What is an Ethereum Layer 2? Rollups, blobs, stages and what “secure” really means in 2026

Who this is for: Institutional allocators, treasurers, founders and policy teams who need to decide which Ethereum L2s are safe to hold assets on, understand what the stage ratings and blob fee market mean for cost and risk, and ask the right questions before deploying capital or an application.

Ethereum settles only a few dozen transactions a second on its base layer, and when demand spikes the fee can exceed what a transaction is worth. Rather than make the base chain faster, which would make it harder to run and verify, the Ethereum community chose to scale by building networks on top of it. These Layer 2s, or L2s, now process the large majority of Ethereum-related activity: L2BEAT’s October 2026 data shows rollups handling roughly 75 times as many user operations as mainnet. They hold tens of billions of dollars and are where most new applications launch.

Two things have changed since 2024. The Fusaka upgrade of December 2025 and the two blob-parameter forks that followed more than doubled Ethereum’s data capacity for rollups, making the cost of posting data almost a rounding error for the big chains. And the shakeout has begun: the three largest optimistic rollups have reached L2BEAT Stage 1, Base has left the Optimism Superchain, and smaller chains such as Blast, Lisk and Botanix have shut down in 2026, leaving some user funds behind delayed or key-controlled exits.

This guide explains how rollups work, what the blob market does after Fusaka and ahead of Glamsterdam, who leads in October 2026, what the stage ratings mean in practice, where the money flows, and the open questions about sequencers, tokens and based or native rollups. It closes with a worked example on the cost of a swap on Ethereum versus three L2s and what a forced withdrawal takes. For how the base layer reaches agreement, see how blockchain consensus works.

Ethereum Layer 2s by the numbers

The rollup model

A rollup is a chain of its own, with its own blocks, its own fee market and usually fees of a fraction of a cent. Users deposit assets by locking them in a bridge contract on Ethereum, which releases a matching balance on the L2. Transactions then happen on the L2, ordered and batched by an operator called a sequencer. Periodically the sequencer posts the batch’s transaction data, compressed, back to Ethereum, along with a commitment to the resulting state. Ethereum stores that data, and because it does, anyone can reconstruct the L2’s state from Ethereum alone.

That is what “inherits Ethereum’s security” means in the narrow sense: the L2’s history cannot be lost or rewritten as long as Ethereum stands, and users can in principle withdraw through the Ethereum contract even if the operators disappear. Whether that holds in practice depends on how Ethereum learns the posted state is correct, whether users can force transactions in when the sequencer refuses, and who holds the keys that can change the contracts. Those three questions are what the L2BEAT stages measure.

Optimistic versus zero-knowledge

The two rollup families differ in how Ethereum knows the posted state is correct. An optimistic rollup assumes it is and opens a challenge window, typically about seven days, during which anyone can submit a fraud proof showing that a state transition was invalid. If no one does, the state is final; this is why a native withdrawal from Arbitrum One or OP Mainnet takes about a week unless a third party fronts the liquidity. Arbitrum’s BoLD protocol, live since 2025, made its challenge game permissionless, so any staked party can dispute a bad state root.

A zero-knowledge (ZK) rollup submits a cryptographic validity proof with each batch, which an Ethereum contract verifies in a single transaction. Finality on Ethereum follows as soon as the proof is verified, so withdrawals can clear in hours rather than days. The cost is generating proofs, which has fallen sharply as proving systems matured: Starknet uses its Stwo prover, ZKsync Era wraps STARKs inside a SNARK, and Linea runs a PLONK-based recursive stack. The camps are converging. Base’s Azul upgrade of 26 May 2026 added a multi-proof system combining trusted-execution-environment (TEE) attestations with SP1 zero-knowledge proofs to an optimistic chain, and Mantle, Morph and Katana use SP1-family validity proofs according to L2BEAT.

