A simple swap on Arbitrum in April 2026 costs a few cents. The same swap, routed directly on Ethereum mainnet, can run fifteen to thirty dollars when the network is busy. Both transactions end up secured by the same validator set — but only one of them actually lives on L1. The other lives on a rollup, and the difference between those two sentences is the story of how Ethereum scaled without changing its security model.
Total Layer 2 TVL is around $41 billion in early 2026, with rollups specifically holding about $34.7 billion. Two chains — Arbitrum One ($16 billion) and Base ($12 billion) — account for roughly 68% of it. Understanding what these systems actually are, how they differ, and how to pick one is now a baseline skill for anyone using Ethereum in practice.
What a rollup actually is
A rollup is a specialised execution environment that does three things, all of them inseparable:
- Executes transactions off-chain — you send a transaction to the rollup's sequencer, not to Ethereum directly.
- Posts state commitments to L1 — the rollup periodically writes a compressed summary of its new state to a smart contract on Ethereum.
- Publishes the underlying transaction data somewhere verifiable — so anyone can reconstruct what happened and challenge it if it's wrong. This is data availability, or DA.
That third leg is what separates a rollup from a sidechain. Polygon PoS and BSC are sidechains: they run their own validator sets, their own security budget, and Ethereum has no idea what's happening on them. A rollup inherits Ethereum's security for settlement and DA. If the rollup's operators vanish tomorrow, the data needed to reconstruct every user's balance is still on Ethereum, and users can exit.
This is the same family of scaling problem Bitcoin's Lightning Network tries to solve, but with a different answer. Lightning uses off-chain payment channels between pairs of users; nothing is ever committed back to Bitcoin until a channel closes. Rollups take the opposite approach: commit everything back, frequently, in compressed form. The trade-offs land in different places, but the goal is the same — more throughput without weakening the base layer.
Rollups run the same EVM semantics and the account model as mainnet (with some exceptions noted below), which is why your wallet, your address, and your Solidity code port over with minimal changes.
Optimistic rollups: fraud proofs and the 7-day window
Optimistic rollups assume the sequencer is honest and give the network time to prove otherwise. Arbitrum, OP Mainnet, and Base all follow this pattern.
The sequencer posts a state assertion to L1. A challenge window opens — roughly 6.4 to 7 days on Arbitrum, about 7 days on OP Mainnet, 3.5 days on Base. During that window, anyone can submit a fraud proof (sometimes called a fault proof) showing the assertion is invalid. If nobody challenges, the state finalises. If someone successfully challenges, an interactive dispute plays out on L1 and the fraudulent proposer is slashed.
The practical consequence: transactions feel instant inside the L2 (the sequencer confirms them in seconds), but withdrawing funds back to L1 through the canonical bridge takes the length of the challenge window. Fast bridges like Across, Hop, and Stargate front the liquidity for a fee, letting you exit in minutes instead of a week.
By early 2026, all three major optimistic rollups run permissionless fraud proofs:
- Arbitrum activated its BOLD permissionless system and remains the largest L2.
- OP Mainnet turned on permissionless fault proofs on June 10, 2024.
- Base (Coinbase's OP Stack chain, launched August 2022) shipped fault proofs in October 2024 with a 3.5-day window and reached L2Beat Stage 1 in April 2025 with a 10-member Security Council requiring 75% consensus.
"Permissionless" is the key word. Early optimistic rollups had a whitelist of approved challengers, which meant fraud-proof security rested on those specific actors staying online and honest. Removing that whitelist is the jump from Stage 0 to Stage 1.
ZK rollups: validity proofs and near-instant finality
ZK rollups take the opposite stance: prove every batch is correct at the moment it's posted. Each batch submitted to L1 comes with a succinct cryptographic proof — a SNARK or STARK — that the new state root is the correct result of applying those transactions. The L1 contract verifies the proof in a single call. There is no challenge window because there is nothing to challenge.
zkSync Era cites roughly 3 hours for full L1 finality via its validity proofs, compared to the 7-day wait for optimistic withdrawals. That latency gap is the headline ZK pitch.
The ZK landscape splits into EVM-compatible and non-EVM designs:
- zkSync Era, Linea (Consensys), and Scroll are zkEVMs that target drop-in Solidity tooling. Most contracts deploy with no changes; edge cases around precompiles, gas accounting, and specific opcodes still exist.
- Starknet is non-EVM. It runs the Cairo VM and uses STARK proofs, which have no trusted setup and are considered quantum-resistant, at the cost of larger proof sizes. Contracts are written in Cairo, not Solidity, so porting is a real engineering task.
The economic reality in early 2026 is that ZK rollup adoption has not yet caught up to the tech: Starknet holds about $529M TVL, Linea $499M, zkSync Era $340M, and Scroll $65M. Collectively they are smaller than either Arbitrum or Base alone. ZK is very likely the long-term end state; it is not the dominant present.
Blobs, EigenDA, and Celestia: where the data lives
Before the Dencun upgrade on March 13, 2024, rollups posted their transaction data as calldata on Ethereum mainnet — expensive, and competing with regular transactions for block space. Dencun introduced EIP-4844, which added a new transaction type carrying blobs: roughly 128 KiB data chunks priced on a separate fee market and pruned from consensus nodes after about 18 days.
The effect on rollup economics was immediate. L1 data costs for rollups fell roughly 10 to 100×. In the first year of blobs, rollups paid approximately 1,020 ETH in blob base fees plus about 2,000 ETH in EIP-1559 fees — against an estimated $34 million per month on calldata in the months before the upgrade. The full mechanics are covered in the piece on blob fee market and EIP-4844 mechanics.
