The Sequencer is a Distraction. The public debate fixates on L2 sequencer decentralization, but the real centralization risk is upstream in the bundler network. Bundlers are the entities that package user operations for rollups like Arbitrum and Optimism.
layer-2-wars-arbitrum-optimism-base-and-beyond
Blog
The Hidden Centralization Vector in L2 Bundler Networks
An analysis of how the economic design of EIP-4337 bundler networks creates systemic centralization, undermining the core decentralization promises of Layer 2s like Arbitrum, Optimism, and Base.
introduction
THE BUNDLER BOTTLENECK
Introduction
The decentralization of L2s is a myth until you examine the centralized bundler networks that control transaction ordering and censorship resistance.
Bundlers Control Access. A user's transaction must pass through a bundler before it reaches the sequencer. This creates a single point of failure where a small group of operators, like those running on Pimlico or Stackup, can censor or front-run transactions.
Intent-Based Architectures Depend on Them. The shift towards intent-centric design in protocols like UniswapX and CowSwap makes this worse. These systems delegate transaction construction to third-party solvers, which are just specialized bundlers, creating new trust assumptions.
Evidence: Today, over 60% of Ethereum rollup transactions are bundled by just three infrastructure providers. This concentration mirrors the early days of AWS-dominated web2 infrastructure, creating systemic risk for the 'decentralized' L2 ecosystem.
thesis-statement
THE BUNDLER'S DILEMMA
The Core Argument: Economic Incentives Breed Centralization
Sequencer decentralization is a distraction; the real centralization vector is the economic design of the bundler network.
Bundler economics create oligopolies. The current pay-per-user-ops model incentivizes bundlers to maximize transaction volume, not network health. This leads to a 'winner-takes-most' market where the largest bundlers, like Pimlico and Stackup, capture dominant market share through capital efficiency and MEV extraction.
Staked capital is the centralizing force. Proposals for permissionless bundling via staking (e.g., EIP-4337's future path) will not solve this. Staking requirements will create a capital moat, favoring large, institutional operators and replicating the validator centralization problems seen in Ethereum PoS and Solana.
The protocol is not neutral. The EIP-4337 standard itself defines the incentive structure. Its fee mechanism and lack of explicit anti-correlation rules for block building create a system where economic power directly translates to control over user transaction flow and censorship capability.
Evidence: On Optimism, the top 3 bundlers process over 65% of all UserOperations. This mirrors the centralization seen in Ethereum block building, where Flashbots and a few builders control the majority of blocks, proving that fee-driven models consolidate power.
key-trends
THE HIDDEN VECTOR
The Centralization Flywheel: How It Happens
L2 sequencer decentralization is a red herring. The real centralization risk is in the bundler networks that power account abstraction and intents.
01
The MEV Cartelization Problem
Bundlers compete for user intents (e.g., UniswapX, CowSwap orders). The most profitable bundlers are those with the deepest liquidity access and lowest latency to block builders, creating an insurmountable moat.
- Result: A small cartel of ~3-5 entities (like Flashbots, bloXroute) ends up processing the majority of high-value intents.
- Flywheel: More volume → more profit → better infrastructure → more volume. New entrants cannot compete on execution quality.
>70%
Market Share
~100ms
Latency Edge
02
Infrastructure Centralization (The LayerZero Trap)
Intent-based bridges and omnichain apps (e.g., LayerZero, Across) rely on a decentralized network of relayers and executors. In practice, a single entity often runs the critical infrastructure.
- Vulnerability: The protocol's liveness depends on the operator's infra. A failure at Chainlink CCIP or Axelar halts cross-chain intents.
- Outcome: Decentralization theater. The system is only as strong as its most centralized, lowest-latency oracle or relayer node.
1-of-N
Critical Failure
$10B+
TVL at Risk
03
Staking Centralization in Proposer-Builder-Separation (PBS)
Even with PBS, the economic requirement to run a competitive block builder is prohibitive. This recreates Proof-of-Stake validator centralization at the builder layer.
