Research
Why Layer 2 Sequencers Are Still Centralized in 2026 - And What's Being Done About It
May 7, 2026
11 mins read
Arbitrum, Base, and Optimism still run single-operator sequencers in 2026. Here's what that means for censorship, MEV extraction, and when decentralization actually arrives.
TL;DR: Every major Ethereum L2 still runs a centralized sequencer in 2026. The Linea halt (June 2024) and Base outage (February 2025) made the real-world risks concrete. Espresso Systems launched Mainnet 0 and remains the leading shared sequencer network after Astria shut down in December 2025. Realistic timeline for production-grade sequencer decentralization across the major L2s: late 2026 to 2027 at the earliest.
Key Takeaways:
- Layer 2 sequencer centralization remains unresolved across every major Ethereum rollup as of mid-2026
- A single operator controlling transaction ordering creates three compounding risks: censorship, MEV extraction, and liveness failure
- Real incidents have already occurred: Linea paused its sequencer in June 2024; Base went down in February 2025
- Espresso Systems is the leading shared sequencer network after Astria shut down in December 2025
- Production-grade decentralized sequencing is realistically 12-18 months away for most major L2s
What Does a Sequencer Actually Do?
A blockchain is, at its core, an ordered ledger. Someone has to decide the order. On Ethereum's base layer, that job belongs to the validator network: distributed, permissionless, no single controller. On Layer 2 rollups, that job belongs to the sequencer.
The sequencer picks up transactions from the L2 mempool, decides whether to execute or discard them, orders them into batches, and submits those batches to Ethereum L1 for final settlement. In most major L2 deployments today, that entire process runs through a single operator.
This is a design choice, not a technical necessity. Rollups could use Ethereum L1 for sequencing directly, with no custom sequencer required. But the Ethereum L1 can't handle the transaction volume that L2s generate, and the latency would negate most of the user-experience benefits. So the major rollups chose speed and cost over decentralization, with a commitment to fix it later. That "later" has not yet arrived.
What Are the Real Risks of a Single-Operator Sequencer?
Three distinct risks come with centralized sequencers.
The first is censorship. A centralized sequencer can exclude or delay specific transactions — either because a regulator asked, because a smart contract exploit is ongoing, or simply because the operator decides to. Users can bypass the sequencer and submit transactions directly to L1, but that process is slower and more expensive. It partially defeats the purpose of using an L2.
The second is MEV extraction. The sequencer sees every transaction before it's finalized. It knows which trades are coming, at what size, and in what sequence. That information has value. Sequencers can reorder transactions to capture arbitrage, run sandwich attacks, or front-run large orders. There's no equivalent of Ethereum's PBS (proposer-builder separation) on most L2s yet, and no external auditor monitoring the mempool in real time. The same trust gap applies to oracle manipulation in prediction markets, where a single controlling entity can influence outcomes before settlement.
The third is liveness. If the single sequencer goes down, the entire chain stops. Transactions don't process. DeFi protocols freeze. Positions can't be managed. The only fallback is submitting directly to L1, which provides continuity in theory but imposes costs that most users won't accept in practice.
These are not theoretical scenarios.
When Has Sequencer Centralization Actually Caused Problems?
Linea, June 2024. Following a Velocore DEX exploit that resulted in roughly $6.8M in losses, Linea's team unilaterally paused the sequencer between blocks 5,081,800 and 5,081,801. The pause lasted approximately one hour. During that window, Linea censored attacker addresses to prevent further bridging while the vulnerability was being assessed. Block production resumed after contact with the Velocore team.
The incident was widely debated. Critics, including Matter Labs' CEO, pointed out the obvious tension: a network marketed as decentralized had been halted by a single decision from a single entity. Linea reaffirmed its decentralization roadmap afterward, but the structural issue remained.
Base, February 2025. Coinbase's sequencer experienced a downtime event that halted the entire Base chain. The exact duration was not detailed in public reports, but the incident reinforced what the Linea case had already shown: a single-operator sequencer is a single point of failure. When the operator goes down, so does the chain.
