Rollups: optimistic vs. ZK, data availability and blobs
The technology behind most of Ethereum's cheap transactions, and the fine print on how safe it is.
A rollup runs transactions off Ethereum, then posts them to Ethereum in compressed batches along with a way to prove they're correct. Optimistic rollups assume batches are valid unless someone proves fraud within about seven days; ZK rollups attach a validity proof to every batch. Since March 2024, rollups post their data in cheap, temporary "blobs" (EIP-4844), which is why layer 2 fees dropped sharply.
New to layer 2s? Start with the beginner guide, what are layer 2 networks?, and layer 1 vs. layer 2 vs. sidechains.
How a rollup works
- Users send transactions to the rollup's sequencer, which orders them and gives a fast "soft confirmation".
- The sequencer compresses hundreds or thousands of transactions into a batch and posts it to Ethereum.
- A contract on Ethereum records the rollup's new state root, and either waits for challenges (optimistic) or checks a proof (ZK).
- Because the data is on Ethereum, anyone can rebuild the rollup's state and, if needed, withdraw through Ethereum without the sequencer's help.
Optimistic vs. ZK rollups
| Compare | Optimistic | ZK (validity) |
|---|---|---|
| How correctness is ensured | Anyone can submit a fraud proof during a challenge window | A zero-knowledge proof shows each batch is valid |
| Withdrawal to Ethereum | About 7 days via the official bridge | Once the proof is accepted, often hours |
| Main weakness | Needs at least one honest watcher; slow exits | Complex math and code; costly proof generation |
| Examples | Arbitrum, Optimism, Base | zkSync, Starknet, Linea, Scroll |
"Fast withdrawal" services that pay you out in minutes on optimistic rollups are third-party liquidity providers: they front you the money and wait out the week themselves, for a fee.
Data availability: the part people skip
A proof says a new state is correct, but users also need the data to know their own balances and to exit if the sequencer vanishes. Posting the data to Ethereum is what makes a rollup a rollup. Systems that keep the data elsewhere, on a committee or a separate data network, are cheaper but weaker: "validiums" or "optimiums". If that data is withheld, users may be unable to prove what they own.
Blobs: EIP-4844 and after
- Dencun, March 13, 2024 (EIP-4844, "proto-danksharding"): added blob-carrying transactions. Each blob holds about 128 KB, has its own fee market separate from gas (see network fees), and is deleted by nodes after about 18 days. Ethereum only needs to keep the data long enough for anyone who cares to download it; a commitment to it stays on chain forever. Target 3 blobs per block, maximum 6.
- Pectra, May 7, 2025 (EIP-7691): raised the target to 6 and the maximum to 9.
- Fusaka, December 3, 2025 (PeerDAS, EIP-7594): nodes check random samples of blob data instead of downloading every blob, which made room to raise capacity further through small "blob parameter only" upgrades, to a target of 10 and maximum of 15 in December 2025 and a target of 14 and maximum of 21 in January 2026.
That's why fees on major rollups often fell to a cent or less for simple transfers, though they still rise when demand spikes.
Example: one swap on a rollup
You swap USDC for ETH on Base. The sequencer confirms it in about a second. Within minutes your swap is part of a batch posted to Ethereum in a blob, alongside thousands of other people's transactions, so you each pay a sliver of that blob's cost. If you later withdraw ETH to Ethereum through the official bridge, you wait out Base's roughly seven-day challenge window before claiming it on Ethereum.
Limitations and risks
- Centralized sequencers. Most rollups run one sequencer, operated by the team. It can't steal funds or forge transactions, but it can delay or censor them and capture MEV. Good rollups offer a way to force transactions through Ethereum.
- Upgrade keys. If a council can upgrade the rollup's contracts instantly, it could in principle change the rules. Check a rollup's stage on L2BEAT.
- Bugs. Fraud-proof and ZK-proof systems are complex code; a bug could accept an invalid state.
- Fragmentation. Each rollup is its own network; moving between them requires bridges.
Frequently asked questions
Why does withdrawing from Arbitrum take a week?
Arbitrum is an optimistic rollup. The official bridge waits about seven days so anyone can challenge an invalid batch before funds are released on Ethereum.
What is a blob?
A chunk of about 128 KB of data attached to an Ethereum block for rollups to use. It has its own fee and is deleted by nodes after about 18 days.
Are ZK rollups better than optimistic rollups?
They exit faster and don't rely on watchers, but they're harder to build and prove. Both are used widely; maturity and decentralization matter more than the type.
What is data availability?
The guarantee that the data behind a block or batch was actually published, so anyone can check it and rebuild the state.
Can a rollup's sequencer steal my funds?
Not in a well-built rollup: it can't forge your signature, and the proofs or fraud checks stop invalid states. It can delay or reorder transactions, and upgrade keys are a separate risk.
Where to next
Sources
- Optimistic rollups (ethereum.org)
- Zero-knowledge rollups (ethereum.org)
- Data availability (ethereum.org)
- EIP-4844: Shard blob transactions
- Pectra upgrade, including EIP-7691 (ethereum.org)
- Ethereum forks timeline (ethereum.org)
- Ethereum's BPO-2 upgrade: blob target 14, limit 21 (BlockEden, February 2026)
- L2BEAT: rollup stages
Written for beginners and checked against the sources listed. Last reviewed November 4, 2026. This page is educational and is not financial, tax or legal advice. How we write and check pages.