Validiums, optimiums and “others”

Not every chain that calls itself an L2 posts its data to Ethereum. Validiums use validity proofs but keep transaction data off-chain with a committee or an external DA network; optimiums do the same with fraud proofs. If the off-chain data disappears, users may be unable to prove their balances and exit. L2BEAT lists these separately and treats them as a weaker guarantee, with a further “others” category for chains that do not meet its L2 definition at all. Know which category a chain sits in before moving assets there.

Blobs, Fusaka and the data market

Until March 2024 rollups posted their data as ordinary calldata, competing with everyone else for block space, which kept L2 fees at tens of cents. The Dencun upgrade of 13 March 2024 introduced EIP-4844, or proto-danksharding, which created a separate, cheap data channel called blobs. Each blob is about 128 kilobytes, priced in its own fee market, and nodes keep blobs for 4,096 epochs, about 18 days, long enough to verify and challenge the rollup but not forever. L2 fees fell by more than 90 percent within days.

From 3 blobs to 21

Blob capacity has since been raised in steps. Dencun set a target of 3 blobs per block with a maximum of 6. The Pectra upgrade of 7 May 2025 raised that to a target of 6 and a maximum of 9. The Fusaka upgrade activated on mainnet at epoch 411,392 on 3 December 2025 and introduced PeerDAS (EIP-7594), which lets each node hold and verify only a slice of blob data rather than all of it, so capacity can rise without raising hardware requirements. Fusaka also introduced blob-parameter-only (BPO) forks (EIP-7892), small scheduled forks that change blob counts without new client code. BPO1 on 9 December 2025 raised the maximum to 15 blobs (target 10), and BPO2 on 7 January 2026 raised it to 21 (target 14), the values in the consensus-layer mainnet configuration as of October 2026.

Fusaka also changed how blobs are priced. Before it, the blob base fee could fall to 1 wei when demand was below target, so rollups were paying almost nothing for data and Ethereum earned almost nothing from them. EIP-7918 ties the blob base fee to execution gas costs, so that a blob can never be cheaper than a fixed amount of regular gas (8,192 gas per blob) at the prevailing base fee. The fee market now reacts to congestion, but at the low gas prices of 2026, with the average Ethereum gas price around 0.38 gwei on 5 October 2026 according to Etherscan, the floor is still a fraction of a cent per blob. Fusaka also raised the default block gas limit to 60 million and capped any single transaction at 16.78 million gas (EIP-7825).

Glamsterdam and Hegotá

The next upgrade, Glamsterdam, activated on the Sepolia testnet on 6 October 2026 and is targeted for mainnet in the fourth quarter of 2026; the Ethereum Foundation’s September 2026 protocol priorities post describes “shipping Glamsterdam in December 2026.” Its headline items are enshrined proposer-builder separation (ePBS, EIP-7732), which removes the off-protocol relays that currently sit between block builders and validators, and block-level access lists (EIP-7928), which let nodes read state in parallel and validate blocks faster. It also reprices state creation and access (EIP-8037 and EIP-8038). None of these is an L2 feature in itself, but together they raise the base layer’s throughput, the ceiling on how much data and settlement rollups can buy. The upgrade after that, Hegotá, is being scoped for 2027 with fork-choice enforced inclusion lists (FOCIL), a censorship-resistance tool that matters for forced L2 transactions.

The landscape in October 2026

By value secured and activity, the leaders are Base, Arbitrum One and OP Mainnet, all optimistic rollups. Base, launched by Coinbase in August 2023 on Optimism’s OP Stack, has become the busiest network in the ecosystem: growthepie counted about 6.6 million transactions and 288,500 active addresses a day in the week to 5 October 2026, with about $4.99 billion of stablecoins on the chain, second among the 27 chains it tracks. Arbitrum One handled about 1.1 million transactions a day in the same week and holds about $4.78 billion in stablecoins. OP Mainnet is a distant third by value, but its software, the OP Stack, runs Unichain, Ink and dozens of other chains.

