A sequencer is one wallet.
That is the most precise explainer of the entire concept that anyone has ever written, and everything else you will read on the topic is either a careful elaboration of that sentence or an evasion of it. The elaboration version usually runs a couple thousand words and ends with the reader more confused than when they started. The evasion version skips the "one wallet" part entirely and talks about "the sequencer set" or "the proposer role" as if there were five of them all taking turns. There are not five of them. On almost every production rollup today, there is one.
I have an Etherscan page open on an L2 batch-submission contract as I write this. Every transaction that landed on the rollup in the last hour was submitted by the same L1 address. The address that paid the gas. The address that ordered the batch. The address that decided which of your transactions went in, and which went in the next batch instead, and which did not go in at all. That is the sequencer. That is the whole thing. What makes the topic interesting is not the definition — the definition fits on a coaster. What makes it interesting is what that one wallet can do, what it usually does not do, and what happens when it goes offline for ninety minutes on a Tuesday and nobody quite knows who to call.
I want to be honest about why this article exists. Most of the sequencer explainers you find when you search the term are written for a reader who is three weeks into crypto and will accept "the sequencer orders transactions efficiently" as a complete answer. I am not writing for that reader. I am writing for the reader who has already read two of those explainers, noticed that none of them tell you who runs the sequencer, and came here looking for the part nobody else wants to put on the page.
The Sequencer Is One Wallet With a Monopoly on Transaction Order
Here is what the sequencer actually does, stripped of the vocabulary that usually surrounds it.
When you submit a transaction to a rollup — a swap on Arbitrum, a mint on Base, a stablecoin transfer on Optimism — that transaction does not go to Ethereum. It goes to a node operated by the rollup team. The node receives it, checks that your signature is valid, places it into a buffer, and waits. A few seconds later, the node takes everything in the buffer, decides the order in which those transactions happened, compresses them, and posts the compressed batch to an Ethereum contract. The Ethereum contract does not execute the transactions. It just records that the batch arrived and in what order the sequencer said it arrived. The execution happens later, off-chain, and can be verified later by anyone who wants to re-run it and complain if the math does not match.
The node that did all of that is the sequencer. On most production rollups it is one machine running one client, signing with one private key that lives in one HSM, in one data center, in one jurisdiction. The rollup docs will sometimes say "sequencer" in the plural as if there is a committee. Look at the L1 batch-submission address and you will see one address.
This is the part of the setup that every explainer covers. It is also the part that is true in the same way that "a bank is a building with money in it" is true. Accurate. Load-bearing for nothing. The sequencer has, by design, three powers that are worth stating plainly. It decides which transactions are included in a batch. It decides the order of transactions within a batch. It decides what delay, if any, sits between the moment you submit and the moment the batch is posted. Each of those three powers is a lever. The rollup's trust story is a claim that none of them will be pulled in a way that costs you money.
A Centralized Sequencer Is a Real Engineering Win. That Is Not the Argument.
I want to concede something up front, because the argument for centralized sequencers is genuinely strong and I am not here to strawman it.
A single, high-availability sequencer delivers sub-second transaction confirmations that are simply not available from a decentralized consensus process. It lets a rollup present users with a UX that feels like a major CEX — you click, the transaction confirms, you see the result — and that UX is the reason most users are willing to touch an L2 at all. If you forced every rollup to run a decentralized sequencer set tomorrow, latency would go from the sub-second range into something measured in seconds, and half the DeFi-native product teams would quietly move back to L1 because their LP bots would not tolerate it. The engineering case is real. I am conceding it fully.
