Ethereum vs. Its Own L2s: Who Really Runs the Show?

In short: Ethereum layer 2s (L2s) are separate networks, mostly rollups such as Arbitrum, Base and OP Mainnet, that execute transactions away from Ethereum and post the data back to it for security. Ethereum provides settlement and data availability, while the L2s capture most user activity and fee revenue, which is why the relationship between them is debated.

At a glance
What L2s get from Ethereum	Settlement, security and data availability
What Ethereum gets	Data (blob) fees, and ETH used as gas on most major rollups
Who runs L2s	Mostly a single sequencer operated by the launching team
Main tension	Activity and fees moving to L2s while mainnet fees fall
Key upgrade	Dencun (March 2024) added blobs and cut L2 data costs
Free trackers	L2BEAT, growthepie

Ethereum’s rollup-centric roadmap has enabled unprecedented scale, but with it comes a new question: Are users trusting Ethereum, or are they trusting a company running an L2?

This article explores the centralization risks across Ethereum’s major Layer 2 rollups: who controls upgrades, who operates sequencers, and what escape hatches users actually have. We’ll use the L2BEAT “Stages” framework to understand which L2s are maturing toward trustless operation, and which still rely heavily on centralized control.

The Framework: Stage 0 to Stage 2

L2BEAT’s Stages framework is a three-level maturity model for rollups:

• Stage 0: Fully custodial or centralized (no fraud/ZK proofs live).
• Stage 1: Proofs are live, but upgrade keys, sequencer control, or escape hatches remain centralized.
• Stage 2: Fully trustless and permissionless. No training wheels.

Today, no major rollup is at Stage 2. Most leading L2s, including Arbitrum, Optimism, Base, and zkSync Era, are still at Stage 1 or below.

Sequencers: Single Points of Failure

Most L2s today are operated by single, privileged sequencers:

• Arbitrum One uses a centralized sequencer, but has plans to move to BoLD for permissionless validation. A Dec 2023 outage revealed risks during high traffic.
• Optimism and Base both rely on a sequencer run by the Optimism Foundation, with outages like Base’s Aug 6, 2025 halt.
• zkSync Era, Linea, and Scroll also rely on single-operator sequencers today.

This architecture enables fast UX but introduces censorship and liveness risks. When sequencers go down, users can’t submit transactions until fallback mechanisms kick in.

Upgradeability & Governance: Who Holds the Keys?

Behind the scenes, most rollups are controlled by upgradeable contracts governed by multisigs or councils:

• Optimism: Controlled by a 2/2 Superchain ProxyAdmin multisig.
• Base: Upgrades flow through two nested 2/2 multisigs governed by Coinbase.
• zkSync Era: Controlled by a 3/3 EmergencyUpgradeBoard with no delay possible.
• Linea: Has 0s upgrade delay; operator can censor or withhold exits.
• Scroll: Upgradeable via multisig with timelocks, but delays may be short.

These upgrade paths are critical for security patches but raise governance centralization concerns. The keys are often held by core teams or foundations.

Escape Hatches: Can Users Exit?

Even if sequencers fail or censor, many rollups include forced transaction mechanisms:

• Optimism allows users to force-include L2 txs after a delay (up to ~12h).
• Arbitrum uses a Delayed Inbox, letting users submit forceInclusion calls after ~24h.

These provide some censorship resistance, but with multi-hour latency and requiring L1 interaction. If upgrade keys are also centralized, these paths can be altered or halted.

Case Studies: When Centralization Breaks Things

• Base outage (Aug 6, 2025): Sequencer failed; block production stopped for 33 minutes. Manual recovery required.
• Arbitrum (Dec 2023): Traffic surge from inscriptions froze the sequencer.
• Optimism (Feb 15, 2024): Sequencer stalled due to unsafe head errors.

These events show how L2s may inherit Ethereum’s settlement security, but not its liveness or censorship resistance.

The Fix: Shared and Decentralized Sequencers

Teams are actively building more decentralized sequencing layers:

• Espresso Systems: Working on shared sequencing to support multiple rollups.
• Astria: Aims for a shared sequencer network with fast finality and composability.
• Flashbots SUAVE: Focuses on cross-domain MEV and decentralized block production.

These systems are early, but promise trust-minimized sequencing and inter-rollup coordination.

Moving Forward

The hope is that more rollups will progress toward Stage 2 on L2BEAT, reducing trust assumptions.

Until then, Ethereum’s scalability comes with trade-offs. And it’s worth asking every time you transact on an L2: who are you really trusting?

Disclaimer

The information provided in this article is for informational and educational purposes only and should not be construed as financial, investment, or trading advice. Onchain News does not provide recommendations to buy, sell, or hold any asset, and nothing here should be taken as a guarantee of future performance. Always conduct your own research and consult a qualified financial professional before making any investment decisions. Cryptocurrency markets are volatile and you are responsible for your own risk.

Common questions

Do Layer 2s take revenue away from Ethereum? They move execution fees to the rollup while still paying the base layer for data. The result is lower mainnet revenue per unit of user activity, which is the substance of the argument.

Is that a problem? It depends whether you think the base layer should capture value or provide settlement. Ethereum’s roadmap chose the second deliberately, so the outcome is a design decision rather than an accident.

How would you measure who runs the show? Compare total fees paid to the base layer against fees captured by rollups, and check how much rollup activity would survive if mainnet data costs rose sharply. That is a dependency test rather than a revenue test.