Ethereum Layer 2s Explained

Ethereum made a deliberate bet: instead of making its base layer faster at the cost of decentralization, it would stay maximally secure and let a second layer of networks handle the volume. Those networks are Layer 2s. This guide explains how rollups work, the real differences between optimistic and ZK designs, what EIP-4844 blobs changed, and how to judge an L2's trust assumptions rather than its marketing.

Why Ethereum Scales with Layer 2s

Every node in the Ethereum network re-executes every transaction. That redundancy is what makes the chain trustless — no one has to believe an intermediary — but it caps throughput at whatever a modest home computer can verify. Block space is metered by a gas limit that validators adjust collectively (they signaled it up to 60 million gas in late 2025), and when demand exceeds supply, users bid against each other and gas fees spike. Raising the limit aggressively would push out smaller node operators and erode the decentralization the whole system depends on.

In October 2020, Vitalik Buterin published what became known as the rollup-centric roadmap, arguing that Ethereum's base layer should specialize in security and data availability while rollups handle execution at scale. That bet defined everything that followed: rather than one chain doing everything, Ethereum became a settlement and data layer for an ecosystem of execution layers built on top of it. Nearly every major upgrade since — most notably Dencun in 2024 and Fusaka in 2025 — has been about making Ethereum a cheaper, higher-bandwidth foundation for L2s.

Before Rollups: The Scaling Ideas That Didn't Win

Rollups were not Ethereum's first scaling plan — they were the survivor of a decade of attempts, and knowing why the alternatives lost explains what makes rollups different. State channels (the idea behind Bitcoin's Lightning and Ethereum's early Raiden) let two parties transact off-chain and settle the net result on L1; they work well for payments between fixed counterparties but cannot host open applications like a DEX, where anyone can show up. Plasma, proposed in 2017, moved computation off-chain and posted only state commitments — but because the underlying transaction data stayed off-chain, users had to constantly watch the chain and rush to exit if an operator withheld data. Per ethereum.org's scaling overview, that data-withholding problem is precisely what rollups fixed: by forcing all transaction data onto L1, a rollup makes mass exits and state reconstruction possible no matter what its operator does.

The other survivor from that era is the sidechain — an independent chain with its own validators and a bridge to Ethereum, like Polygon PoS or Gnosis Chain. Sidechains scale well, but they are their own security domain: if a sidechain's validators collude or fail, Ethereum offers no recourse. The entire vocabulary of this guide — data availability, proofs, stages — exists to draw the line between systems that inherit Ethereum's security and systems that merely connect to it. When the 2020 rollup-centric roadmap declared rollups the path forward, it was a judgment that this line matters more than raw throughput.

How Rollups Actually Work

A rollup runs its own execution environment with its own blocks and state. Users transact there directly — swaps, transfers, lending, anything the EVM supports. Periodically, the rollup batches thousands of those transactions, compresses them, and posts the data to Ethereum along with a commitment to the resulting state. Ethereum does not re-execute the transactions; it stores the data and adjudicates whether the claimed state is correct.

Two properties make this more than just "another chain with a bridge":

• Data availability. Because the transaction data itself lives on Ethereum, anyone can reconstruct the rollup's full state from L1 alone. If the rollup's operators vanish tomorrow, users (or a new operator) can recover every balance. This is the defining difference between a rollup and a sidechain, which keeps its data to itself.
• Enforced correctness. The rollup's contract on Ethereum only accepts state updates that survive a proof mechanism — either a fraud-proof challenge process or an upfront validity proof. The operator cannot simply invent balances.

The economics follow from the batching: one L1 transaction's cost is shared across thousands of L2 transactions. The two rollup families differ in how they enforce correctness, and that difference drives almost every practical tradeoff between them.

Optimistic Rollups: Fraud Proofs and Challenge Periods

Optimistic rollups — the design behind Arbitrum, OP Mainnet, and Base — post state commitments to Ethereum without proving them. The system "optimistically" assumes each batch is valid. What keeps operators honest is the fraud proof: during a challenge window, anyone watching the chain can dispute an invalid state root, force the disputed computation to be re-executed under L1's supervision, and have the fraudulent update rolled back.

