Layer 2 Fees Compared: Optimistic vs ZK Rollups in Practice

Rollups cut Ethereum fees by doing the work somewhere else and posting a summary back. They differ in how they prove that summary is correct, and that difference shows up in fees, withdrawal times, and what happens when a proof fails. This compares the two designs by how they actually behave.

The shared trick

Both types execute transactions off the main chain, where blockspace is cheap, and then publish a compressed summary to Ethereum. The main chain does not re-execute your transactions. It checks a claim. What differs is how that claim is checked.

One number dominates everything else: cost per transaction. On the main chain, fees are an auction for scarce blockspace, so a busy minute makes every transaction expensive. Rollups replace that auction with a fixed fee set by the sequencer, which is why a swap costs cents instead of tens of dollars and does not spike when the rest of the network is congested.

Optimistic rollups assume honesty

An optimistic rollup publishes transaction data and asserts the resulting state is correct. During a challenge window, anyone can submit a fraud proof showing the sequencer lied. If nobody challenges it, the state is accepted.

This is why withdrawals take time. You are not waiting for a mathematical proof; you are waiting for the window to close, which used to be about a week. Most venues solved this for users by providing fast liquidity on the destination in exchange for a fee, but that is a third party taking on the risk for you, not the protocol settling faster.

ZK rollups prove it

A zero-knowledge rollup publishes a cryptographic proof that the batch of transactions was executed correctly. Validity is demonstrated rather than assumed, so there is no challenge window and withdrawals settle as soon as the proof lands on the main chain.

The cost of that proof is computation, which is why proving was the slow part for years. Recent hardware and circuit improvements have narrowed it substantially, though proving remains the heaviest part of the stack.

How the trade-offs show up in practice

• Transaction cost. Both are orders of magnitude cheaper than the main chain. ZK proving adds a fixed overhead per batch, which is spread across its transactions.
• Withdrawal time. ZK is near-instant once proven. Optimistic is fast with a liquidity provider, slow if you withdraw yourself.
• Raw data posted to the main chain. Optimistic rollups historically post full transaction data. ZK rollups post a proof and some compressed data. That difference matters a lot for main chain capacity and long-term cost.
• Proof systems. Optimistic rollups generally rely on fraud proofs. ZK rollups vary widely, with different proof systems carrying different proving costs and hardware requirements.
• Ecosystem maturity. Newer ZK systems have smaller ecosystems: fewer integrations, more missing bridges, rougher developer tooling. If your application only supports one, that overrides every fee advantage.

Choosing practically

Judge a rollup on more than the headline number. Check the fixed cost per transaction on a typical swap, whether the application you use supports it, whether the bridge you need is available, and how much data the chain posts to the main chain. The last one affects the entire ecosystem's cost curve and is rarely advertised.

The optimistic-versus-ZK debate matters for how these systems evolve. For a user deciding where to transact today, both are cheap enough that convenience and stability usually decide the question.

Frequently asked questions

Are all rollup transactions really cents?

Most are, and the number varies by network, by batch load, and by what the transaction does. A swap is cheap on every major rollup. The difference shows up when you bridge back to the main chain, which costs a main chain transaction regardless of which rollup you came from.

Why do some rollups still withdraw in seven days?

That is the optimistic challenge window. If you use a venue that offers instant withdrawals, they are fronting the funds and covering the delay themselves for a fee. You can also bridge straight to the main chain yourself and wait out the window.

Does choosing a ZK rollup reduce risk?

It removes one category of risk, the challenge period during which a fraudulent batch could be accepted. It does not remove sequencer centralisation, contract bugs in applications, or bridge risk. Every layer adds its own failure modes.