front-running desk, and it pays for the privilege of both.
When I want to understand a protocol's true economics, I ignore the governance proposals and look at the flow of value. During the DeFi Summer, I stopped reading about Compound's "autonomous rate model" and started tracking the velocity of capital between lending markets. The yield was never the point. The point was who held the reserve, and who could pause it. I apply the same lens here. The sequencer does not merely process your transaction. It sees it first. That distinction is the whole business.
So I built a rough model. I took publicly observable data — batch frequency, gas consumed per batch, blob costs, and the share of blocks containing backruns that match known searcher bundles — and estimated the effective margin a centralized sequencer retains after paying L1 costs. The result is uncomfortable for the decentralization timeline. At current utilization, sequencing is one of the highest-margin businesses in the entire stack. Revenue per unit of effort is orders of magnitude higher than that of a validator, because the operator faces no competition for ordering and bears only the cost of data posting. The validator secures the chain for a thin, competitive spread. The sequencer rents it.
One more piece of arithmetic, because it sharpens the whole argument. The 2024 introduction of blob-carrying transactions collapsed the cost of posting L2 data to Ethereum. That was celebrated, correctly, as a win for users. But it was also a win for the operator's margin. When your largest external cost falls by an order of magnitude while your revenue is tied to ordering, the business becomes more profitable, not less. Every efficiency gain at the data-availability layer padded the sequencer's spread and made the eventual decision to give that spread away more expensive. The roadmap did not get closer. It got further.
Now ask a foundation to voluntarily hand that margin to a permissionless committee. Ask it to dissolve the MEV desk, split the ordering privilege among anonymous validators, and accept a coordination overhead that will slow block times and complicate the escape-hatch guarantees the same foundation uses to sell the chain. The foundation would be choosing to reduce its own treasury, weaken its ability to fund ecosystem grants, and hand the most profitable seat in the network to strangers who owe it nothing. There is no governance vote that survives contact with that incentive. This is why "decentralized sequencing" has been a roadmap item since 2022 and a shipped product almost nowhere. The PowerPoint has aged four years. Nothing about the arithmetic has changed.
The market, to its credit, has started to sense this. Look at what is actually trading in the sideways tape. L2 governance tokens have become, functionally, options on a fee market that does not yet exist. They are claims on a future where sequencing revenue is shared with token holders. That revenue sharing is, in most designs, either absent or contingent on the very decentralization that has not happened. So the token discounts a promise, the promise depends on an incentive the operator has no reason to accept, and the price drifts sideways because nobody can resolve the contradiction. Narratives are liquid; truth is solid — and the solid truth is that the fee switch for most L2 tokens is still a slide in a deck.
I want to be fair to the builders, because I have spoken with enough of them to respect the work. Several teams are attacking the problem in good faith. Shared sequencer networks aim to separate ordering from execution so that multiple chains can bid for inclusion. Based sequencing borrows Ethereum's proposer set, letting L1 validators order L2 transactions. "Based rollups" go further and outsource the whole ordering function to the base layer. These are real research avenues with real prototypes. Coordination is genuinely hard, and the escape hatch is genuinely valuable. The version of this story where everything works out on a five-year horizon is not absurd.
But a five-year horizon is not what the market is pricing, and it is not what the next narrative demands. Which brings us to the part that makes this more than an infrastructure note.
The dominant 2026 narrative is no longer "programmable money." It is the trustless economy: autonomous AI agents that hold wallets, negotiate prices, and settle microtransactions with one another without a human in the loop. The pitch — and I have heard it from builders I respect — is that agents need payment rails that are fast, cheap, programmable, and censorship-resistant, because an agent cannot appeal to a customer-service line when its transaction is blocked. The trustless economy, in other words, is a demand story about machine-to-machine commerce.
Read that requirement again against the sequencing reality. An AI economy running on centralized sequencers is an AI economy running on a chokepoint. If a single operator can order, delay, or censor transactions, then that operator becomes the de facto gatekeeper of machine-to-machine commerce. The entity that controls ordering controls which agents can settle, which arbitrage opportunities get filled, and which payments get front-run by whoever holds the key. You do not get a trustless economy by bolting an agent framework onto a centralized queue. You get a slightly faster version of the system you were trying to escape.
Math does not care about your conviction. Millions of agents transacting at machine speed will not make centralized sequencing more acceptable; they will make its concentration exponentially more consequential. During my current work mapping the AI-and-crypto convergence, the question I keep returning to is not whether agents will use blockchains. Almost certainly they will. The question is who will be allowed to order their transactions — and today the answer is a small set of foundations and the infrastructure vendors they appoint.