Now here is what the concession does not cover. A single sequencer with the three powers above has a set of extractive options available to it that a decentralized sequencer set, almost by construction, does not. It can front-run any transaction it sees before the transaction has been ordered. It can censor any transaction it prefers not to include. It can re-order pending transactions to insert its own transactions in between, capturing the MEV that would otherwise have leaked to builders on L1. None of those things require a malicious operator. They can be done by a piece of middleware running next to the sequencer, by a subcontractor running the sequencer on behalf of the rollup team, or by a future operator that inherits the keys after an acquisition. The set of people who can do these things is exactly the set of people who touch the private key. That set is small. It is, in almost every case, smaller than the set of people who sign off on fee changes at Binance — where at least the maker and taker fees sit at 0.1% and 0.1%, on a public fee page, and a change to either of them has to survive a corporate process that leaves an audit trail. The sequencer extraction surface does not sit on a public fee page. It sits in code, and in the contract between the rollup foundation and whoever they quietly hired to run the machine.
This is the part of the argument that the "centralized sequencer is fine" camp does not want to have. They want to have the latency conversation, which they win. They do not want to have the extraction-surface conversation, which they do not.
The Cost of a Sequencer Is Not the Gas Fee. It Is the Order.
I want to redirect the entire framing of what a sequencer costs you.
The visible cost of using a rollup is the gas fee. It is small and it is the number every explainer puts in front of you as the punchline. "Sequencer does all of this, and you pay two cents." Accurate and beside the point. The invisible cost is the order. When a sequencer decides your swap is included in a batch a few positions later than the transaction you submitted after — and yes, sequencers can do exactly that, because they are the ones writing the order — that is a decision that is worth something to somebody. On a directional move, a few positions of delay on a swap against a moving pool can be the difference between a fill at one price and a fill at a worse one. You will never see that cost. It will not appear in your transaction receipt. Nobody will email you about it. It is priced into the fill, and the fill is all you see.
Compare that to the CEX fee model for a moment, because the comparison is what makes the problem visible. On Binance, the maker and taker fee is 0.1% and 0.1%. On OKX it is 0.08% and 0.1%. On MEXC the maker fee is 0.0% and the taker is 0.02%. MEXC's visible taker cost is roughly five times lower than Binance's, and that is the first number somebody writing a CEX-comparison article would notice and lead with. The delta is small, public, and arguable. There are other costs the CEX fee page does not show — spread markup, funding-rate smoothing, liquidation-engine discretion — and those costs are real, but at least the headline fee is on a page you can screenshot and compare against the next exchange over.
On a rollup, the sequencer extraction surface is the opposite. The headline gas fee is tiny and honest. The real cost is in the ordering, and the ordering is never on a page you can screenshot. There is no "sequencer MEV fee: 0.03% per block" line that you can compare across Arbitrum and Base and Optimism. The number exists. It is just that none of the rollup teams publish it, and the people who can compute it from on-chain traces disagree on the methodology by an order of magnitude.
This is why I find it strange that the Crypto Twitter consensus on sequencers has settled into a comfortable "decentralized sequencers are coming, don't worry about it." They are coming. Several rollup teams have roadmaps. The actual deployments are mostly still testnets, or production deployments with enough operator-council overrides that the centralization has moved into the council instead of the node. The thing that is in production today is one wallet, three powers, and a trust story.
---
This started as a straightforward explainer — what is a sequencer, how does a rollup work, here is the diagram with the batch and the arrows. What I kept running into as I drafted was that the definition is trivial and everything interesting lives in the thing the definition hides. Four drafts later it is an opinion piece about why the definition is the evasion. That is a different article than I started writing, and the shape of it is less tidy than the explainer shape would have been. I think the untidier version is the honest one.
There is a question I do not know the answer to, and I want to leave it on the page rather than pretend I do. If a rollup sequencer is technically capable of extracting MEV from user transactions and technically capable of hiding the extraction inside the fill, and no operator publishes a quarterly "here is what we extracted" number — how would the ecosystem even detect the extraction? Dune dashboards can estimate it. Some have. The estimates vary by an order of magnitude depending on how you define the counterfactual. Whether the actual extraction on any given L2 sits at the low end of those estimates or the high end is a question that nobody in the public data has answered. If you know, write.