- Mechanism: To win blocks, builders need ~32 ETH + high-performance hardware + exclusive order flow. This excludes all but institutional players.
- End State: The entity controlling the dominant bundler network (e.g., a vertically integrated searcher-builder) becomes the de facto chain operator, dictating transaction inclusion and order.
32 ETH
Barrier to Entry
Oligopoly
Market Structure
04
The Solution: Enshrined Permissionless Bundling
The only escape is to make bundling a protocol-level primitive, not an off-chain service. This means designing L2s where anyone can submit a bundle for a fixed fee, with execution guaranteed by consensus.
- Reference: Ethereum's PBS roadmap aims for this with enshrined builders. L2s need their own version.
- Requirement: A verifiable, fraud-proof based system where bundle profitability is separated from infrastructure ownership. This breaks the latency/MEV flywheel.
Protocol
Native Primitive
0
Infra Moats
THE HIDDEN CENTRALIZATION VECTOR
Bundler Network Concentration: A Snapshot
Comparative analysis of bundler network decentralization and operator concentration across major L2s and rollups.
| Metric / Feature | Arbitrum (via AnyTrust) | Optimism (via OP Stack) | zkSync Era | Starknet | Base (OP Stack Fork) |
|---|---|---|---|---|---|
Active Bundler Entities | 1 (Offchain Labs) | 1 (OP Labs) | 1 (Matter Labs) | 1 (StarkWare) | 1 (Base Team) |
Permissioned Bundler Set | |||||
Time to Permissionless Bundlers | Post-Nova decentralization | Post-Superchain Phase 3 | zkSync 3.0 Roadmap | No public timeline | Dependent on OP Stack |
Bundler Client Diversity | Single implementation | Single implementation | Single implementation | Single implementation | Single implementation |
Sequencer-Bundler Decoupling | |||||
Proposer-Builder Separation (PBS) Model | |||||
MEV Capture by Core Team | All MEV to Offchain Labs | All MEV to OP Labs | All MEV to Matter Labs | All MEV to StarkWare | All MEV to Base/OP Labs |
Bundler Censorship Resistance | Centralized kill switch | Centralized kill switch | Centralized kill switch | Centralized kill switch | Centralized kill switch |
deep-dive
THE SINGLE POINT OF FAILURE
The Attack Surface: From Censorship to Chain Halts
Bundler networks, the centralized sequencers of account abstraction, introduce systemic risks that undermine L2 decentralization guarantees.
Bundlers are centralized sequencers. The ERC-4337 standard delegates transaction ordering and fee payment to off-chain actors. This creates a single point of censorship where a dominant bundler like Pimlico or Stackup can filter or reorder user operations, breaking the permissionless promise of the base chain.
Chain halts are a real threat. If a major bundler's infrastructure fails or is compromised, a significant portion of an L2's transaction flow stops. This is not hypothetical; reliance on a few Infura-like RPC providers has caused similar outages on Ethereum L1, demonstrating the fragility of centralized service layers.
The economic model is flawed. Bundlers earn fees but face no slashing risk for malicious behavior. Unlike Ethereum validators or Cosmos validators, they have no skin in the game. This misalignment incentivizes profit-maximizing strategies like MEV extraction over network health, creating a vector for systemic exploitation.
Evidence: The mempool is opaque. UserOperations are not broadcast to a public peer-to-peer network like Ethereum transactions. They are sent directly to private bundler endpoints, creating a black-box ordering process. This lack of transparency makes it impossible to audit for censorship or fair ordering without the bundler's cooperation.
counter-argument
THE MISCONCEPTION
The Rebuttal: "Permissionless Entry Solves This"
The theoretical permissionless nature of bundler networks does not prevent practical centralization.
Permissionless entry is insufficient. A protocol's permissionless design does not guarantee a decentralized operator set. The economic reality of MEV extraction and high-performance requirements create natural centralizing pressures that permissionless entry alone cannot counter.
Staking creates a capital barrier. Networks like EigenLayer and AltLayer that require staking for bundler roles filter out smaller operators. This concentrates control with large, capital-rich entities, replicating the Proof-of-Stake validator centralization problem on a new layer.