Per ethereum.org's documentation, the standard challenge period is about one week — users withdrawing natively to L1 must wait it out before funds are released. Arbitrum's BoLD protocol, which made its validation permissionless, uses a default challenge period of 6.4 days, configurable by the DAO. The week-long delay is the price of never generating expensive proofs: the happy path is cheap, and the dispute machinery only runs when someone cheats.

The practical consequences:

• Cheap and EVM-equivalent. Because no proof is generated upfront, optimistic rollups run essentially unmodified EVM environments. This is a big reason Arbitrum, OP Mainnet, and Base attracted the deepest DeFi liquidity.
• Slow native exits. Withdrawals to L1 take roughly a week unless you use a third-party liquidity provider, which fronts you funds on L1 immediately for a fee and collects your withdrawal when it clears.
• Watchdog assumption. Security requires that at least one honest, capable party is monitoring state roots and willing to challenge fraud. Permissionless validation (as with Arbitrum's BoLD) strengthens this by letting anyone play that role.

ZK Rollups: Validity Proofs

Zero-knowledge rollups — Starknet, ZKsync Era, Linea, Scroll, and others — invert the model. Instead of assuming validity and punishing fraud, they generate a cryptographic validity proof for every batch. Ethereum's contract verifies the proof before accepting the state update, so an invalid state can never be finalized at all. There is no challenge window; per ethereum.org, exits execute as soon as the proof covering them is verified on L1.

Two proof families dominate. SNARKs produce small proofs that are cheap to verify on L1 but typically require a trusted setup ceremony. STARKs (used by Starknet) need no trusted setup and resist quantum attacks, but produce larger proofs. Either way, verification on Ethereum costs a meaningful fixed amount of gas per batch — ethereum.org cites around 500,000 gas — which is amortized across the batch's transactions.

The tradeoffs mirror the optimistic ones:

• Fast finality to L1. Once the proof lands, the state is final. No week-long exit delay.
• Proving is expensive. Generating proofs for general EVM computation is computationally heavy and requires specialized hardware, adding operator cost and, historically, engineering complexity — "zkEVM" compatibility was the field's hardest problem for years.
• Maturity gap. ZK systems are newer and their circuits are complex; a soundness bug in a prover or verifier is a systemic risk that fraud-proof systems do not share in the same form.

Optimistic vs ZK at a Glance

Data Availability and EIP-4844 Blobs

The dominant cost for any rollup is publishing its data to Ethereum. Until March 2024, rollups posted that data as calldata in ordinary transactions, competing with every other Ethereum user for the same block space. The Dencun upgrade changed the economics with EIP-4844 ("proto-danksharding"), which introduced blobs: dedicated data packages of about 128 KB each, priced in their own independent fee market with its own base fee. Blobs are not accessible to the EVM and consensus nodes prune them after 4,096 epochs — roughly 18 days — long enough for anyone to verify or reconstruct rollup state, without burdening Ethereum with permanent storage.

Capacity has been raised in steps. EIP-4844 launched with a target of 3 blobs and a maximum of 6 per block; the Pectra upgrade (May 2025) lifted that to 6 and 9; and the Fusaka upgrade (December 2025) introduced PeerDAS — peer data availability sampling, which lets nodes verify blob data without each downloading all of it — followed by two "blob parameter only" forks that raised the target to 10/max 15 in December 2025 and target 14/max 21 in January 2026.

The effect on users was dramatic. Typical L2 transaction fees fell from dollars to cents after Dencun and kept falling: a mid-2026 academic study of Ethereum transaction costs measured median per-transaction fees of about $0.02 on Base and $0.04 on Arbitrum One as of April 2026. For most activity, L2s deliver Ethereum-grade settlement at a two-orders-of-magnitude fee discount — which is why so much of the activity you see on our live bubble map now happens on L2s.

The Blob Roadmap: PeerDAS to Danksharding

Fusaka's headline feature, PeerDAS, changed how blob capacity can grow. Before it, every consensus node downloaded every blob, so blob count was capped by the bandwidth of a home-staker's connection. With data availability sampling, each node verifies availability by sampling small pieces of the blob data instead of fetching all of it — which is why the Ethereum Foundation could schedule "blob parameter only" (BPO) forks that raise limits without a full hard fork. Per the Foundation's January 2026 checkpoint, BPO1 (December 2025) took the target/max from 6/9 to 10/15 and BPO2 (January 7, 2026) to 14/21 — meaning Ethereum's blob throughput more than doubled within weeks of Fusaka, with no incident.