By default, major rollups now publish their data to Ethereum blobs. That keeps the rollup's security model pinned to Ethereum's validator set and gives you the purest "rollup" security posture.
Alt-DA layers exist for rollups willing to trade some security for cheaper throughput:
- Celestia holds roughly half of the overall data-availability market outside Ethereum blobs in 2026.
- EigenDA, an AVS on EigenLayer, targets around 100 MB/s throughput in V2 and is substantially cheaper than Celestia at high volumes. This connects directly to EigenLayer restaking and AVSs — EigenDA is the flagship application of restaked ETH.
- Avail is the Polygon spinout playing in the same space.
Rollups that post to alt-DA are, strictly speaking, validiums or hybrids — they rely on an external committee for data availability rather than Ethereum itself. L2Beat flags this clearly on every chain page. For high-frequency app-specific rollups (gaming, specialised DEXes), the economics of alt-DA can make sense. For a general-purpose chain holding billions in user deposits, Ethereum blobs remain the default.
The Stage ladder and the sequencer problem
L2Beat's Stages framework is the most useful decentralisation rubric in the ecosystem. It categorises rollups by how much trust users still place in the operator versus in smart contracts and permissionless actors.
- Stage 0 — Full Training Wheels: Operator-run. The system self-identifies as a rollup, posts state roots, provides DA on L1, and makes its software source-available.
- Stage 1 — Limited Training Wheels: Smart-contract governance with a Security Council of at least 8 members (≥50% external, 50% threshold). Functional proof system with at least 5 external fraud-proof submitters. Operator-independent exits. A 7-day exit window for unwanted upgrades.
- Stage 2 — No Training Wheels: Fully contract-managed. Permissionless proof system. A 30-day exit window. The Security Council's role is restricted to on-chain-detected errors only.
In early 2026, Arbitrum One, OP Mainnet, and Base have all reached Stage 1 with live permissionless fraud proofs. No major rollup has yet reached Stage 2. Every chain below Stage 1 is, for all practical purposes, a trusted operator running software that happens to post to Ethereum.
Then there is the sequencer problem, which the Stage framework only partially addresses. Almost every L2 in production — Arbitrum, OP Mainnet, Base, zkSync Era — runs a single permissioned sequencer: Offchain Labs, the OP Foundation, Coinbase, and Matter Labs respectively. A single sequencer can delay your transaction, reorder it to extract value, or go down entirely. OP Mainnet, Arbitrum, and Base have all had sequencer outages measured in hours.
The mitigation Stage 1 rollups offer is the force-include escape hatch: if the sequencer won't process your transaction, you can submit it directly to the rollup's L1 contract and force it into the queue. That keeps the worst-case censorship bound. But day-to-day UX still depends on one operator staying online and honest. Shared-sequencer projects like Espresso and Astria had limited adoption; Astria shut down in 2025. Arbitrum's decentralised-sequencer roadmap targets 2026. This is the frontier, not a solved problem.
Picking an L2 as a user
A practical, ordered framework:
- Wallet and chain support. Does your wallet derive addresses on the L2? Most do, because rollups reuse Ethereum's address format. For reaching L2 dApps directly, Zelcore with WalletConnect v2 covers essentially any chain with a WalletConnect surface.
- Bridging. The canonical bridge is safest — its security reduces to the rollup's own L1 contracts — but slow on optimistic chains (roughly 7 days). Fast bridges charge a fee for liquidity-fronted instant finality. Bridges are where large money historically gets stolen: Ronin ($625M, March 2022), Nomad ($190M, August 2022), Wormhole ($325M, February 2022), and the Kelp DAO rsETH bridge drained for about $292M in April 2026. Treat every bridge as a distinct, unaudited-enough trust assumption.
- Stage and proof system. Stage 1 with permissionless fraud proofs — Arbitrum One, OP Mainnet, Base — is the practical bar for holding meaningful capital. Stage 0 rollups with centralised operators are fine for amounts you can lose to a rugpull.
- DA check. Ethereum blobs is the standard. Alt-DA is an extra trust assumption. L2Beat labels every chain. If the rollup posts to Celestia or EigenDA, you're one layer further from Ethereum's security.
- When the L2 is worth it at all. Frequent DeFi, NFT mints, stablecoin payments — yes. A single long-term ETH hold you touch once a year — no. Keep it on L1 or in cold storage. The bridge round-trip is friction, and every bridge is a risk surface.
The total L2 TVL of $41 billion still sits below Ethereum L1's on-chain value. Rollups won transaction volume; L1 remains the settlement and store-of-value layer. Treat L2s as where you act, not where you hold.
Key takeaways
- A rollup is off-chain execution plus on-chain state commitments plus verifiable data availability. Miss any leg and it's a sidechain.
- Optimistic rollups (Arbitrum, OP, Base) dominate L2 TVL and enforce a 3.5–7 day withdrawal window. ZK rollups (zkSync, Linea, Scroll, Starknet) compress finality to hours via validity proofs but hold far less capital today.
- Dencun's EIP-4844 blobs cut rollup DA costs by roughly an order of magnitude and made sub-cent L2 transactions economically sustainable.
- For meaningful balances, L2Beat Stage 1 plus permissionless fraud proofs plus Ethereum-blob DA is the minimum realistic bar. Sequencers are still single points of failure almost everywhere.
- Bridges are the most frequently exploited surface in crypto. Use canonical bridges for size, fast bridges for speed, and never treat any bridge as neutral infrastructure.
Part 6 — the capstone — turns to practical self-custody: holding ETH in Zelcore, gas hygiene, L2 access, and what Pectra and Fusaka change for users.