The ZK side is more fragmented. Starknet is the largest ZK rollup by value secured and the only major one at Stage 1. Mantle, which L2BEAT now classifies as a validity rollup using SP1 Hypercube proofs, and Lighter, a perpetuals-focused chain with its own prover, each secured well over $1 billion in October 2026 while remaining at Stage 0. Linea, built by Consensys, and ZKsync Era, the zkEVM pioneer, have fallen to the low hundreds of millions in value secured. Rollup kits such as the OP Stack, Arbitrum Orbit and the ZK Stack have made launching a chain a matter of weeks, which is why exchanges (Kraken’s Ink), DeFi protocols (Uniswap’s Unichain, launched February 2025) and brokerages (Robinhood Chain) now run their own L2s.

Source: L2BEAT project pages, October 2026. Value secured is rounded. Stage definitions are explained in the next section.

What “secure” means: the L2BEAT stages

L2BEAT, the ecosystem’s reference monitor, grades rollups from Stage 0 to Stage 2 by how much their security depends on operators rather than on Ethereum. Stage 0 means the chain posts data to Ethereum, but the proof system may be incomplete or permissioned, users may be unable to force transactions through, and a multisig can upgrade contracts instantly. Stage 1 requires a working, permissionless proof system, a way to exit without the operators, and a Security Council of at least 8 members with a 75 percent threshold. Stage 2 means the system is enforced by code: non-bug upgrades give users at least 30 days to exit, and the council can act only on provable bugs.

The test L2BEAT applies at Stage 1 is the “walkaway test”: can users exit in the presence of malicious operators even if the Security Council disappears? In October 2026 Base, Arbitrum One, OP Mainnet, Starknet, Ink and Unichain pass it. None of the major chains is at Stage 2. Arbitrum One is closest, with a 17-day delay on ordinary upgrades and a 10-day exit window, but its Security Council retains an emergency path with no delay, which it used on 24 May 2026 to patch a governance denial-of-service vulnerability and, a month earlier on 21 April 2026, to recover roughly 30,766 ETH from the KelpDAO exploiter after the rsETH incident. Those interventions protected users; they also show that a Stage 1 chain is still a system with humans holding keys.

Where the risk actually sits

1. Upgrade keys. On Base, OP Mainnet and Unichain, upgrades have no on-chain delay, so a malicious or compromised upgrade could in theory move funds before users could react. L2BEAT lists “funds can be stolen if a contract receives a malicious code upgrade” on every one of these chains. Arbitrum’s delay and ZKsync’s 4-to-8-day standard path are the exceptions.
2. Forced inclusion. If the sequencer censors you, can you submit your transaction on Ethereum and have the rollup forced to include it? On Base and OP Mainnet the answer is yes within 12 hours; on Arbitrum One within about a day; on Linea there is no such mechanism, which is why L2BEAT rates its censorship risk as critical.
3. Proof system completeness. Blast’s fraud-proof system “never fully activated,” so a malicious state proposer could have taken funds. Several Stage 0 chains remain in that position.
4. Data availability. A rollup that posts all data to Ethereum blobs can always be reconstructed. A validium cannot if its DA layer fails.
5. Liveness. Linea halted for about an hour on 10 September 2025. ZKsync Era’s operator intervened over a proof-system vulnerability on 30 July 2025. Centralised sequencers can and do stop.

Sequencers: the decentralisation that has not happened

Every major L2 in October 2026 is run by a single sequencer operated by the team or its parent: Coinbase for Base, Offchain Labs for Arbitrum One, OP Labs for OP Mainnet, Uniswap Labs for Unichain. The sequencer decides transaction order, which lets it extract value by front-running and, more importantly, censor or halt. L2BEAT notes on each project page that the operator can “frontrun user transactions” and that “a live and trustworthy operator is vital to the health of the system.” Forced inclusion limits the damage, since a user can bypass the sequencer through the L1 bridge, but it is slow, costs Ethereum gas, and on some chains does not exist.

