We didn’t fly to Davos to talk about decentralization. We flew there to nod politely while VCs told us that the only way to scale Ethereum is to give a single company control over the transaction ordering. I sat in a side event, listening to a founder explain how their rollup’s sequencer was "temporarily centralized for speed" – a phrase that has become the industry’s most comfortable lie. The room nodded. I pulled out my phone and checked the source code. Three multisig addresses. Two of them belonging to the same venture firm. The third, a dead link. This wasn’t an exception. It was the rule.
Two weeks later, I wrote an internal memo for my education platform’s research team, titled "The Sequencer Shell Game." It started with a simple question: If the sequencer controls the order of transactions, and the sequencer is operated by a single entity, what exactly has been decentralized? The answer, after auditing the top five rollups by TVL, was nothing. Not one of them had a working, trustless decentralized sequencer in production. Arbitrum had a plan. Optimism had a thesis. zkSync had a blog post. But the live code told a different story – a story where a small group of people could reorder, censor, or front-run transactions at will.
This article isn’t about calling out projects. It’s about the uncomfortable truth that the bull market euphoria has papered over: we are building the same old financial system with new branding, and the sequencer is the Trojan horse.
The Context: Why Sequencing Matters
To understand why this centralization is a disaster waiting to happen, we need to revisit the basic architecture of a Layer 2 rollup. A rollup takes transactions off the main Ethereum chain, executes them in a separate environment, and then posts a compressed proof back to Ethereum. The entity that decides which transactions go into each batch, and in what order, is the sequencer. In a fully decentralized system, the sequencer role would be distributed among many participants, much like Ethereum’s validator set. In practice, almost every major rollup today uses a single sequencer run by the project team.
Why does that matter? Because the sequence of transactions determines everything: which trades go through first, which liquidations trigger, which MEV opportunities get captured. If a single sequencer can reorder transactions, they can effectively front-run every user on the chain. They can censor transactions from certain addresses. They can extract value without permission. The whole point of blockchain – trustless, permissionless, censorship-resistant – breaks the moment the sequencer becomes a gatekeeper.
Yet the industry has accepted this as a temporary trade-off. "We’ll decentralize later," they say. But later never comes. I’ve been tracking this since 2021, and the story is always the same: after launch, the sequencer becomes a revenue source, and the team finds reasons to keep control. It’s not malice; it’s incentive misalignment. Why give up a profitable monopoly when your competitors haven’t either?
The Core: What the Code Actually Shows
Over the past month, I audited the sequencer architecture of six leading rollups: Arbitrum One, Optimism, Base, zkSync Era, Starknet, and Linea. I didn’t rely on whitepapers or Medium posts. I read the actual contracts on Etherscan and L2 explorers. Here’s what I found:
Arbitrum One: The sequencer is operated by Offchain Labs. The sequencer feed is a single endpoint. While Arbitrum has a forced inclusion mechanism (anyone can bypass the sequencer by submitting a transaction directly to L1 via the forceInclusion function), this mechanism is slow (up to a day) and gas-prohibitive for normal users. In practice, the sequencer has total control. The upgrade keys for the rollup contracts are held by a 2-of-4 multisig where two signers are Offchain Labs employees. I checked the addresses – one is an account that hasn’t signed a transaction in three months.
Optimism: OP Mainnet uses a single sequencer run by Optimism Foundation. Their "decentralized sequencing" roadmap, announced in early 2023, promised a peer-to-peer sequencer network by Q4 2023. It’s now Q2 2025. Still not implemented. The current sequencer can censor transactions at will. There is no forced inclusion mechanism. You either go through their sequencer or you don’t transact.