Earlier incidents include a roughly five-and-a-half-hour Arbitrum sequencer outage in 2022 during the Nitro upgrade. No funds were lost, but the outage demonstrated how dependent the rollup's liveness was on one operator staying online.
The pattern across all of these: the failure was not catastrophic in isolation, but the structural dependency is. Base generated over $75 million in sequencer revenue through 2025. The economic stakes of who controls that sequencer, and whether they can be paused, pressured, or compromised, are not small.
What Is the Current Sequencer Status Across Major Rollups?
As of mid-2026, here is the honest picture:
Arbitrum runs a sequencer operated by Offchain Labs. BoLD permissionless fraud proofs are live or rolling out in 2026 phases, which advances Arbitrum toward Stage 1 and 2 status on L2Beat's framework. A "Censorship Timeout" feature is in development to reduce the impact of sequencer-driven censorship via faster force-inclusion to L1. Full sequencer decentralization has no confirmed mainnet date.
Optimism (OP Mainnet) operates a single centralized sequencer run by the Optimism Foundation. Integration with shared sequencing infrastructure (including Espresso and Flashbots) is targeted for mainnet in 2026, aligned with the Pectra upgrade cycle. It remains fully centralized today.
Base has Coinbase as its sole sequencer operator. It reached Stage 1 in April 2025 with permissionless fault proofs, a genuine milestone. But the sequencer itself remains centralized. Base is evolving its own unified stack and has committed to faster decentralization milestones, with no confirmed sequencer decentralization date as of this writing.
zkSync Era is operated by Matter Labs. Earlier roadmaps referenced multi-node sequencer testnets in late 2025 and open participation in 2026. The current 2026 public roadmap prioritizes institutional and privacy features (Prividium, ZK Stack, Airbender). Decentralization is described as "in progress."
Linea is fully centralized, run by Consensys. The roadmap targets Stage 1 in Q4 2025 and a permissioned-set block-building structure via QBFT consensus in 2026. No fully permissionless sequencer yet.
Polygon zkEVM runs a centralized sequencer. The broader Polygon 2.0 and AggLayer work focuses on interoperability across CDK chains; each chain can customize its sequencer but they remain centralized in production deployments.
Scroll operates a team-run centralized sequencer and has published a decentralization roadmap that covers both sequencer and prover decentralization. No confirmed mainnet date for sequencer decentralization.
The L2Beat risk framework treats sequencers as the primary remaining trust assumption across all of these chains. The realistic production horizon for decentralized sequencing across the major L2s is late 2026 to 2027 at the earliest. For a deeper look at how ZK-Data-Rollups fit into this landscape, Orochi's technical overview covers the architecture in detail.
What Is the State of Shared Sequencer Networks in 2026?
Shared sequencers are decentralized networks that handle transaction ordering for multiple rollups simultaneously. The idea is to replace each rollup's single-operator sequencer with a distributed validator set, improving liveness, censorship resistance, and enabling cross-rollup composability.
Espresso Systems is the most developed project in this space. Mainnet 0 is live, running a permissioned set of approximately 100 geographically distributed nodes on HotShot consensus with EspressoDA. It provides fast confirmations (around six seconds, with sub-second latency on the roadmap) and has integrations with the Arbitrum Nitro stack, Cartesi, and the OP Stack. Espresso has processed over 20 million transactions and holds more than $300M in total value secured. The full permissionless proof-of-stake model (Mainnet 1) is targeted for late 2025 into 2026, with the $ESP token now live following an airdrop in early 2026.
Astria, once a leading shared sequencer effort built on Celestia, shut down in December 2025 at block 15,360,577. The project raised approximately $18 million but ceased operations due to funding and adoption challenges. There is no direct replacement. Espresso's momentum has expanded to fill much of the vacuum.