Several approaches to decentralising sequencing are live or in testing. Shared sequencers such as Espresso aim to order transactions for many rollups at once. Based sequencing, below, hands ordering to Ethereum’s validators. Optimism’s Superchain interop, enabled on OP Mainnet in Upgrade 16 on 14 July 2025 and extended with “super roots” in Upgrade 20 on 24 September 2026, lets OP Stack chains verify each other’s messages, a step toward shared security even with one sequencer per chain. Base’s decision to leave the Superchain on 4 March 2026 and run a “unified, Base-operated stack” with six hard forks a year moves the other way, toward faster iteration under a single operator. None of the top three chains has published a dated plan for multi-party sequencing.

Based rollups and native rollups

Two research directions would change the L2 relationship with Ethereum. A based rollup gives up its own sequencer and lets Ethereum’s block proposer order its transactions, inheriting Ethereum’s liveness and censorship resistance at the cost of slower confirmations and less fee revenue. Taiko launched as a based rollup in 2024 and remains the largest example. A native rollup goes further: instead of each rollup maintaining its own proof contracts, Ethereum would verify rollup execution directly through a precompile, so the L2 would inherit Ethereum’s exact execution rules and upgrade with it. Native rollups are a proposal, not a scheduled upgrade, as of October 2026, but they would make the Stage 2 question moot, because there would be no separate proof system to govern. Both ideas also reduce what a rollup can earn, which is why the chains with the most revenue have shown the least interest.

The economics: fees, rent and the token question

A rollup’s revenue is the fees its users pay. Its main variable cost is the “rent” it pays Ethereum: blob fees for data and gas for posting state roots and verifying proofs. The spread between them is its operating profit. Before Dencun, rent could eat most of a rollup’s revenue. After Dencun and especially after Fusaka’s capacity increases, rent has become small. growthepie’s ecosystem-wide “rent paid” figure for the latest day in early October 2026 was only a few thousand dollars across all tracked chains, while Base alone collected about $54,400 in user fees that day and Arbitrum One about $18,800. On DefiLlama’s chain fee rankings, Base earned about $4.4 million in fees over the 30 days to early October 2026 against Ethereum’s roughly $16.8 million.

The counterintuitive result is that cheap blobs made L2s profitable and Ethereum poorer. Ethereum’s own fee revenue fell as activity moved to L2s, and before EIP-7918 the blob base fee often sat at its 1 wei minimum. This is the heart of the “value accrual” debate: do L2s capture the activity while Ethereum sells them settlement and data at marginal cost, or does ether benefit because every L2 settles in it and holds it on the bridge? Fusaka’s fee floor and the rising L1 gas limit are a partial answer. Those who hold ETH for yield should read our guides to crypto staking and restaking; the KelpDAO rsETH recovery on Arbitrum shows how L2 bridges and restaked collateral interact.

Who has a token and who does not

Most large L2s have a governance token: ARB (Arbitrum DAO), OP (Optimism Collective), STRK (Starknet), ZK (ZKsync), and, since September 2025, LINEA. The tokens vote on treasury spending and, on Arbitrum, elect the Security Council, but users do not need them to transact; gas on every major L2 is paid in ETH. The largest L2, Base, has no token. Base said in September 2025 that it was exploring one, and as of October 2026 that remains the most-asked question in the sector: a Base token would either hand governance of a $16 billion chain to holders or, without governance rights, raise the securities questions our guide to SEC versus CFTC regulation covers, and Coinbase would have to explain how it relates to fee revenue that already flows to shareholders. For how to read any L2 token’s unlock schedule, see how to read a token supply schedule.

