Ethereum's main chain handles about 1.8 million transactions a day, and every node re-checks each one, which keeps the chain secure but limits how much it can carry. Layer 2 networks take most of that work elsewhere. They run their own fast, cheap chains and lean on Ethereum for the part that matters most, a public record that lets anyone verify the result and get their money out. In October 2026, Base alone processed about 8.3 million transactions a day. This guide explains the two kinds of rollup, what blobs did to fees, who actually controls these networks, how long withdrawals take and what can go wrong, with data checked on November 2, 2026. The crypto hub covers the wider market, and how a blockchain works explains the base layer itself.
What is a layer 2 blockchain?
A layer 2 blockchain processes transactions off Ethereum's main chain and records their data and results on Ethereum, borrowing its security instead of building its own.
The common design is the rollup. A rollup's operator collects thousands of transactions, executes them, compresses the data and posts it to Ethereum along with a claim about the new state, meaning every account balance after those transactions. Because the data sits on Ethereum, anyone can replay it, check the claim and, if the operator disappears, rebuild the chain and exit. That last property is what separates a rollup from a sidechain, which has its own validators and leaves users relying on them.
The scale is now large. L2BEAT counted $37.91 billion of assets secured by layer 2 projects at 01:00 UTC on November 2, 2026, and Base alone runs more than four times as many transactions as Ethereum itself. Most charge gas in ETH and look identical to Ethereum from a wallet, with the same address format, which is convenient and also the root of many costly mix-ups. The reason the design exists at all is the trade-off between scale, security and decentralization covered in the post on the blockchain trilemma.
How do optimistic and zero-knowledge rollups differ?
Optimistic rollups assume their results are correct unless someone proves fraud within a challenge window, while zero-knowledge rollups attach a cryptographic proof that the results are correct.
That single difference drives most of what users notice. An optimistic rollup is cheaper to run and easier to make compatible with Ethereum's software, but it must hold withdrawals back long enough for an honest watcher to spot and dispute a false claim. A zero-knowledge, or ZK, rollup does heavy computation to produce a validity proof for each batch; once Ethereum's contract verifies the proof, the batch is final and nothing needs to wait.
| Optimistic rollup | Zero-knowledge rollup | |
|---|---|---|
| How results are checked | Assumed valid; anyone can challenge with a fraud proof during a window | A validity proof is verified on Ethereum for every batch |
| Official withdrawal to Ethereum | Days, to cover the challenge window | Once the batch's proof is verified, plus any safety delay |
| Running costs | Lower; proofs are only computed in a dispute | Higher; every batch needs a proof |
| Examples | Arbitrum One, OP Mainnet, Base | Starknet, ZKsync Era, Linea, Scroll |
The line is blurring. Base, the largest rollup by value secured, now resolves its state claims with either a trusted-hardware attestation or a ZK proof, and if both kinds are posted its challenge window shrinks from five days to one.
How did blobs change layer 2 fees?
Blobs gave rollups a cheap, separate place on Ethereum to post their data, and median rollup fees fell about 95% within a day of the March 13, 2024 Dencun upgrade.
Before Dencun, rollups posted their data as ordinary transaction input, competing for the same block space as every other Ethereum user, and that data was most of what their users paid for. EIP-4844 created blobs: chunks of about 128 kilobytes attached to blocks, priced in their own fee market and deleted after roughly 18 days, long enough for anyone who needs the data to copy it. Data availability is the technical name for this guarantee that the data was published, and it is what lets a rollup's users rebuild its state without the operator's help.
The effect was immediate. The day after the upgrade, DL News, using growthepie data, reported that median fees had fallen 96% on Base, 95% on OP Mainnet and Arbitrum One, and 98% on Starknet, which went from about $1.35 to $0.0196. Capacity has risen since, from a target of 3 blobs per block to 14, through Pectra, Fusaka and two follow-up forks, and Fusaka also set a floor under blob prices tied to the main chain's base fee; what Ethereum is lists those upgrades.
Data has become a small part of a rollup's costs. In October 2026, Base collected an average of $83,740 a day in user fees and paid about $700 a day to Ethereum, $128 of it for blobs, according to growthepie. Arbitrum One collected about $20,212 a day and paid about $353. Most of what users pay now covers the rollup's own demand and the operator's margin.
How do layer 2 fees compare with Ethereum mainnet?
Base and Arbitrum One charged a median of a fraction of a cent per transaction in October 2026, about 10 to 30 times less than Ethereum mainnet.
| Network | Type | Median fee, October 31, 2026 | Average daily median, October 2026 | Average daily transactions, October 2026 |
|---|---|---|---|---|
| Ethereum mainnet | Layer 1 | $0.027 | $0.052 | 1.79 million |
| Base | Optimistic rollup | $0.0014 | $0.0017 | 8.33 million |
| Arbitrum One | Optimistic rollup | $0.0029 | $0.0037 | 1.35 million |
| OP Mainnet | Optimistic rollup | under $0.00001 | $0.00003 | 1.50 million |
| Starknet | ZK rollup | $0.0048 | $0.0068 | 143,000 |
| Scroll | ZK rollup | $0.0011 | $0.0031 | 12,950 |
| Linea | ZK rollup | $0.013 | $0.030 | 22,900 |
Source: growthepie. The median is for all transactions on each network, not a plain transfer.
Two things stand out. The gap with mainnet has narrowed because Ethereum's median base fee stayed under 0.1 gwei on 20 of the 30 days from October 2 to 31, 2026, so mainnet is no longer the dollars-per-transaction network of earlier years; gas fees explained shows the arithmetic. And rollup fees are not fixed: Ethereum's October average was pushed up by a surge on October 29, and a rollup's fee rises the same way when its own blocks fill. Busy apps on a single rollup can push its fees above its neighbors' for hours at a time.