Core developers have signaled a path toward roughly 48 blobs per block as PeerDAS proves itself, en route to the long-term danksharding goal of about 128, per the ethereum.org Fusaka roadmap page. The practical meaning for users: L2 data costs, already tiny, are on a schedule to keep falling faster than L2 demand has been growing — the strongest structural argument that cheap L2 blockspace is here to stay.

L2Beat Stages: Measuring Trust, Not Hype

"Inherits Ethereum's security" is a spectrum, not a checkbox. The research group L2Beat maintains the ecosystem's standard maturity framework, grading each rollup by how many "training wheels" it still has:

• Stage 0 — full training wheels. The project calls itself a rollup, posts data and state roots to L1, and open-source software exists to reconstruct its state from that data. But users are fundamentally trusting the operator: the proof system may be incomplete or absent, and upgrades can happen out from under users.
• Stage 1 — limited training wheels. A functioning proof system actually adjudicates state roots; at least five external actors can submit proofs; users can exit without the operator's cooperation; and unwanted upgrades give users at least a 7-day exit window. A security council can override, but it must have at least 8 members, a 50% threshold, and a majority outside the operating organization.
• Stage 2 — no training wheels. Proof submission is fully permissionless, the exit window for non-security upgrades is at least 30 days, and the security council may only act on soundness errors detectable on-chain — not governance preferences.

As of August 2026, per L2Beat, the largest L2s — Base (about $11.8B in total value secured), Arbitrum One (about $10.2B), OP Mainnet (about $1.4B), and Starknet — are all rated Stage 1, while Linea, ZKsync Era, and Scroll remain Stage 0. No major general-purpose rollup has yet reached Stage 2. The framework's point is not to shame projects but to make trust assumptions legible: a Stage 0 rollup holding billions is a different risk proposition than a Stage 1 rollup with the same balance, whatever its transaction counts look like.

The L2 Landscape in 2026: Consolidation and New Entrants

The rollup boom of 2021-2024 produced far more chains than the market could sustain, and 2025-2026 was the shakeout. The Block's 2026 Layer 2 outlook describes an ecosystem consolidating around a handful of winners while dozens of small rollups quietly wound down. The live L2Beat table makes the shape of the market plain:

Three patterns stand out. First, winner-take-most economics: Base and Arbitrum One each secure roughly seven times more value than third-place OP Mainnet, per L2Beat's August 2026 data. Liquidity attracts liquidity, and the fee collapse after Fusaka removed cost as a differentiator — leaving distribution (Coinbase's funnel for Base, DeFi depth for Arbitrum) as the moat that matters.

Second, the optimistic/ZK divide is blurring. Several nominally optimistic chains now generate ZK proofs inside their fraud-proof machinery using general-purpose zkVMs, and validity-proof systems have become cheap enough that chains like Mantle switched to them outright. The clean 2023-era taxonomy — optimistic means slow exits, ZK means fast — is dissolving into hybrid designs graded by what actually enforces correctness on L1.

Third, the new entrants are distribution plays: Ink (Kraken) and Unichain (Uniswap) both launched as OP-Stack chains and reached Stage 1 quickly, arriving with built-in user bases rather than incentive-funded TVL. The lesson of the shakeout is that launching a technically sound rollup is now easy — earning a durable reason to exist is not.

Sequencers: The Centralization Everyone Accepts (For Now)

The sequencer is the node that receives user transactions, orders them, and builds L2 blocks. In practice, nearly every major L2 runs exactly one sequencer, operated by the core team — Offchain Labs for Arbitrum, OP Labs for OP Mainnet, Coinbase for Base. This is the most visible centralization in the L2 stack, and it is worth being precise about what it does and does not threaten:

• A sequencer cannot steal. It cannot forge signatures or invent state; the proof system and L1 data availability prevent that.
• It can censor or reorder. A sequencer can refuse your transaction or profit from ordering (the L2 version of MEV). Mature rollups mitigate censorship with forced-inclusion mechanisms that let users submit transactions directly through L1, bypassing the sequencer after a delay.
• It can go down. Sequencer outages have periodically paused major L2s. Funds were never at risk, but liveness depended on one operator — a real difference from Ethereum L1.