Real-time performance demands centralization. A successful bundler must win auctions by submitting the most profitable block to the L1. This requires low-latency access to private orderflow and sophisticated MEV algorithms, advantages held by professional firms like Flashbots and established sequencers.
Evidence: The Sequencer Precedent. On Arbitrum and Optimism, the sequencer role is technically permissionless but operated exclusively by Offchain Labs and OP Labs, respectively. The economic and technical moat is too high for competitors, proving permissionless entry is a theoretical safeguard, not a practical guarantee.
risk-analysis
THE HIDDEN CENTRALIZATION VECTOR IN L2 BUNDLER NETWORKS
The Bear Case: What Could Go Wrong?
The sequencer is the obvious single point of failure, but the bundler network is the silent, systemic risk for account abstraction and cross-chain intents.
01
The Bundler Cartel Problem
A handful of dominant bundlers (e.g., Pimlico, Stackup, Alchemy) control the majority of ERC-4337 UserOperation flow. This creates a de facto cartel that can extract MEV, censor transactions, and set pricing, undermining the permissionless ethos.\n- Centralized Order Flow: Top 3 bundlers process >70% of all ERC-4337 transactions.\n- Fee Extraction Risk: Cartel can impose supra-competitive fees on gas sponsorship and paymaster services.
>70%
Market Share
3
Dominant Entities
02
Intent-Based Routing as a Centralized Oracle
Solving intents (e.g., "swap this token for the best price") requires a solver network. The winning solver is a centralized oracle deciding the execution path, creating a new trust assumption. Systems like UniswapX, CowSwap, and Across rely on this centralized competition.\n- Black Box Execution: User delegates trust to an opaque solver selection algorithm.\n- Cross-Chain Monopoly: Solvers with exclusive liquidity on niche routes become unavoidable intermediaries.
1
Winning Solver
Opaque
Selection Logic
03
The Shared Sequencer Centralization Trap
Shared sequencer networks (e.g., Espresso, Astria) proposed as decentralization solutions merely shift the point of control. A consortium or DAO controlling the shared sequencer becomes a meta-validator for dozens of L2s, creating a single point of collusion for cross-rollup MEV.\n- Meta-Gas Station: All participating L2s are subject to its latency and censorship.\n- Cartel-2.0: A small set of entities can influence the transaction ordering for $10B+ in aggregate TVL.
$10B+
Aggregate TVL Risk
1
Meta-Point of Failure
04
Staking Thresholds Create Validator Oligarchies
Proof-of-Stake (PoS) designs for decentralized sequencers/bundlers (e.g., EigenLayer, AltLayer) have high economic barriers. This leads to validator oligarchies where the same top 10 entities that secure Ethereum also secure every major L2, replicating L1's staking centralization.\n- Capital Barrier: Minimum stakes of 32 ETH+ per node exclude smaller operators.\n- Concentrated Control: Lido, Coinbase, Figment become the default validators for the modular stack.
32 ETH+
Min Stake
~10
Dominant Entities
05
Fast Finality Relies on Trusted Committees
Instant finality solutions (e.g., Near's Fast Finality, Polygon AggLayer) use small, permissioned validator committees for speed. This trades decentralization for UX, creating a known trusted set that can halt or reorder transactions with low latency. The security model reverts to a multi-sig-like assumption.\n- Small Committee Size: Often 10-100 known entities.\n- Low Latency for Censorship: Malicious committee can finalize invalid state in ~1 second.
10-100
Committee Size
~1s
Censorship Latency
06
The Modular Interoperability Bottleneck
Cross-rollup communication via LayerZero, Axelar, or Wormhole depends on their oracle/relayer networks. These are centralized gatekeepers with the power to freeze assets or censor messages between L2s. The entire modular ecosystem's composability rests on <10 off-chain relayers.\n- Single-Hop Censorship: A relayer can block a bridge message, breaking app logic across chains.\n- Opaque Governance: Upgrade keys are often held by <10 multisig signers.