Who controls a layer 2, and what are L2BEAT's stages?
On most rollups, one company-run sequencer orders transactions, and multisig signers or a security council can upgrade the contracts holding users' funds, often instantly.
The sequencer is the operator that puts transactions in order and gives the fast confirmations users see; on Base and OP Mainnet, blocks come every 2 seconds, with unofficial preconfirmations every 200 to 250 milliseconds. A single sequencer can go offline or ignore transactions. The protection is forced inclusion: a user sends the transaction to the rollup's contract on Ethereum, and the rollup must include it after a delay.
L2BEAT, an independent research group, grades rollups on how far they have removed these "training wheels." Its stages framework, introduced in June 2023, works like this:
- Stage 0. The project calls itself a rollup, posts its state and all the data needed to rebuild it to Ethereum, publishes software that can do the rebuilding and has a proof system, but a few entities still control it.
- Stage 1. Apart from bugs, the only way to block or forge a withdrawal is to compromise at least 75% of a security council of at least eight members, and upgrades by anyone outside that council must give users time to exit first.
- Stage 2. Anyone can submit fraud proofs, users get at least 30 days to exit before an unwanted upgrade, even one a DAO votes for, and the council can act only on bugs detected on chain.
| Network | Proof type | L2BEAT stage | Value secured, November 2, 2026 | If the sequencer censors |
|---|---|---|---|---|
| Base | Optimistic, with hardware and ZK proofs | Stage 1 | $14.68 billion | Forced inclusion within 12 hours |
| Arbitrum One | Optimistic | Stage 1 | $9.53 billion | Forced inclusion within about 1 day |
| OP Mainnet | Optimistic | Stage 1 | $1.57 billion | Forced inclusion within 12 hours |
| Starknet | ZK | Stage 1 | $348 million | No forced inclusion; a council minority can include it |
| ZKsync Era | ZK | Stage 0 | $232 million | An Ethereum queue exists, but inclusion cannot be forced |
L2BEAT stresses that stages measure dependence on operators, not code quality. A Stage 1 rollup can still have a bug, and upgrades on several of them take effect the moment the signers approve: L2BEAT lists no exit window for Base or OP Mainnet.
How long do layer 2 withdrawals take?
Official withdrawals from optimistic rollups take about five to seven days, while deposits from Ethereum arrive within minutes and ZK rollups pay out once a proof lands.
The wait applies only to the canonical bridge, the rollup's own contract on Ethereum. On Arbitrum One the challenge period is about 6.4 days (45,818 Ethereum blocks). OP Mainnet combines a 3.5-day challenge clock with a 3.5-day buffer, for 7 days. Base cut its delay in May 2026: a withdrawal now takes 5 days when one proof is posted, or about 1 day when both a hardware attestation and a ZK proof agree. Starknet adds no delay once a batch's proof is verified on Ethereum.
Worked example: Three withdrawals started at noon UTC on Monday, November 2, 2026
- Arbitrum One: 45,818 Ethereum blocks × 12 seconds = 549,816 seconds, about 6 days 8.7 hours, so the ETH can be claimed on Ethereum from about 20:45 UTC on Sunday, November 8
- OP Mainnet: 3.5 + 3.5 = 7 days, so from about noon on Monday, November 9, after a separate proving transaction
- Base, one proof: 5 days, so from about noon on Saturday, November 7; with both proofs, from about noon on Tuesday, November 3
- In each case the funds wait in the bridge contract until a final claim transaction is sent on Ethereum, which costs Ethereum gas
These are normal cases. L2BEAT's worst-case estimates, which assume a censoring operator and a maximally delayed dispute, run to 15 days 18 hours on Arbitrum One and 21 days 3 hours on OP Mainnet. Third-party bridges skip the wait by paying out from their own funds on the other side, which adds trust in whoever runs them; crypto bridges explained compares the two routes and their costs. People buying ether on an exchange can often withdraw straight to a rollup and avoid the round trip entirely, a choice how to buy ethereum walks through.
What are the risks of layer 2 networks?
The main risks are operators and councils with power over funds, bugs in bridge and proof contracts, slow exits, and the ordinary price risk of the assets held there.
- Upgrade and emergency powers. Councils exist to fix bugs, and they use those powers. On April 21, 2026, Arbitrum's 9-of-12 Security Council froze about 30,766 ETH held on the network by the attacker behind the KelpDAO bridge hack. After a DAO vote and a May 8 order from a federal court in New York, the ETH went on June 15 to a wallet controlled by Aave LLC, which holds it under the court's supervision for the recovery effort. The freeze kept the funds from the attacker, and it also showed that a small group can change balances on the chain.
- Single sequencers. An operator outage halts fast confirmations for everyone, and forced inclusion takes hours to a day.
- Contract and proof bugs. The bridge contracts on Ethereum hold billions of dollars, and a flaw in them or in a proof system could let an attacker withdraw funds that are not theirs. Base's design also relies on trusted hardware for one of its two proof types.
- Slow and one-way exits. A started official withdrawal cannot be sped up or cancelled, and funds are unusable during the wait.
- Network mix-ups. Identical addresses across rollups make it easy to send funds to the wrong one; sending crypto to the wrong network covers recovery, and how to store crypto safely covers the keys behind every address.
- Token and price risk. Governance tokens such as ARB, whose launch is covered in the post on Arbitrum's native token, are volatile on their own. Ether held on any network can lose most or all of its value, as the ether price page shows.