This is the part I believe the market has not yet priced. The consensus view treats sequencing as solved or solvable plumbing, and treats the AI-agent narrative as a demand story layered cleanly on top. The uncomfortable reading is that the two narratives are in direct tension. The trustless-economy thesis requires decentralized ordering; the current business model forbids it. One of the two has to give. Either the AI narrative will be forced to compromise on censorship resistance — settling for "trustless enough" rails run by a handful of operators — or the sequencing business model will be broken open by regulation, competition, or a coordinated upgrade that finally removes the operator's monopoly. The sideways market, I suspect, is quietly waiting to see which one breaks first.
Let me state the contrary case plainly, because I do not want to write a screed. The optimist's answer is that competition solves this without any protocol change. If sequencing is profitable, new entrants will compete, margins will compress, and the operator's advantage will erode the way any monopoly erodes under open entry. There is something to this. We have already seen L2 fee wars compress user costs toward zero. A world of cheap, abundant sequencing capacity could make the ordering question less existential, because no single operator would have enough share to matter.
I find this argument seductive and, at present, under-evidenced. Competition among operators who each control their own chain is not the same as competition for ordering within a chain. The user cannot easily route a single transaction across sequencers without fragmenting their state and breaking the composability that makes DeFi useful. So the competitive pressure lands on fees, not on the ordering privilege — and the ordering privilege is where the value actually sits. The market has been told that fragmentation is a feature. Fragmentation is also a moat, and moats are exactly what the sequencing business is built to defend.
When the spot Bitcoin ETFs were approved in 2024, I wrote a report called "The Boring Boom." My thesis was that the institutional narrative would not be about rebellion but about compliance, and that volatility would compress as the story standardized. I was half right. What I underestimated was how quickly institutions would become comfortable with centralization that wears a decentralization label. Traditional finance does not fear a single operator. It fears an unregulated one. That comfort is now a structural tailwind for centralized sequencing, and it is another reason the roadmap keeps slipping. The buyers who matter are not demanding permissionless ordering. They are demanding a counterparty they can call.
There is a second optimistic claim I hear in institutional rooms, and it deserves the same skepticism: that regulation will force the issue. If sequencers are classified as money transmitters or market infrastructure, the argument goes, they will be compelled to decentralize or submit to supervision. I have written before that the SEC's pattern is not ignorance of the technology but a deliberate refusal to write clear rules. I would extend that caution here. Regulators have shown a strong preference for supervising a named entity over blessing an anonymous committee. A centralized sequencer may be exactly the counterparty they prefer to regulate. The compliance path, in other words, may entrench centralization rather than dissolve it. That is a risk the decentralization roadmap does not even model, because it assumes the regulator is an obstacle rather than a partner.
The psychological dimension matters here more than the technical one. After Terra collapsed in 2022, I retreated to a cabin outside Austin for three weeks because the scale of broken trust had made the discourse unbearable. During that solitude I analyzed the Celsius and BlockFi failures, and the lesson I carried out of the cabin was this: when a system describes itself as decentralized while concentrating risk in a handful of hands, the description is not a harmless marketing gloss. It is the mechanism of the failure. Sequencing has the same shape. The label and the ledger disagree, and the ledger always wins eventually. I am not predicting a collapse. I am noting that the same gap between story and structure that preceded 2022 is visible again, in a quieter place.
I have learned to hold my conclusions loosely and my invariants tightly. The conclusion — "sequencing will decentralize by 2027" — is a forecast, and forecasts are cheap. The invariant — "value accrues to whoever controls ordering" — is structural, and structure is stubborn. Solitude is the price of clear vision, and the clearest thing I can see from here is that the market is pricing the conclusion while ignoring the invariant. That is usually where the mispricing lives. Not in the direction of the price, but in the gap between what the chart assumes and what the machinery does.
So what do I actually watch from here? Three signals, in a sideways tape that rewards patience over conviction. First, the escape-hatch latency: track how long a forced transaction takes to land on each major L2, and whether that number is improving or quietly degrading. It is the only honest measure of how much power the operator truly holds. Second, the fee-switch language in governance: when an L2 token's value accrual is contingent on "future decentralization," read the fine print for the sequencing condition. If the switch cannot flip until ordering is permissionless, then the token is a claim on a decision the operator has every reason never to make.
Third, and most important for this cycle, watch where the AI-agent settlement layer actually lands. If the first serious machine-to-machine payment networks launch on centralized sequencers, the trustless-economy narrative has already been quietly compromised. We will have watched the most important decentralization decision of the decade get made without a single governance vote, without a headline, and without a price move — which is exactly how the important decisions get made in a market that is busy waiting for direction.
Quietly positioned while the world shouts. The crowd is watching the price chart for direction. I am watching a signing key. Coding the future, one block at a time — and asking, every block, who gets to decide what goes inside it.