How we got here: a timeline

Worked example: a swap on L1 versus three L2s, and a forced exit

How to evaluate an Ethereum L2: a checklist

• Which L2BEAT stage, and why? Stage 1 with a passed walkaway test is the minimum for holding meaningful assets. Read the specific unmet requirements, not just the number; two Stage 0 chains can differ enormously.
• Is the proof system live and permissionless? Blast’s never was. Ask whether anyone can challenge or whether only a whitelist can, and whether the proof covers the whole virtual machine.
• Can you force a transaction through Ethereum, and how fast? 12 hours on Base and OP Mainnet, about a day on Arbitrum One, never on Linea. This is your only recourse if the sequencer censors you.
• Who holds the upgrade keys, with what threshold and what delay? An 8-of-11 or 10-of-13 council with zero delay protects against a single rogue actor but not against a coordinated or compelled upgrade. Arbitrum’s 17-day delay is the current best practice.
• Where is the data posted? Ethereum blobs or calldata mean the chain can be rebuilt from Ethereum alone. An external DA committee means it may not.
• What is the chain’s revenue and runway? Blast was earning about $110 a day in chain revenue when it shut. A chain that cannot cover its sequencer, prover and engineering costs from fees will eventually close or be sold.
• Who operates the sequencer, and is there a dated decentralisation plan? A single operator is the norm; ask whether governance has committed to a date for changing it.
• How does the token, if any, relate to control? On Arbitrum, token holders elect the Security Council; on other chains the token is advisory. Check what the token can actually change.
• What happens on shutdown? Read the exit procedure in advance. Know the bridge contract addresses, the withdrawal delay and whether your assets are plain ETH or something, like staked ETH, with its own unbonding period.

Risks and open questions

Key-holder risk at scale. Roughly $20 billion of value on Base, OP Mainnet and Unichain sits behind contracts that a Security Council plus a foundation or company multisig can upgrade with no delay, according to L2BEAT in October 2026. The councils are reputable and have so far acted to protect users, but the architecture means a compromise of enough signers, or a legal order to them, is a route to funds. Stage 2 would close that route; no major chain has committed to a date for it.

Consolidation and shutdowns. The 2026 closures of Blast, Lisk and Botanix show that rollup software is cheap to launch and expensive to run. Each shutdown is a live test of exit guarantees, and Blast’s shows the gaps: locked staked ETH with its own unbonding, a user deadline, and keyholders who can still change contracts. More closures are likely.

Fragmentation. Liquidity and users are split across dozens of chains. Superchain interop, ZKsync’s interop upgrades and intent-based bridges address this piecemeal, and the rsETH exploit that led to Arbitrum’s April 2026 recovery shows cross-chain messaging remains a major attack surface.

Ethereum’s own economics. If L2s keep nearly all user fees and pay rent that rounds to zero, Ethereum’s security budget depends on issuance and on L1 activity that is itself migrating. Fusaka’s fee floor and Glamsterdam’s repricing are the first protocol-level responses. Whether they are enough, and whether based or native rollups shift the balance back, is unresolved.

Regulation. A listed company operating the largest L2, and perhaps issuing a token for it, puts L2 governance in front of securities and market-structure regulators. How US rules treat L2 tokens and sequencer operators is still open in October 2026.

What to watch next

Glossary

Rollup

A chain that executes transactions off Ethereum and posts compressed transaction data and state commitments back to it, so Ethereum can verify or reconstruct the chain.

Optimistic rollup

A rollup whose state is assumed valid unless someone submits a fraud proof during a challenge window of about a week.

Zero-knowledge (ZK) rollup

A rollup that posts a cryptographic validity proof with each batch, which an Ethereum contract verifies before accepting the new state.

Blob

A roughly 128-kilobyte unit of data attached to an Ethereum block, priced in its own fee market and kept by nodes for about 18 days. Rollups post their data in blobs.

PeerDAS

Peer data availability sampling, introduced in Fusaka, which lets each node verify a slice of blob data rather than download all of it, allowing more blobs per block.

BPO fork

A blob-parameter-only fork: a small scheduled change to blob target and maximum that needs no new client code. Two have occurred, in December 2025 and January 2026.

Sequencer

The operator that orders L2 transactions, builds L2 blocks and posts batches to Ethereum. On every major L2 it is a single party.

Forced inclusion

A mechanism letting users submit an L2 transaction through an Ethereum contract so the rollup must include it within a set delay, bypassing a censoring sequencer.

Security Council

A multisig of named members that can upgrade or pause a rollup’s contracts, usually with a 75 percent threshold, and in Stage 2 only to fix provable bugs.