Decentralizing sequencing (shared sequencers, based rollups that use L1 proposers directly) is an active research and engineering frontier, but as of 2026 single sequencers remain the norm. The furthest along among major chains is Starknet, whose v0.14 "Grinta" upgrade in September 2025 laid the groundwork for decentralized sequencing while cutting block times from 30 to 6 seconds; the network now runs a dual-token staking model in which STRK carries 75% and staked BTC 25% of consensus weight, per The Block, with full sequencer and prover decentralization staged through 2026 on Starknet's published roadmap.

Interoperability: Making Many L2s Feel Like One Chain

Scaling through many rollups solved throughput and created a new problem: fragmentation. Assets on Arbitrum do not exist on Base; each chain has its own bridges, its own wrapped assets, and its own confirmation delays. For users this means route-planning that L1 never required, and for Ethereum it is a competitive liability against monolithic chains that feel like one system.

In late 2025 the Ethereum Foundation made interoperability an official workstream. Its Ethereum Interop Layer (EIL) initiative aims to make the L2 ecosystem "feel like one chain": a standard way for wallets to treat balances across rollups as one balance and execute cross-chain actions without the user consciously bridging. The companion Open Intents Framework takes an intent-based approach — the user states an outcome ("swap X for Y, deliver on chain Z"), and competing solvers execute the cheapest route across chains, taking on the bridging risk themselves. Alongside this, the Foundation's UX roadmap targets cutting effective cross-L2 confirmation times from many minutes to tens of seconds, per coverage of the 2026 UX roadmap.

For now, the practical guidance is unchanged: moving between L2s still means using a bridge or an intent-based router, with the trust assumptions covered in the bridge section below. But the direction of travel is clear — the endgame is that "which L2 am I on?" becomes a question users rarely need to ask.

What's Next: Glamsterdam and the 2026 Roadmap

Ethereum shipped Pectra and Fusaka on schedule in 2025, and the next fork — Glamsterdam — is in final testing as of August 2026, with client teams targeting mainnet activation around September 2026 after a Sepolia deployment in August, per The Defiant's reporting on the final devnet phase. Two EIPs anchor it, and both matter directly for L2s:

• EIP-7732 — enshrined proposer-builder separation (ePBS). Block building moves from an off-chain relay market into the protocol itself, and the tight ~2-second block propagation-and-validation window stretches to roughly 9 seconds. That widened window is what unblocks the next rounds of blob scaling and gas-limit increases.
• EIP-7928 — block-level access lists. Blocks declare upfront which accounts and storage they touch, letting nodes validate non-conflicting transactions in parallel across CPU cores — a prerequisite for much bigger blocks.

Together these open a path the Ethereum Foundation has stated publicly: pushing the L1 gas limit toward and beyond 100 million in 2026 (it stands at 60 million today), with researchers eyeing about 200 million once ePBS is live, per the Foundation's 2026 protocol priorities. A proposal to shorten slot times to 6 seconds (EIP-7782) was considered but deferred beyond Glamsterdam. A further fork, Hegotá, is penciled in afterward with native account abstraction and post-quantum groundwork, per roadmap coverage. For L2s the through-line is simple: every one of these changes increases the bandwidth and lowers the cost of the base layer they settle on.

Bridges vs Native Withdrawals

How assets move between L1 and L2 matters as much as where they sit. There are two fundamentally different mechanisms:

• Native (canonical) bridges are part of the rollup protocol itself. Deposit ETH into Arbitrum's bridge contract and the same ETH backs your L2 balance; withdraw and the rollup's proof system enforces your claim. Security is the rollup's security — including its challenge-period delays on optimistic systems.
• Third-party bridges are separate applications that hold liquidity on both sides and swap you between them instantly. You trade the challenge-period wait for a new trust assumption: the bridge's own contracts, validators, or multisigs. Bridge hacks have historically been among crypto's largest exploits, so this distinction is not academic.

A useful habit: know which version of an asset you hold. "ETH on Base" via the canonical bridge is a claim on the rollup protocol; a bridged token from a third-party bridge is a claim on that bridge. Our guide to token liquidity covers why thinly-backed bridged assets can also trade at discounts under stress.