Base: Built on OP Stack, same single sequencer model. Coinbase runs it. Coinbase can – and has, during network congestion – reordered transactions to prioritize their own internal traffic. I found a transaction audit showing that during the memecoin frenzy in March 2024, Base’s sequencer delayed third-party transactions by an average of 12 seconds while processing Coinbase-owned addresses instantly. Coincidence? Perhaps. But the architecture allows it.
zkSync Era: Matter Labs controls the sequencer. They have a "Validator" role on L1 that can stop state updates. The upgrade mechanism is a 2-of-3 multisig. One key is held by Matter Labs CEO. I recall the ETHDenver 2024 panel where they promised a "multi-prover" system within six months. Nothing shipped.
Starknet: Sequencer is run by StarkWare. Their decentralized sequencer plan was announced in 2022. Still single. The "StarkNet Alpha" tag is still on the website.
Linea: ConsenSys runs the sequencer. ConsenSys runs the majority of Ethereum infrastructure (Infura). Now they run your L2 too. The conflict of interest is staggering.
The pattern is clear: not a single rollup has a live, trustless, decentralized sequencer. What they have are promises, blog posts, and token governance proposals that never pass because the team holds enough tokens to veto.
But the deeper problem isn’t just centralization. It’s the lack of any real economic security. In a traditional blockchain, validators stake capital and face slashing if they misbehave. In these L2s, the sequencer operates with no stake. If it goes down, the chain stops. If it’s compromised, the chain is compromised. There is no downside for the sequencer operator – they capture all the upside of MEV and transaction fees, and users bear all the risk.
A Contrarian Angle: Is Temporary Centralization Actually Better?
Let me play devil’s advocate for a moment. Some argue that a single sequencer is more efficient, allows faster upgrades, and has lower operational costs. They say that decentralizing sequencing introduces latency, complexity, and potential for forking. And they’re not entirely wrong.
For example, during the zkSync launch in 2023, the chain processed over 2,000 TPS with a single sequencer. If it had been decentralized with consensus overhead, that number would have been lower. New users don’t care about sequencing decentralization – they care about cheap gas and fast confirmations. In a bull market, where speed and low fees drive adoption, centralization is actually the optimal strategy for growth.
But here’s the catch: growth built on centralization is fragile. If the sequencer operator decides to change the rules – say, to comply with a regulatory request – there is no recourse. If the operator goes offline for a day (like Solana did multiple times, and like Arbitrum’s sequencer did in December 2023), the entire chain freezes. The trade-off isn’t technical; it’s political. You are trusting a single entity with your financial freedom.
And I’ve seen firsthand what happens when trust breaks. In 2020, I trusted a yield farm protocol that had a single admin key. The admin key was exploited. I lost $15,000. The lesson was brutal: if a system has a single point of control, it is not a blockchain – it’s a database with a token.
The Takeaway: What Needs to Happen
We are five years into the L2 narrative, and the industry has not shipped a single production-ready decentralized sequencer. The bull market has masked this failure, but the bear will expose it. When regulators start asking who controls the sequencer, the answer will be "we do" – and that will be the end of the "trustless" claim.
I’m not saying L2s are useless. They are necessary for scaling. But we need to be honest about the trade-offs. If you use an L2 today, you are not using a decentralized application. You are using a centralized service that posts occasional proofs. Treat it as such.
What can be done? First, demand proofs of decentralization. Not roadmaps. Not tweets. A verifiable, non-upgradeable sequencer rotation contract on L1. Second, support projects that actually ship. I’ve seen some promising work from teams like Espresso and Astria, who are building shared sequencer networks. They aren’t tied to any single rollup. They could become the neutral settlement layer we need. But they are still in testnet.
Third, and most importantly, stop accepting the "temporarily centralized" excuse. It’s been temporary for four years. At some point, temporary becomes permanent. Truth in blockchain isn’t found in the whitepaper; it’s found in the deployed bytecode. And the deployed bytecode today says: single sequencer, single point of failure, single point of control.
Let’s not wait for the next black swan to prove what we already know. Read the contracts. Verify the keys. Demand better. Because if we don’t, we’re just rebuilding the same old walled gardens, this time with zero-knowledge proofs.