Walkaway test

L2BEAT’s Stage 1 criterion: users can exit even if the operators are malicious and the Security Council disappears.

Validium

A chain using validity proofs but keeping data off Ethereum, so exits depend on an external data availability provider.

Based rollup

A rollup whose transactions are ordered by Ethereum’s own block proposers rather than a dedicated sequencer.

Native rollup

A proposed design in which Ethereum verifies rollup execution directly through a precompile, so the rollup needs no separate proof system of its own.

Rent

What a rollup pays Ethereum for blob space, state-root posting and proof verification; the cost side of an L2’s business.

Why it matters

Ethereum’s bet is that the base layer should be a slow, maximally verifiable court of final settlement, with everything else happening in networks that rent its security. By October 2026 that bet has largely paid off on the numbers: tens of billions of dollars and most new applications live on L2s, and the biggest have matured to where exits no longer depend on the operator’s goodwill. After Fusaka, data is unlikely to be the bottleneck for years.

But the model has moved the hard problems rather than solved them. Security now depends on a handful of multisigs with instant upgrade power, sequencing on a handful of companies, and the economics on a rent that has fallen close to zero. The shakeout of 2026 is a reminder that “secured by Ethereum” describes the data, not the business, and that the checklist in this guide, not the throughput chart, is what should decide where capital goes. For the base-layer mechanics behind all of this, see our guides on blockchain consensus and reading on-chain data, and for live prices on ETH and the major L2 tokens see the CCS price tracker. Related interviews from the CCS archive include our conversations with Aztec’s co-founders on privacy-focused ZK L2s and Mantle’s Arjun Kalsy on L2 scaling.

Sources

1. L2BEAT: Scaling summary (value secured and stages), October 5, 2026
2. L2BEAT: Activity (user operations and scaling factor), October 5, 2026
3. L2BEAT: Base project page (Stage 1, Azul, Superchain exit), October 5, 2026
4. L2BEAT: Arbitrum One project page (BoLD, upgrade delays, 2026 interventions), October 5, 2026
5. L2BEAT: OP Mainnet project page (Upgrades 16, 17 and 20), October 5, 2026
6. L2BEAT: ZKsync Era project page, October 5, 2026
7. L2BEAT: Linea project page, October 5, 2026
8. L2BEAT: Unichain project page, October 5, 2026
9. Ethereum EIPs: EIP-7607 Fusaka meta (activation epoch, BPO1 and BPO2 dates), December 3, 2025
10. Ethereum consensus-specs: mainnet.yaml blob schedule (15 and 21 max blobs), January 7, 2026
11. Ethereum EIPs: EIP-7892 Blob Parameter Only hardforks, 2025
12. ethereum.org: Fusaka upgrade (PeerDAS, EIP-7918, gas limit), December 3, 2025
13. ethereum.org: Pectra upgrade (blob target 6, maximum 9), May 7, 2025
14. ethereum.org: Glamsterdam upgrade (ePBS, block-level access lists), October 6, 2026
15. Ethereum Foundation: Glamsterdam testnet announcement (Sepolia 6 October 2026), September 17, 2026
16. Ethereum Foundation: Protocol current and emerging priorities (Glamsterdam December 2026, Hegotá), September 7, 2026
17. ethereum.org: Danksharding (blob retention of 4,096 epochs), 2026
18. Base: The next chapter for Base Chain (leaving the Superchain), March 4, 2026
19. growthepie: Base chain metrics (transactions, fees, stablecoins), October 5, 2026
20. growthepie: Arbitrum One chain metrics, October 5, 2026
21. DefiLlama: Fees by chain, October 5, 2026
22. Etherscan: Gas tracker (0.383 gwei average, ETH $2,719), October 5, 2026
23. Crypto Coin Show: Ethereum Layer 2 Blast shuts down with $51 million in user funds locked by five keyholders, October 2, 2026

Disclosure: This guide is for education only and is not investment, legal or tax advice.