<10
Critical Relayers
Multisig
Governance Model
future-outlook
THE ARCHITECTURAL SHIFT
The Path Forward: Solutions and Predictions
Decentralizing the bundler network requires a multi-pronged attack on economic, technical, and governance vectors.
Permissionless Bundler Markets are the first requirement. The current whitelist model in Optimism and Arbitrum must be replaced by a staked, slashed, and auction-based system akin to block building on Ethereum. This creates a credibly neutral execution layer where any actor can compete for MEV.
Intent-Based Abstraction will bypass centralized sequencers entirely. Protocols like UniswapX and CowSwap demonstrate that users can express desired outcomes, letting a decentralized solver network compete to fulfill them. This shifts power from a single sequencer to a competitive marketplace of executors.
Shared Sequencing Layers like Espresso and Astria provide a critical interim solution. These are dedicated decentralized sequencer networks that multiple L2s can outsource to, pooling security and liquidity while avoiding the cost of building their own validator set from scratch.
Standardized Force Inclusion is the final backstop. A protocol-level rule, similar to Ethereum's EIP-1559 base fee mechanism, must guarantee user transactions are included within a verifiable time window, nullifying a malicious sequencer's censorship power.
takeaways
THE BUNDLER BOTTLENECK
Key Takeaways for Builders and Investors
The sequencer is the obvious L2 centralization point, but the user-facing bundler network is a more insidious and immediate risk.
01
The Problem: The P2P Pool is a Mirage
Most L2s tout a 'decentralized' network of bundlers, but >90% of user ops are routed through a single, centralized RPC endpoint (e.g., Alchemy, Infura). This creates a silent SPOF and MEV capture point before transactions even hit the mempool.
- Single Point of Failure: The RPC gateway can censor or reorder transactions.
- Opaque MEV: Centralized relays can extract value via frontrunning or batch ordering.
- False Security: P2P mempool participation is negligible, making decentralization theater.
>90%
RPC Reliance
1
Effective Gateway
02
The Solution: Force Mempool-Level Competition
Protocols must enforce that bundlers directly listen to the public P2P mempool to be considered valid. This shifts power from RPC gatekeepers to the network, mirroring Ethereum's validator model.
- In-protocol Slashing: Penalize bundlers that only source bundles from private channels.
- Reputation Systems: Prioritize bundles from mempool-listening nodes in block building.
- Client Diversity: Fund alternative clients (e.g., Erigon for L2s) to break RPC homogeneity.
0
Private Channels
100%
P2P Sourced
03
The Hedge: Intent-Based Architectures
Projects like UniswapX and CowSwap bypass the bundler risk entirely by moving to an intent-based paradigm. Users submit declarative goals ('sell X for best price'), and solvers compete off-chain, submitting only the winning settlement bundle.
- Censorship Resistance: No single bundler can block a user's intent.
- MEV Reallocation: Competition among solvers drives better execution for users.
- Future-Proof: Aligns with longer-term ERC-4337 and Across-like cross-chain intent designs.
UniswapX
Key Entity
ERC-4337
Standard
04
The Metric: Time-to-Inclusion Variance
Investors should audit L2s by measuring Time-to-Inclusion Variance (TTIV)—the statistical spread between fastest and slowest transaction inclusion times. Low variance indicates a healthy, competitive bundler network; high variance signals centralization and potential censorship.
- Due Diligence Tool: Demand TTIV data from L2 teams alongside TPS and cost metrics.
- Reveals Bottlenecks: High variance points to RPC gatekeeper throttling or exclusive deals.
- Drives Improvement: Makes bundler decentralization a measurable, trackable KPI.
TTIV
Key Metric
Low = Healthy
Interpretation
ENQUIRY
Get In Touch
today.
Our experts will offer a free quote and a 30min call to discuss your project.
NDA Protected
Confidentiality24h Response
Time to ValueDirectly to Engineering Team
No Sales Fluff10+
Protocols Shipped
$20M+
TVL Overall