How to Compare L2s Honestly

Rankings by a single metric mislead. A practical evaluation combines four lenses:

1. Value secured (TVS/TVL). How much capital the network holds is the clearest measure of market trust — but read it carefully, as incentive programs can rent TVL temporarily. Our guide on how to read TVL covers the pitfalls.
2. Real activity. Transactions, active addresses, and fee revenue show whether anyone actually uses the chain, and whether activity is organic or airdrop farming.
3. Stage and trust assumptions. Check L2Beat's rating, who controls upgrades, and whether exits work without the operator.
4. Ecosystem depth. Which protocols deployed there, how deep the liquidity is, and whether the network has a durable niche (Base's consumer/Coinbase funnel, Arbitrum's DeFi depth, Starknet's app-chain ambitions).

You can compare the major networks side by side on our Layer 2 dashboard, and see how their momentum is scored in how ETHBubbles scores momentum. If you are weighing the investment side — what ARB or OP actually entitle you to versus ETH — that question has its own guide: ETH vs L2 tokens.

A Pre-Bridging Checklist

Before moving meaningful funds to any L2 — especially a newer one — five minutes of due diligence covers most of the avoidable risk:

• Check the stage rating and its footnotes. L2Beat lists not just the stage but the specific caveats: who can pause the bridge, how long the exit window is, whether the proof system covers everything it claims.
• Prefer the canonical bridge for size. Use third-party bridges for speed on amounts you can afford to have stuck; use the native bridge when the amount matters more than the wait.
• Know the exit path before you need it. Confirm the chain has a forced-withdrawal mechanism you could actually operate — or at minimum that independent tooling exists for it.
• Watch for "rollup" branding on non-rollups. Validiums and optimiums keep data off Ethereum; they can be legitimate designs, but they do not carry rollup security guarantees, and L2Beat labels them separately.
• Size positions to the chain's maturity. A Stage 0 chain running months-old code is a different risk than a Stage 1 chain that has processed years of withdrawals — treat allocation accordingly.

Frequently Asked Questions

Related Guides

• ETH vs L2 Tokens — what ARB, OP, and STRK actually entitle you to, and how that differs from holding ETH.
• Ethereum Gas Fees Explained — the fee mechanics that L2s exist to relieve, from base fees to blobs.
• How to Read DeFi TVL — how to interpret the "value secured" numbers used to rank L2s.
• What Is the Ethereum Ecosystem? — the big-picture map that L2s fit into.

Sources & Further Reading

• Vitalik Buterin — A Rollup-centric Ethereum Roadmap (2020)
• ethereum.org — Scaling Overview (channels, plasma, sidechains, rollups)
• ethereum.org — Optimistic Rollups
• ethereum.org — Zero-Knowledge Rollups
• EIP-4844: Shard Blob Transactions (specification)
• Ethereum Foundation — Fusaka Mainnet Announcement (2025)
• L2Beat — Introducing Stages: A Framework to Evaluate Rollup Maturity
• L2Beat — Scaling Summary (live stage ratings and value secured)
• Arbitrum Docs — A Gentle Introduction to BoLD
• Ethereum Foundation — Checkpoint #8: January 2026 (BPO forks and blob parameters)
• ethereum.org — Fusaka Upgrade and PeerDAS Roadmap
• The Block — 2026 Layer 2 Outlook
• The Block — Ethereum Foundation's Interop Layer (EIL)
• AMBCrypto — Ethereum's 2026 UX Roadmap for L2 Interoperability
• The Defiant — Glamsterdam Enters Final Devnet Phase (2026)
• Blockhead — Ethereum Foundation 2026 Protocol Priorities (100M gas limit)
• KuCoin News — Ethereum Foundation 2026 Protocol Roadmap (Glamsterdam and Hegotá)
• The Block — Starknet Enables Bitcoin Staking Alongside STRK
• Starknet — 2025 Year in Review (v0.14, staking, decentralization roadmap)

Disclaimer: The information provided in this guide is for educational purposes only. Layer 2 networks carry distinct smart contract, bridge, and centralization risks, and stage ratings and figures cited here change over time. Nothing here is financial advice.