Based rollups (rollups that use Ethereum L1 validators directly for sequencing) are attracting increasing attention as a simpler architectural alternative. No major production deployments have shipped using this model.
Flashbots is building verifiable sequencing infrastructure that centralized sequencers can adopt without becoming a full shared sequencer network. It is one component of the OP Stack decentralization work planned for 2026.
The overall picture: shared sequencing exists and is advancing, but production cross-rollup usage remains limited. The transition from centralized to decentralized sequencing is underway, but slower than early timelines suggested.
What Does This Mean for Institutional Deployments?
The institutions and protocols building on L2s — particularly those dealing with real-world assets, stablecoin reserves, or tokenized securities — face a specific version of this problem. They need sequencer neutrality and liveness guarantees that centralized sequencers cannot currently provide by design.
A centralized sequencer operator faces the same regulatory pressures as any financial intermediary. The U.S. BSA framework could classify sequencer operators as money transmitters, imposing KYC and reporting requirements. That creates a compliance "feature" for some operators, and a censorship risk for others. The Linea incident demonstrated that the line between security response and censorship is narrow when a single entity holds the power.
For protocols moving real-world assets on-chain, the question of who controls transaction ordering is not an abstract concern. It is a counterparty risk. Verifiable, trustless data infrastructure (including how transactions are ordered and settled) is a prerequisite for institutional-grade deployments, not an optional upgrade.
zkDatabase addresses this problem at the data layer: by generating Zero-Knowledge Proofs at the source of each data state, it removes the need to trust any single operator's ordering or attestation. The sequencer problem and the data verification problem are related. Both are instances of the same underlying question: who do you trust, and why?
What Should You Watch in the Next 12 Months?
The meaningful milestones to track through late 2026 and into 2027:
Espresso's Mainnet 1 rollout. The transition to permissionless PoS will be the first real test of whether shared sequencing can operate at production scale without a permissioned validator set.
OP Stack's interop upgrade. Optimism's planned Pectra-aligned release will determine whether shared sequencing becomes a default option for the OP ecosystem, which includes Base and a growing number of L2s built on the stack.
L2Beat Stage 2 ratings. As fraud proof systems mature and sequencer decentralization advances, watch which chains earn Stage 2 status. That rating will become a proxy for institutional due diligence.
Cross-rollup MEV. As multiple L2s begin to interoperate, the MEV problem gets more complex. Shared sequencers become the focal point for cross-domain MEV extraction. Whether that is a feature or a risk depends entirely on how decentralized the sequencer set actually is.
Want to understand how zkDatabase removes single-operator trust assumptions at the data layer? See how zkDatabase works — architecture, use cases, and getting started.
FAQ
Why haven't major L2s decentralized their sequencers yet?
Sequencer decentralization is technically harder than deploying fraud proofs or improving data availability. A decentralized sequencer network requires consensus mechanism design, economic incentive structures, validator coordination, and liveness guarantees that don't depend on any single node. The major L2s have prioritized shipping working products first, then decentralizing. The competitive environment further slows decentralization — teams that spend engineering resources on it fall behind on product features. The result is that every major rollup acknowledges the problem and none have fully solved it.
What happens if a centralized sequencer is ordered to censor transactions by a regulator?
The sequencer operator complies or risks legal consequences. The Linea June 2024 incident showed this in a security context: the sequencer was paused and specific addresses were censored in response to an exploit. A regulatory order would follow the same logic. Users can submit transactions directly to L1 to bypass sequencer censorship, but the gas costs and latency make this impractical for most use cases.
Is Espresso's shared sequencer live and usable today?
Espresso's Mainnet 0 is live with a permissioned validator set. It has processed over 20 million transactions. Integrations with Arbitrum Nitro, Cartesi, and the OP Stack are available. Mainnet 1 (the fully permissionless proof-of-stake version) is in rollout through 2026. Cross-rollup atomic composability via shared sequencing is planned but not yet a live production feature.
Related toResearch
